CodeGem is people-first engineering management software. Employee engagement and satisfaction are supported through a healthy culture; wellness is prioritized by top managers, and development analytics can help leaders focus on what matters. With CodeGem, you can expect a performance boost from your engineering team. Drive continuous improvement by connecting industry-leading KPIs, actionable alerts, career development, and wellness analytics into a 1-on-1 framework that unifies it all.

CodeGem

codegem

How to quickly get past ubiquitous «echo bots» tutorials and build an assistant that actually does something. Just clone, tweak and launch. Can it become a «bot-end» framework? *a reference to Kraftwerk’s single “The Robot” (1978) It started with a screw up I have to admit: as a former journalist, I still write faster than I think. Turns out, it applies to code too. in my series. Even though saw it and some readers used it to create their own projects, no one seemed to have noticed the gaping design flaw. I misused gem in a way that when several users try talking to the tutorial bot at the same time, their messages get mixed up. When one user is in the middle of the interaction, bot is hung up on the task in progress and any other message from any other user would be treated as the next step in the conversation thread. and I desperately wanted to make up for it. I had to rethink the architecture to handle requests from different users and remember their place in conversation. A simple refactoring led to more refactoring, and more refactoring, and more… As a result, I created a simple tool that solves a number of design problems with bots, so you don’t have to solve them yourself**.** I royally screwed up “Build your first Messenget bot in Ruby” more than 6000 people facebook-messenger That sucked M — your starting point to mess with Ruby chatbots for Messenger, the largest messaging platform out there. At this early stage it’s more a and I haven’t packaged it as a separate gem yet, so when you download the boilerplate, you can see how its internals are implemented inside the folder. It also contains some example code that you’re free to tweak and replace. eet Rubotnik Facebook proof of concept /rubotnik What it does You can bind keywords (“help”, for instance, as in or to your bot’s functions (we will call them “commands” from now on) with a straightforward syntax: . Forget about with a dozen of ’s to parse your input. “help!” “Would you help me, please” bind ‘help’, to: :give_help case when You write all your bot’s logic as separate methods inside boilerplate’s module and you can chain them together to create threads of dialogue with Commands next_command :method_name You can generate UI elements with the help of convenience classes: (it also makes those Facebook’s Generic Templates a bit easier to generate) UI::ImageAttachment.new(‘http://your.image/url’).send(@user) You can set up a nested persistent menu and a greeting screen for your bot. Alter and to create your own. rubotnik/persistent_menu.rb rubotnik/bot_profile.rb Its will walk you through all the steps necessary to hook up your bot on Facebook and bring it to life. README It’s a standalone service that runs on Rack (it uses Puma server with default config). Think of it as framework for your web app, if you have one. It has enabled by default so you can create your own webhooks and handle incoming API connections. “bot-end” Sinatra It’s primed for easy deployment with Heroku. Just do and once you’re done testing on localhost. heroku create heroku push origin master What it doesn’t do It has no persistence, but can be extended to work with any database. By default, Rubotnik has simple in-memory storage that keeps track of connected users and their state (see and to see how it’s implemented). rubotnik/user.rb rubotnik/user_store.rb It doesn’t work inside of a Rails project (this is one of the features that can be thought about in future). On the other hand, it can talk to your existing Rails app through REST API. How it works So how do you build your bot with Rubotnik? First, make sure you have your and installed — we will need this for development, as well as for moving your bot to the cloud. Then follow this steps: Heroku CLI ngrok git clone :progapandist/rubotnik-boilerplate.git git@github.com mv rubotnik-boilerplate NAME_OF_YOUR_PROJECT cd NAME_OF_YOUR_PROJECT rm -rf .git  # remove boilerplate’s git history git init     # track your own project Open the project in your favorite editor and take a look at the file. There’s a bunch of stuff already written, that constitutes logic for a that you can play with . bot.rb demo bot here The that demoes boilerplate features bot It implements an example of a simple API call to Geocoding to determine details about your location, a sample questionnaire to demonstrate gathering data from user message by message, and it also renders some of Facebook’s UI elements to get a taste what your own bot can offer in terms of user experience. Google If you don’t change anything at this point and just follow in the boilerplate’s readme, you’ll get the exact copy of the demo bot. It lets you understand what does what inside the boilerplate, but you probably want to build your own thing, right? Then just take a quick glance at the sample code and replace it with your own. instructions for bot setup Note that all the heavy lifting (sending data to Facebook Messenger Platform and handling incoming requests to your bot’s webhook) is done by the gem ( from Norway did a terrific job, and a community around his project is growing steadily). Rubotnik merely piggybacks on its API, like so: facebook-messenger that guy Same logic applies to postbacks, which are events triggered when the user interacts with buttons in your bot’s UI. Let’s take a closer look, two things are happening in this snippet: is an example of Rubotnik’s . bind ‘string' DSL is a helper function that allows you to send a text message to connected user with a simplicity of say say "Hello!" Take a look at to see how commands can be chained together to form threads. /commands/questionnaire.rb Check the and read through comments in boilerplate’s code to learn more about DSL, helper functions and where to apply them. README How it’s done Rubotnik’s main classes are called and , they can be found inside of folder. Here’s a snippet that shows how message dispatching is implemented: MessageDispatch PostbackDispatch /rubotnik A instance variable is accessible everywhere through common Rubotnik namespace and represents any given incoming message. It’s an instance of facebook-messenger’s class and has all its properties (refer to facebook-messenger’s ) @message Facebook::Messenger::Incoming::Message README @message.id          # => 'mid.1457764197618:41d102a3e1ae206a38'@message.sender      # => { 'id' => '1008372609250235' }@message.seq         # => 73@message.sent_at     # => 2016-04-22 21:30:36 +0200@message.text        # => 'Hello, bot!'@message.attachments # => [ { 'type' => 'image', 'payload' => { 'url' => 'https://www.example.com/1.jpg' } } ] Same goes for that’s an instance of @postback Facebook::Messenger::Incoming::Postback references the connected user and is an instance of class defined in . The main point of this class is to hold onto user’s facebook ID and keep track of commands bound for execution. You can add your custom containers for state (as in the case with property, used to track user’s answers in ). @user User user.rb answers questionnaire.rb That’s pretty much all the not-so-secret sauce that ensures all users are served independent of each other and the conversation with them continues exactly at the point it stopped, even after a long interruption. How to go live Once you’re done building and testing your bot, just create a heroku app from your console and run . Then update your webhook in Facebook developer console and that’s it! git push heroku master How to say if you ❤️ it or 👺 it For now Rubotnik is just a bunch of code organised in some folders. It’s raw. It’s a proof of concept. It needs polishing. Some of the code has to do with actual “framework” logic, while other bits just demo example bot’s features. It has statements sprinkled everywhere. There’s a bunch of debug statements that are designed to help with understanding how bot performs. But they can’t be allowed to stay there for long. require_relative Some of my practices may be frowned upon by seasoned programmers. That’s understandable, I only discovered Ruby magic . last year That’s exactly why I invite everyone to collaborate. Feel free to fork the boilerplate and make PR’s or just and I’ll add you as a collaborator for the main repo. email me I’m also looking for ways to test Rubotnik with some real world examples, so if you are planning to integrate your service with Messenger, contact me and I’ll help you out. What next? Obviously, Rubotnik should be turned into a gem. But as it’s tightly coupled with , I don’t know where to start yet. It should also probably have -style generators to create directory structure for new projects. facebook-messenger rails new I’m also looking into and tools to enable easy interactions with NLU engines that would make your bot a better conversationalist. wit.ai api.ai Recommend 💚 this article on Medium and stay tuned! 🤖 BOT POWER! 🤖

Ya tvoy Rubotnik*. Beginner-friendly Ruby boilerplate to build your own Messenger bot

For the past 6 months, I’ve been working on GPT Pilot ( ) to understand how much we can really automate coding with AI, so I wanted to share our learnings so far and how far it’s able to go. https://github.com/Pythagora-io/gpt-pilot When I started building GPT Pilot, I wrote on how it is envisioned. Now, I want to share my revised thinking and go into what worked and what didn’t work. this blog post Finally, you’ll see examples of apps that were created with GPT Pilot and how you can create an app with a AI pair programmer. real What Is the Idea Behind GPT Pilot? GPT Pilot is envisioned as a r – . Rather, it is a developer who creates a plan for how your app or feature should be built and starts coding. It wants to do most of the coding by itself, but when it gets stuck, it needs clarification about the given requirements, or requires a code review, it asks you for help. eal AI developer not an autocomplete or a chatbot Is AI Like a Junior Developer? Or… I often see CodeGen GPT-4-based tools that say they are building an AI junior developer. Somehow, I’ve always had a problem with that because when I use ChatGPT for coding, it gives me answers and ideas that only a super-senior person could give – something that absolutely no junior dev would even be able to grasp. Still, no LLM can build an app nearly as well as a senior developer can, but the knowledge GPT-4 has about coding is way beyond any junior developer. I would say that GPT-4 has so much knowledge about every part of software development like it’s the most senior developer in the world but with the memory of a goldfish. I picture it as a superhuman robot that just stands in the middle of a room and can only do a single small action at a time, but it cannot combine many actions and work repetitively. You must tell it exactly what it should do next. This is what we’re after with GPT Pilot – we want to create a framework of thinking for the LLM that gets that superhuman robot to continuously work by revising its previous actions, have a feedback loop, and determine what should it do next in order to finish the end goal, which is to build a production-ready application. , I outlined the main pillars on which GPT Pilot was built. But these have changed a bit based on our team’s learnings, so here are the revised pillars: In the blog post I mentioned above not only because AI is not good enough but also because you might want to change how something works or looks after it’s implemented. It’s common for a developer or product manager, once they see what an implementation looks like, to decide to change it. A human is needed to supervise the AI Or, you realize there are more edge cases than you initially anticipated and think it’s easier to refactor your current implementation than to fix every issue. The problem is when you finish the entire app and then try to refactor it – this is when it becomes much harder because every change will impact all the other features. On the other hand, if you do the refactor before you commit your changes, you’ll be able to proceed with the next features on top of well-written code. This is why . it’s crucial for an AI developer to have a human in the loop whenever a task is implemented This way, the human can review the implementation of each task (just like a code review before merging a PR) before GPT Pilot continues on to the next task. If a human tells GPT Pilot what is wrong, it will be much easier to fix the issues within the task itself. At the same time, the LLM has the context of what needs to be done in the task and what has been done so far. I have a feeling that many people judge ChatGPT’s ability to write code by how well it delivers the first time you ask it to code something. If it doesn’t produce working code, many will think it’s not impressive. In reality, . Instead, you write code, run it, see the errors, and iterate. This is exactly what GPT Pilot enables GPT-4 to do – after it writes code, GPT Pilot can run the code, take the output, and ask the LLM if the output is correct, if something should be fixed, and if so, how. AI can iterate over its own mistakes. humans almost never write working code on the first try . There are many repetitive routines that all developers go through when building an app. One of the routines can be – write code, run it, read the errors, change code, rerun it, etc. Another higher-level one can be – take a task, implement it, test the implementation (repeat until all tests pass), send it for review, fix the issues (repeat until the reviewer approves), and deploy. Many of these routines can be orchestrated if we have an intelligent decision-maker in the loop (like an LLM). Software development can be orchestrated . When we created the first version of GPT Pilot, we thought it would need to iterate over tasks, implement code, fix it, and move on. In reality, you don’t continuously progress when coding an app – you rewrite your code all the time. The coding process is not a straight line Sometimes, you refactor the codebase because, after the initial implementation, you realize there is a better way to implement something. Other times, you do it because of a change in requirements. Like I mentioned in #1, after you see that a solution isn’t working, you sometimes need to roll back a bunch of changes, think about an alternative solution to the problem, and try solving it that way. To make GPT Pilot, or any other AI developer, work at scale, it needs to have a mechanism that will enable it to go back, choose an alternative path, and reimplement a task. What Did We Learn? LLMs, in general, are a new technology that everyone is trying to understand – how it works, what can be done better, how to do proper prompt engineering, etc. . Our approach is to focus on building the application layer instead of working on getting LLMs to output better results The reasoning is that LLMs will get better, and if we spend weeks optimizing a prompt, it might be completely solved with the new version of GPT. Instead, we’re focusing on what the user experience should look like and which mechanisms are needed to control the LLM to enable it to continuously work, getting closer and closer to the final solution. So, here are our learnings so far: . Our original thinking was that, with the human’s input, GPT Pilot would be able to navigate in the right direction and get closer and closer to the working solution, even if the initial description was vague. The initial description of the app is much more important than we thought However, GPT Pilot’s thinking branches out throughout prompts, beginning with the initial description. And with that, if something is misleading in the initial prompt, all the other info that GPT Pilot has will lead in the wrong direction. So, when you correct it down the line, it will be so deep into this incorrect way that it will be almost impossible to get it onto the right path. Now, as I’m writing this, it seems so obvious, but that is something we needed to learn – to focus much more on the initial description. So, we built a new agent called “Spec Writer,” which works with you to break down the project requirements before it starts coding. . As I mentioned above in the pillars section, refactoring happens all the time, and GPT Pilot must do so as well. We haven’t implemented a solution for this yet, but it will likely work by adding the ability for GPT Pilot to create markers around its decision tree so that whenever something isn’t working, it can review markers and think about where it could have made a wrong turn. Coding is not a straight line . My thinking was that if an agent reviews what the other agent did, it would be redundant because it’s the same LLM reprocessing the same information. But it turns out that when an agent reviews the work of another agent, it works amazingly well. We have 2 different “Reviewer” agents that review how the code was implemented. Agents can review themselves One does it on a high level, such as how the entire task was implemented, and another one reviews each change before they are made to a file (like doing a git add -p). compared to multiple problems in a single prompt. For example, if you tell GPT Pilot to make 2 different changes in a single description, it will have difficulty focusing on both. So, we split each human input into multiple pieces in case the input contains several different requests. LLMs work best when they can focus on one problem . This is very obvious now, but initially, we didn’t tell GPT-4 to add any logs around the code. Now, it creates code with verbose logging, so that when you run the app and encounter an error, GPT-4 will have a much easier time debugging when it sees which logs have been written and where those logs are in the code. Verbose logs help . This is also an obvious conclusion, but we had to learn it. It’s much easier for GPT-4 to implement features and fix bugs if the code is split into many files instead of a few large ones. Just like we, humans, think – it’s much easier to process smaller chunks of code rather than big ones. We do that by creating levels of abstraction – functions, classes, etc. Splitting the codebase into smaller files helps a lot One of the easiest ways to get the LLM to create a more manageable abstraction is just to tell it to create more modular code and split it into more files. It works amazingly well, and the end result is also better for us, developers. . GPT Pilot’s goal is to do 90% of all coding tasks and leave the other 10% to a human. This 10% usually comprises fixes or small changes that the LLM struggles to notice, but for the human, it might be a simple change. For a human to be able to fix the code, they need to be clearly shown what was written in the task and the idea behind it However, the problem is that it’s not easy to tell the human what is not working and what code they should look at. If GPT Pilot writes 3,000 lines of code, the human developer, if they want to help GPT Pilot, needs to understand the entire codebase before diving into the actual code. In future versions of GPT Pilot, the human developer will have a detailed explanation of what code has been added to the current task and the reasoning behind it. This way, you will be able to help GPT Pilot. . LLMs are better off asking humans questions rather than letting humans think about all the possible errors. Again, it’s very obvious looking back, but one of the things we noticed was that people were much more willing to answer questions that GPT Pilot asked them instead of having an open-ended input field where they could write unconstrained feedback. Humans are lazy . This was one of the biggest learnings for me. I thought you could prompt GPT-4 by giving it a couple of solutions it had already used to fix an issue and tell it to think of another solution. However, this is not as remotely easy as it sounds. It’s hard to get an LLM to think outside the box What we ended up doing was asking the LLM to list all the possible solutions it could think of and save them into memory. When we needed to try something else, we pulled the alternative solutions and told it to try a different, but specific, solution. Apps Created With GPT Pilot Currently, you can create simple but non-trivial apps with GPT Pilot. We’ve also seen people create apps with GPT Pilot that are very impressive, such as an app that can fine-tune a ResNet model to count palm trees and then, when you upload an image, count the trees in it. Here are a couple of apps we created, along with the code, stats, and demos: Prompt Lab ( ) DEMO https://www.youtube.com/watch?v=UXhC4nx0pvs&embedable=true Think of this as OpenAI Playground on steroids – it enables you to load a conversation from a JSON array or enter it manually, run the conversation with the LLM X number of times, save the conversation to the database, and load it back in. We actually use this app when we engineer a prompt and want to see how it behaves. The Playground is not good enough because it’s time-consuming to repetitively rerun a single prompt. With Prompt Lab, you can choose how many times to run it and let the app run the prompt repeatedly. Once it’s finished, you can analyze the results. This might be useful for people who are building an AI app and need to optimize their prompts. ⏳ Time spent: ~2 days 💾 Github repo ( ) SQLite Database Analyzer Tool DEMO https://www.youtube.com/watch?v=gKgswTm5tFo&embedable=true This is also an app we use internally to analyze a local SQLite database. It pulls the data from the database in a format that’s very specific to the GPT Pilot database structure, but it can easily be modified to fit other structures. It reads and uploads your SQLite database, splits the rows by a specific field, unpacks the rows into values, loads the LLM conversation data into a form, and enables you to simply change one of the messages and submit the LLM request to GPT-4 to see how the results will look. This way, you can analyze the conversations GPT Pilot’s agents have with the LLM and easily explore what would happen if the prompts were different. ⏳ Time spent: ~2 days 💾 Github repo Code Whisperer ( ) DEMO https://www.youtube.com/watch?v=8qzq6F5DtyE&embedable=true Code Whisper is a fun project we created as an example to showcase. The idea is that you can use it to ask the LLM questions about your codebase. You paste in the link to a public Github repo. Then, it clones the repository, sends the relevant files to the LLM for analysis, which creates a description for each file about what the code does, and saves those descriptions into the database. After that, you can ask the app questions about the codebase, and the codebase shows you the response. In this demo, we use GPT-3.5. ⏳ Time spent: 7 hours 💾 Github repo Star History ( ) DEMO https://www.youtube.com/watch?v=mjc4vnlLz3M&embedable=true I’ve been releasing open-source projects for years now, and I’ve always wanted to see how fast my Github repo is growing compared to other successful repositories on . The problem is that on Star History, I’m unable to zoom into the graph, so a new repo that has 1,000 stars cannot be compared with a big repo that has 50,000 because you can’t see how the bigger repo does in its beginning. https://star-history.com/ So, I asked GPT Pilot to build this functionality. It scrapes Github repos for stargazers, saves them into the database, plots them on a graph, and enables the graph to be zoomed in and out. ⏳ Time spent: 6 hours 💾 Github repo Conclusion I hope you gained some insight into the current state, problems, and findings that we deal with at GPT Pilot. So, to recap: We think that a real AI developer tool should be based on the following pillars: A human is needed to supervise the AI. We should enable the AI to iterate over its own mistakes. , which should be implemented on a layer on top of LLMs. Software development can be orchestrated because, in reality, the coding process is not a straight line. The AI developer should be able to refactor the codebase So far, we’ve learned that: The initial app description is much more important than we thought Coding is not a straight line, agents can review themselves LLMs work best when they focus on one problem compared to multiple problems in a single prompt Verbose logs do miracles Splitting the codebase into smaller files helps a lot For a human to be able to fix the code They must be clearly shown what has been written and the idea behind it Humans are lazy It’s hard to get the LLM to think outside the box What do you think? Have you noticed any of these behaviors in your interactions with LLMs, or do you have a different opinion on any of these? I would be super happy to hear from you, so leave a comment below or shoot me an email at . zvonimir@gpt-pilot.ai You can try creating an app with AI with GPT Pilot here: GPT Pilot open source repo GPT Pilot VS Code extension If you liked this post, it would mean a lot if you star the , and if you try it out, let us know what you think. Your feedback is really important to the entire GPT Pilot Team. GPT Pilot Github repo Also published here

After 6 Months of Working on a CodeGen Dev Tool (GPT Pilot), This Is What I Learned

কোডজেন ডেভ টুলে (GPT পাইলট) 6 মাস কাজ করার পর, আমি এটাই শিখেছি

Nach 6 Monaten Arbeit an einem CodeGen-Entwicklungstool (GPT-Pilot) habe ich Folgendes gelernt

Después de 6 meses de trabajar en una herramienta de desarrollo CodeGen (piloto GPT), esto es lo que aprendí

Après 6 mois de travail sur un outil de développement CodeGen (pilote GPT), voici ce que j'ai appris

कोडजेन डेव टूल (जीपीटी पायलट) पर 6 महीने काम करने के बाद, मैंने यही सीखा

CodeGen 開発ツール (GPT パイロット) を 6 か月間使って学んだこと

6개월간 CodeGen 개발 도구(GPT 파일럿) 작업을 하면서 배운 점은 다음과 같습니다.

Após 6 meses trabalhando em uma ferramenta CodeGen Dev (GPT Pilot), foi isso que aprendi

Вот что я узнал после 6 месяцев работы над инструментом разработки CodeGen (пилотная версия GPT)

CodeGen Geliştirme Aracı (GPT Pilot) Üzerinde 6 Ay Çalıştıktan Sonra Öğrendiğim Şey Bu

Sau 6 tháng làm việc trên CodeGen Dev Tool (GPT Pilot), đây là điều tôi học được

使用 CodeGen 开发工具（GPT Pilot）6 个月后，这是我学到的东西

The world of artificial intelligence (AI) is undergoing a seismic shift, largely driven by the emergence of Large Language Models (LLMs). These open-source LLMs are pushing the boundaries of what AI can achieve, and in this blog post, we'll delve into some of the most remarkable models that are shaping the future of technology and communication. Each LLM offers unique strengths and capabilities, making them indispensable tools for developers, researchers, and organizations. Let's embark on a journey to discover the potential of these cutting-edge models. : Empowering AI Interactions LLaMA 2 LLaMA 2, developed collaboratively by Meta AI and Microsoft, is a groundbreaking AI language model with three available sizes, ranging from 7 to a staggering 70 billion parameters. It's not just an upgrade but a monumental leap in AI capabilities. LLaMA 2 possesses the ability to comprehend both text and images, making it ideal for multimodal tasks. Supported on platforms like Azure and Windows, this model democratizes AI access. Safety is at its core, with extensive training to minimize harmful outputs. Key Advantages: Multimodal capabilities for text and image understanding, suitable for diverse tasks. Availability in three sizes, catering to a wide range of use cases. Seamless integration with platforms like Azure and Windows. Rigorous safety measures to ensure responsible AI use. Open availability for fine-tuning on multiple platforms. Diverse training data for comprehensive understanding, reducing biases. : Elevating AI Performance Claude 2 Claude 2, developed by Anthropic, is a model designed to elevate AI performance to new heights. This model achieved an impressive score in the Bar exam, surpassing its predecessor. In GRE reading and writing exams, Claude 2 performed above the 90th percentile, showcasing its proficiency in comprehending and generating intricate content. It excels in processing extensive documents and demonstrates enhanced coding capabilities. Safety is paramount, ensuring responsible AI use. Key Advantages: Remarkable performance in academic evaluations, highlighting its competency. Ability to process inputs of up to 100K tokens, enabling in-depth analysis. Enhanced coding proficiency for complex programming tasks. Focus on reducing harmful content generation, prioritizing ethical AI. Plans for global availability in the near future, expanding its reach. : A Versatile Text-To-Text Model T5 T5, or Text-To-Text Transfer Transformer, is a versatile pre-trained language model developed by researchers at Google AI. It’s based on the Transformer architecture and designed to handle a wide range of natural language processing tasks through a unified “text-to-text” framework. With 11 different sizes, T5’s models vary from small to extra-large, with the largest having 11 billion parameters. Key features of T5: Encoder-decoder architecture: T5 employs an encoder-decoder architecture, treating almost all NLP tasks as a text-to-text problem. This results in enhanced consistency in model design. Pre-training for diverse tasks: In T5’s pre-training process, the model generates target text from the source text, which includes various tasks like translation, summarization, classification, and more. This approach results in a versatile and unified model. Flexible input-output paradigm: It operates in a “text as input, text as output” paradigm. Framing tasks in this manner reduces complexity and allows fine-tuning for specific objectives. Adapter-based architecture: T5 uses a modular architecture that adapts to new tasks through several additional parameters. Contextual consistency: T5 maintains coherence in lengthy interactions and produces natural-flowing conversations. : The Open-Source Powerhouse GPT-NeoX-20B GPT-NeoX-20B, developed by EleutherAI, is a formidable open-source AI model with 20 billion parameters. It builds upon the architecture of GPT-3 while introducing innovations like synchronous data parallelism and gradient checkpointing. GPT-NeoX-20B is known for its ability to produce coherent and contextually relevant content, efficient multi-GPU training, and fine-tunability for various applications. Key Advantages: Innovative features like synchronous data parallelism enhance training efficiency. Efficient multi-GPU training for faster model development. Coherent and contextually relevant content generation, ensuring high-quality outputs. Fine-tunable for various applications, making it adaptable to specific needs. : Scaling Down Without Sacrificing Quality GPT-J GPT-J is a model with 6 billion parameters, making it more accessible compared to larger models. Trained on the Pile dataset, it shares its roots with the GPT-2 architecture. GPT-J employs parallel decoders for efficient text processing, excelling in powerful text generation capabilities. With a user-friendly API, it's a cost-effective alternative to larger models. Key Advantages: Accessibility with 6 billion parameters, balancing cost and performance. Efficient text processing with parallel decoders, speeding up text generation tasks. Powerful text generation capabilities for natural language processing tasks. User-friendly API, simplifying integration into various applications. Cost-effective alternative to larger models, reducing computational expenses. : Efficiency and Scale Unleashed OPT-175B OPT-175B boasts a colossal size of 175 billion parameters and is primarily trained on unlabeled text data containing English sentences. It utilizes gradient checkpointing for memory efficiency, excels at few-shot learning, supports mixed precision training, and is committed to reducing its carbon footprint. Key Advantages: Massive size with 175 billion parameters, enabling comprehensive learning. Memory-efficient gradient checkpointing, reducing memory consumption during training. Excellence in few-shot learning, allowing the model to adapt quickly to new tasks. Support for mixed precision training, optimizing training speed and efficiency. Focus on reducing carbon footprint, promoting environmental responsibility. : Fostering Scientific Collaboration BLOOM Description: BLOOM, developed by BigScience, is a monumental achievement with 176 billion parameters. It's designed to foster scientific collaboration and breakthroughs. BLOOM relies on 46 natural world languages and 13 programming languages, ensuring inclusivity. With advanced contextual comprehension and ethical communication, it prioritizes responsible AI use and cultural sensitivity. Key Advantages: Large-scale with 176 billion parameters, enabling in-depth scientific research. Multilingual competence for global collaboration, breaking language barriers. Advanced contextual comprehension, providing nuanced responses. Ethical communication and cultural sensitivity, promoting responsible AI. Inclusive language for a diverse user base, reducing biases. : China's AI Contender Baichuan-13B Baichuan-13B, introduced by China's Baichuan Inc., is a formidable open-source LLM designed to compete on the global stage. With 13 billion parameters and a pre-training dataset of 1.3 trillion tokens, it excels in both English and Chinese AI language processing. It empowers applications spanning sentiment analysis to Mandarin content creation, aligning with Baichuan's mission to democratize generative AI. Key Advantages: Proficiency in understanding and generating Chinese content. Simplified data interaction for research and trend analysis. A vast capacity of 13 billion parameters for nuanced communication. Industry-grade performance across various applications. : Bidirectional Context Understanding BERT BERT (Bidirectional Encoder Representations from Transformers) was created by researchers at Google AI. With a model size of up to 340 million parameters, BERT has been trained on a diverse dataset comprising 3.3 billion words, including BookCorpus and Wikipedia. Key features of BERT: Bidirectional context: BERT comprehends context from both directions in a sentence, enhancing its grasp of nuanced relationships and improving understanding. Attention mechanism: It employs attention mechanisms focusing on relevant words, capturing intricate dependencies, and enabling the model to give context-aware responses. Masked language model: During training, BERT masks certain words and predicts them using surrounding context, enhancing its ability to infer relationships and meaning. Next sentence prediction: BERT also learns to predict whether a sentence follows another in a given text. It enhances BERT’s understanding of sentence relationships, which is beneficial for tasks like question answering and summarization. Task agnostic: BERT’s pretraining and fine-tuning approach enables easy adaptation to different tasks. It can achieve remarkable results even with limited task-specific data by fine-tuning the pre-trained model on specific tasks. : Streamlining Software Development CodeGen CodeGen, a creation by Salesforce AI Research, is inspired by the GPT-3.5 architecture and offers a range of sizes, including 350 million, 2 billion, 6 billion, and an impressive 16 billion parameters. It has been trained on a diverse set of programming languages and frameworks, making it a valuable tool for generating accurate and reliable code solutions. Key Advantages: Accurate and reliable code generation using a vast training dataset. Flexibility in understanding and generating code in multiple programming languages. Error identification and handling, improving code quality. Potential to streamline software development processes and enhance developer productivity. *** These open-source LLMs are transforming the landscape of AI, from enhancing language understanding to promoting ethical AI use. As they continue to evolve, they hold the potential to redefine the possibilities of technology and communication. Explore these models and embark on your journey into the future of AI. Here at Linguix we’ve experimented with lots of those LLMs and our team is happy to consult you or help you implement them. Just us here to get started! email

From LLaMA 2 to CodeGen: Navigating the World of Open-Source LLMs

Our company faced enormous challenges when the full-scale Russian invasion of Ukraine started. But instead of cutting costs, we decided to continue actively expanding. India became one of our top-priority markets, so after research and planning, we went for it. The team is still under a lot of stress: most employees stay in Ukraine and sometimes have to work from bomb shelters, and others are spread across the globe. But expansion is going well, and I’m ready to share some insights and lessons learned. Choosing the market: the paradise for Java developers We help people learn Java because we know this profession can be fulfilling and well-paid simultaneously. Over the last ten years, we've gained tremendous experience in working with international students (over 30,000 people from 40 countries have already become our graduates) and entered many markets (the USA, China, and European countries). So, we weren't new to this business. Nevertheless, India seemed like a big challenge. To name a few reasons, the huge cultural gap between Ukraine and India, and the circumstances in which our employees have to work. After the war started, some people fled the country, and some moved to other regions of Ukraine. It got harder to communicate daily and hold meetings. But we knew the potential of this market and were ready to put extra effort into this idea. What made India so attractive to us? This country's economy is growing fast, and the share of the young population is pretty high ( ). There are over 658 million Internet users in India, which explains the high demand for online education and the booming growth of IT companies. people between 15 and 29 years old constituted 27,2% in 2021 India's the world's leader in IT outsourcing, and the demand for programmers here continues to increase. For example, in 2021, India witnessed the launch of 14,000 new startups. With every year, we can expect this number to get bigger, so the competition for skilled employees will get tougher. One of the top hiring websites in India, , offers over 100,000 job openings for IT specialists. And the salaries are appealing: Java developers here can earn up to 2 million rupees per year (the average annual salary in India is 240,000 rupees, according to ). Naukri.com AmbitionBox.com It's only logical that young people want to acquire a profession of a Java developer. But unfortunately, universities can't meet this demand. They aren't flexible enough to change the curriculum as fast as the market requires, and it takes too much time (several years!) to get a degree. On the other hand, online courses can offer up-to-date programs, a lot of practice, and sometimes even help with a job search. That's why EdTech has been flourishing in India in the last few years. This market is expected to reach , with 37 million paid users. Seven among 36 in the world are Indian. Moreover, the biggest EdTech unicorn is also Indian – it’s ByJu’s with a market value of $22 billion. $10.4 billion by 2025 EdTech unicorns Some segments of the EdTech market in India are more competitive than others. For example, K-12 is quite filled compared to additional professional education. It's not a blue ocean for us, but also not a red ocean. In our segment, we see big and small companies, but the "market pie" isn't sliced yet, and the leaderboard isn't final. And what's more important, no Java learning platforms offer such a high-quality product as we do. With these considerations in mind, we decided to start our journey. Research and preparation: better thorough than sorry Having an idea is one thing, and transforming it into a business plan is the other. That’s why we decided to enter the Indian market only after detailed research. How did we conduct it? First, we examined every study we could find on the Indian EdTech market, its potential, and our soon-to-be competitors. We analyzed the information, shared our conclusions with team members, and discussed our findings. Second, we had multiple meetings with market experts. We reached out to some representatives of Indian EdTech companies and asked them questions to get a better perspective on the local market. Third, we couldn’t proceed without talking to potential customers. So, we interviewed Indian clients of our main product – a self-paced online Java course. Also, we surveyed them. We already had several hundred users, and it was beneficial to ask for their feedback. Moreover, we talked to Indian employers and browsed local job search websites. We wanted to understand if the requirements for Java developers differ in India compared to other countries. Our team made everything possible – and impossible – to gather as much information as we needed. And we did it under air raid sirens which are pretty common in Ukraine, and from bomb shelters. Our employees adjusted even to these circumstances. Some of them had to catch the Internet connection to work. But their motivation remained strong despite the high level of stress and discomfort. Eventually, this stage gave us tons of valuable insights. For example, we discovered that the main thing our potential Indian students desired was employment. It’s not enough for them to gain Java knowledge or improve their coding skills. They want to get a job offer after they complete the course. And it’s the main reason Indian customers aren’t satisfied with existing courses. Most of them last just a few months, and finding a job is impossible after such a brief study. Also, we came to another conclusion: we really can deliver what the target audience needs. As our surveys say, 95% of our graduates find their first job within three months after completing the course. We were sure that in 12 months, we could train Java developers who were able to get job offers immediately. The first hypothesis, first results… Our research helped us create the first hypothesis, i.e., if we offer potential students a job guarantee, they will gladly choose us and be willing to pay for the first month or the entire course at once. To test it, we did several steps: Talked to employers to identify their needs and reach preliminary agreements to hire our graduates. Surprisingly, some companies wanted to start hiring our graduates immediately! It proved our point that the demand for Java developers in India is huge, and our reputation is substantial and works for us. Tailored the existing product (self-paced course) for the Indian market. CodeGym Java University is a 12-month course with mentors and a job offer guarantee at the end. It combines all the benefits of our online course (gamification, lots of practice, automatic verification of tasks completed, etc.) with a guided approach (live lectures and mentors’ help). Created a landing page for CodeGym Java University and started holding free online events about IT in general, Java, and our offer. We invited local speakers to gain more trust from the audience. Over 1,500 people visited our online events, and about 100 filled out applications for our course. As a result, we had our first group of students in a month. They are studying now, and so far, they are satisfied with what they get. So, our first hypothesis worked, and we moved to the next challenge – growing and scaling. … and first complications Not everything was going as smoothly as we had hoped. The first problem became evident when the sales started. We had a sales team, but it wasn't local. Our employees live in different countries and mostly use Western approaches to communicate with potential clients. For example, go straight to business, discuss the offer, and close the deal quickly. In India, on the other hand, people prefer an entirely different approach. To close the deal here, you must establish personal communication first and talk about family, relatives, life, etc. Only then can you start discussing your offer. Also, we discovered that the trust issue was crucial. If a potential client hears an unfamiliar accent, they immediately understand that there's a foreigner on the other end and start asking questions. The USA or European experience isn't relevant here. Therefore, potential customers may doubt your expertise. And, of course, we didn't know much about the cultural differences between Indian states. For instance, there are many local holidays, different for different states. And selling something on these dates is pointless. All these things created barriers to communication. As soon as we understood that, we decided to hire a local team. Now, we already have four sales managers in India, several career consultants, and Java mentors. They help us understand the cultural features, comprehend the needs of our target audience, and process applications more efficiently. Our mentors are professional Java developers who work for companies like SalesForce, Walmart Tech, and Microsoft. We also trained them to use our teaching methods, and now they combine understanding the local culture with CodeGym's DNA. Typically, we realize the need for a particular employee first and then start looking for them. But sometimes, we find an interesting person for a position we don't have yet. If this person can make us stronger, and we can afford them, we go for it. For example, that's how Shubham Dumbre joined our team. He's an IT entrepreneur, an author, and a speaker with immense expertise in EdTech companies. He helps us to hold our open events and communicate with potential students. Another complication was out of our control. Russians started bombing the civil infrastructure with missiles and drones, and electricity became a luxury. Blackouts can last for several hours, even days (and we can’t complain because, in some regions, people live without electricity for months). Again, we had to adjust. We’ve bought equipment (power generators, power banks, WiFi routers that can work without electricity, etc.) and learned to keep all the devices charged. And we continue working. Getting closer to success, iteration by iteration As I've said before, scaling is our next big challenge. To keep the process going and fix all the problems as soon as they appear, we conduct special meetings. Every week (usually on Friday), we get together with the "Indian" team and discuss the previous week's results. We've divided the entire project into funnel stages to discuss each part's success separately. For example: Awareness and advertising. What experiments did we conduct last week? What else did we do? What results did we get? Our free webinars. How did the chatbots, emails, and SMS messages work? What was the attendance of the webinar? What questions did people ask there? What do the statistics about each webinar say? During these meetings, we analyze how the audience's interest changes and what data show. The sales process. What insights did we get from our leads and sales managers? What are the sales metrics? Who among the sales managers was the most efficient? The learning process. How are students of current groups learning? What's their performance? How many tasks per week did they solve on our platform? What feedback did students give to mentors? And what feedback did mentors give to students? In wartime, not every employee can attend a meeting, and it's not their fault. To keep everyone on the same page, we take notes and share them with people who couldn't participate in a discussion. Our internal communications have become somewhat asynchronous, but it doesn't affect our results. The last months were challenging for us, but also, they taught us many valuable lessons. For example, we should always be prepared for anything, both on personal and company levels. We are much stronger than we thought. People really are our main asset (that's why we've kept the whole team and are proud of this decision). Sure, there will inevitably be other challenges as we continue to pursue our "Indian dream.” But we stay together and stay true to our mission – teach people Java and give them an in-demand profession. In this uncertain world, all we can count on is ourselves. By Alex Yelenevych, Co-Founder & Chief Marketing Officer at , a platform for learning Java CodeGym.cc Also Published Here

How a Ukrainian Startup  CodeGym Managed to Enter a New Market During the War

Over the last few years in the developer world, November has been taken over by a creative coding challenge called Codevember. I’m attempting Codevember myself this year, and hopefully, after reading this you will challenge yourself too! What is Codevember? Codevember is a challenge by and which invites developers to post a new creative piece of code for every day of November. Arthur Rob Valentin Daguenet The has a list of hints for each day of the month, e.g. , and . Using the hint, participants come up with a creative bit of code inspired by the word. For example, for the hint , you could create a usable piano with JavaScript. Codevember website 1. Infinity 2. Time 3. Carrot Music Why should I do Codevember? Here are a few reasons why you should partake: Gives you a chance to experiment with new technologies that you might not always have a chance to Helps expand your knowledge of HTML, CSS and JavaScript whilst having fun After 30 days you will have a collection of code creations that you can be proud of How can I do Codevember? The Codevember team aim to make the challenge as accessible as possible and allow you to submit your creations using any platform/tool you like. If you want to get your work featured on the Codevember website, you can do so in a few ways. To attempt , post a solution each day with the hashtag #codevember and the day of the month e.g. . CodePen users can get their pen’s featured with the same title format. Alternatively, you can directly to the Codevember site. Codevember on Codier #codevember — 12 — My awesome work submit your work And don’t worry if you have missed a day or two, you can join in whenever!

Challenge yourself this #codevember!

In this article, I want to share my experience of creating a terraform provider from scratch, show how to add it to the Terraform and OpenTofu registries, and share some useful links and tips. I wrote the provider for Dodo pizza, and I was inspired by . Instead of Dominos, Dodo pizza doesn't allow you to create an order via API - only retrieving some data, but for learning purposes, this is more than enough. Terraform Dominos Pizza provider Which SDK to use? First of all, you need to decide which HashiCorp SDK you will use. For today HashiCorp offers 2 Go programming language Software Development Kits (SDKs) for building Terraform providers: and . For newly created providers they recommend using the framework, and for existing providers to migrate from SDKv2 to the framework. The Terraform Plugin Framework  “offers significant advantages as compared to the SDKv2”, as HashiCorp says. I decided to use the old, good SDKv2 for two reasons: almost all examples and existing providers that I know are using SDKv2, and it will be interesting for me to understand how to migrate from SDKv2 to the new framework. Terraform Plugin Framework SDKv2 How to interact with API? All articles that I found provide examples of limited sample APIs or include API calls inside the provider itself. But all “large” providers use Go libraries to interact with existing API. For example, Cloudflare provider uses their own , and GitHub provider uses library maintained by Google. And I decided to do the same - create my own library and use it in the provider. cloudflare-go go-github uses swagger, which is such a widely used tool, that I believe it should have some library generator for it. I started to google it and found about swagger-codegen - the standard way to generate client SDK for swagger. Dodo Global API an article In my case, I used only one command to generate an almost fully functional client SDK for Dodo API: swagger-codegen generate -i https://globalapi.dodopizza.com/api/swagger/v2/swagger.json -l go -o . You can find dodo-go here: https://github.com/Nmishin/dodo-go How to start writing the provider? I found several articles about terraform provider creation, from HashiCorp and independent writers: article from Mark McDonnell was the most useful for me; “Beginner's guide to creating a Terraform Provider” a good example from HashiCorp; “Writing Custom Terraform Providers” article contain not only info about terraform provider, but also a lot of useful links; “Terraform Custom Provider with Data Source” I realized that some terraform providers use API libraries very similar to my generated dodo-go library, and it was super helpful to compare my provider with existing provider, for example: . terraform-provider-authentik also can be a good example, since it’s well maintained and not as complicated as, for example, . Cloudflare terraform provider terraform-provider-aws What to do next after creating the provider? When your provider is ready, or you are happy with existing functionalities, you need to publish it to the Terraform registry and OpenTofu registry if you plan to use it with opentofu. All the needed steps are well described in HashiCorp’s documentation: https://developer.hashicorp.com/terraform/registry/providers/publishing#terraform-registry-manifest-file And this article from Kevin Wang can be useful as well: https://thekevinwang.com/2023/10/05/build-publish-terraform-provider After publishing provider to the Terraform registry, publishing to OpenTofu will be much easier, because you have already done all the necessary presteps. You need to create two issues: the first one is the provider itself, and second one is about the provider’s GPG Key. Example of adding provider: https://github.com/opentofu/registry/issues/228 Example of adding GPG Key: https://github.com/opentofu/registry/issues/230 And that's it! After your provider is added to all registries, you can add it to the config as any regular provider. Let’s check how it works for my terraform-provider-dodo. In my test configuration, I have three files and I want to retrieve data about available brands: cat provider.tf 

terraform {
  required_providers {
    dodo = {
      source = "Nmishin/dodo"
      version = "0.0.1"
    }
  }
}

provider "dodo" {} cat main.tf 

data "dodo_brand" "test" {} cat outputs.tf 

output "brands" {
  value = data.dodo_brand.test.names
} tofu plan

data.dodo_brand.test: Reading...
data.dodo_brand.test: Read complete after 1s [id=bf4b2096fab07279e4a6a9db5fb704b5]

Changes to Outputs:
  + brands = [
      + "dodopizza",
      + "doner42",
      + "drinkit",
    ]

You can apply this plan to save these new output values to the OpenTofu state, without changing any real infrastructure. Conclusion Right now everyone with basic programming skills and terraform/opentofu experience can create its own provider. There is so much information about how to start to write code and publish provider, but in this article I wanted to provide high level instruction with good useful links. And I hope my story will help you not to be afraid to start creating a provider from scratch and I am absolutely sure that you will be successful with it!

My Path for Terraform Provider Creation

How are you, hacker?
🪐 has got you covered with fresh content from our trending stories of the day! Set email preference . Want to know what's trending right now?: The Techbeat by HackerNoon here To Focus or Not Focus: How to Find the Right Product-Market Fit By [ 3 Min read ]
To all the product managers out there searching for that elusive product-market fit, remember to keep your eyes on the prize and focus on what matters most! @densmr Read More. Everything You Need to Know to Implement Generative AI for Your Enterprise By [ 8 Min read ]
The easiest way to conceptualize what is possible with Generative AI is to imagine a customized Large Language Model running inside your firewall. @minio Read More. Elevating Product Management with Marketing Insights By [ 9 Min read ]
Learn how to having marketing skills makes you a better product manager and how they help build great products. @dashadobrego Read More. Breaking the CPO Bubble: My Strategy for Staying Close to Our Products By [ 3 Min read ]
Explore a CPO's journey in "Breaking the CPO Bubble," detailing strategies for connecting with products and teams and enhancing customer alignment. @densmr Read More. How to Manage a Product Backlog in a Fast-Growing B2B Startup By [ 7 Min read ]
In this article, I’d like to share my tips for working as a product manager in fast-growing startups. I’ll also explain how these tips could be used in B2B prod @dashadobrego Read More. How to Teach People to Make the Right Decisions and Delegate Important Business Tasks With no Stress By [ 6 Min read ]
Discover effective strategies for decision-making and delegation in business, fostering a culture of responsibility and continuous improvement. @densmr Read More. Is AGI Getting Closer? Anthropic's Claude 3 Opus Model Shows Glimmers of Metacognitive Reasoning By [ 5 Min read ]
Anthropic's new LLM told prompters it knew they were testing it @mikeyoung44 Read More. Mixing the Cross-Chain magic with Tokenized Real Worlds Assets By [ 4 Min read ]
The future of finance is with tokenized RWAs and cross-chain DeFi; dig into real world assets on Avalanche and ICON. @helloicon Read More. Sam Mangel to Sam Bankman-Fried: Acceptance is the Path Forward By [ 4 Min read ]
The approach SBF’s legal team are taking toward sentence mitigation makes one thing clear: the former FTX CEO has not yet accepted what he did was wrong. @drewchapin Read More. 7 Monetization Options to Consider for Your Blog By [ 5 Min read ]
Looking for ways to generate stable revenue through your blog? The article showcases a few noteworthy options for you to consider. @griffinjt Read More. How To Create a Static, Personal Website for Less Than $1 a Month By [ 12 Min read ]
The goal of this post is to show you how you can create your own static personal website for under $1 a month. @afrocloud Read More. GitHub Copilot Review: Does it Really Give a 55% Speed Boost to Development? By [ 7 Min read ]
This article presents a GitHub Copilot review and assesses its impact on enhancing development speed in real-life scenarios. @elekssoftware Read More. Chrome Password Manager Betrayed My Trust 13 Years Ago. I Never Forgot By [ 10 Min read ]
Chrome's broken trust led to Neulock: a password manager built on user control, not encryption & blind faith. Learn the 3 security principles behind it. @lucasneves Read More. Tokenizing Real-World Assets (RWA) Part 1: A Deep Dive into ERC-3643 By [ 6 Min read ]
Discover the concept of  Real-World Assets (RWA) tokenization and its benefits, along with an examination of the ERC-3643 standard tailored for this purpose. @leonidcryptonstudio Read More. After 6 Months of Working on a CodeGen Dev Tool (GPT Pilot), This Is What I Learned By [ 13 Min read ]
What I learned in 6 months of working on the most powerful CodeGen dev tool GPT Pilot @zvone187 Read More. From Leadership to Lines of Code: A Team Leader's Guide By [ 8 Min read ]
To balance leading and coding, block time for tech tasks, prioritize, and delegate. Embrace continuous learning and personal projects to stay sharp. @lookingforere Read More. 100 Days of AI, Day 13: How Instruction Finetuning Improves a Pre-trained LLM By [ 6 Min read ]
Instruction finetuning is a finetuning technique used to convert base models like GPT3 into a chat-gpt like product. @sindamnataraj Read More. How to Use RunLoop in IOS Applications By [ 5 Min read ]
Run Loop is a very important tool in iOS development. Let's explore how to use it. @alekseimarinin Read More. Top Crypto To Buy This Month: BlockDAG's Presale Hits $3M, Challenging Kaspa Amid Cosmos' ATOM Rise By [ 3 Min read ]
Discover BlockDAG, hailed as the 'Kaspa Killer,' breaks all presale records, reaching $3 million and grabbing the attention of Cosmos investors. @btcwire Read More. Testing in Godot: How I Personally Approach It By [ 4 Min read ]
Some time ago, I participated in a game jam and made this game – Of Mice and Bad Choices, I think this is a good opportunity how automated testing can be done 🧑‍💻 What happened in your world this week? It's been said that , , . Feeling stuck? We got you covered ⬇️⬇️⬇️ We hope you enjoy this worth of free reading material. Feel free to forward this email to a nerdy friend who'll love you for it.
See you on Planet Internet! With love,
The HackerNoon Team ✌️ @dlowl Read More. writing can help consolidate technical knowledge establish credibility and contribute to emerging community standards ANSWER THESE GREATEST INTERVIEW QUESTIONS OF ALL TIME

The TechBeat: To Focus or Not Focus: How to Find the Right Product-Market Fit (3/10/2024)

How are you, hacker?
🪐 has got you covered with fresh content from our trending stories of the day! Set email preference . Want to know what's trending right now?: The Techbeat by HackerNoon here To Focus or Not Focus: How to Find the Right Product-Market Fit By [ 3 Min read ]
To all the product managers out there searching for that elusive product-market fit, remember to keep your eyes on the prize and focus on what matters most! @densmr Read More. Elevating Product Management with Marketing Insights By [ 9 Min read ]
Learn how to having marketing skills makes you a better product manager and how they help build great products. @dashadobrego Read More. Breaking the CPO Bubble: My Strategy for Staying Close to Our Products By [ 3 Min read ]
Explore a CPO's journey in "Breaking the CPO Bubble," detailing strategies for connecting with products and teams and enhancing customer alignment. @densmr Read More. How to Manage a Product Backlog in a Fast-Growing B2B Startup By [ 7 Min read ]
In this article, I’d like to share my tips for working as a product manager in fast-growing startups. I’ll also explain how these tips could be used in B2B prod @dashadobrego Read More. Everything You Need to Know to Implement Generative AI for Your Enterprise By [ 8 Min read ]
The easiest way to conceptualize what is possible with Generative AI is to imagine a customized Large Language Model running inside your firewall. @minio Read More. How to Teach People to Make the Right Decisions and Delegate Important Business Tasks With no Stress By [ 6 Min read ]
Discover effective strategies for decision-making and delegation in business, fostering a culture of responsibility and continuous improvement. @densmr Read More. Is AGI Getting Closer? Anthropic's Claude 3 Opus Model Shows Glimmers of Metacognitive Reasoning By [ 5 Min read ]
Anthropic's new LLM told prompters it knew they were testing it @mikeyoung44 Read More. Nintendo Breaks Its Silence On Palworld By [ 4 Min read ]
Nintendo finally broke its silence after Palworld’s launch raised controversy among the Pokemon community. @playerauctions Read More. Sam Mangel to Sam Bankman-Fried: Acceptance is the Path Forward By [ 4 Min read ]
The approach SBF’s legal team are taking toward sentence mitigation makes one thing clear: the former FTX CEO has not yet accepted what he did was wrong. @drewchapin Read More. Data Lake Mysteries Revealed: Nessie, Dremio, and MinIO Make Waves By [ 6 Min read ]
Let's see how Nessie, Dremio and MinIO work together to enhance data quality and collaboration in your data engineering workflows. @minio Read More. 7 Monetization Options to Consider for Your Blog By [ 5 Min read ]
Looking for ways to generate stable revenue through your blog? The article showcases a few noteworthy options for you to consider. @griffinjt Read More. 5 Low-Cost Cybersecurity Measures SMBs Can’t Afford to Not Implement By [ 4 Min read ]
A successful cyber attack can be devastating; many cybersecurity measures that can help SMBs can be implemented at surprisingly little cost. @jonstojanmedia Read More. GitHub Copilot Review: Does it Really Give a 55% Speed Boost to Development? By [ 7 Min read ]
This article presents a GitHub Copilot review and assesses its impact on enhancing development speed in real-life scenarios. @elekssoftware Read More. After 6 Months of Working on a CodeGen Dev Tool (GPT Pilot), This Is What I Learned By [ 13 Min read ]
What I learned in 6 months of working on the most powerful CodeGen dev tool GPT Pilot @zvone187 Read More. What is the Language Processing Unit (LPU)? Is It GPU's Rival? By [ 7 Min read ]
So: What is LPU, how does it work, and where is Groq (such an unfortunate name, given Musk’s Grok is all over the media) coming from? @kseniase Read More. How to Analyze Anything - Master Data Analysis With ChatGPT (Beginner's Tutorial) By [ 3 Min read ]
Today, we’re diving into an exciting feature within ChatGPT that has the potential to enhance your productivity by 10, 20, 30, or even 40%. @proflead Read More. Chrome Password Manager Betrayed My Trust 13 Years Ago. I Never Forgot By [ 10 Min read ]
Chrome's broken trust led to Neulock: a password manager built on user control, not encryption & blind faith. Learn the 3 security principles behind it. @lucasneves Read More. Mixing the Cross-Chain magic with Tokenized Real Worlds Assets By [ 4 Min read ]
The future of finance is with tokenized RWAs and cross-chain DeFi; dig into real world assets on Avalanche and ICON. @helloicon Read More. Tokenizing Real-World Assets (RWA) Part 1: A Deep Dive into ERC-3643 By [ 6 Min read ]
Discover the concept of  Real-World Assets (RWA) tokenization and its benefits, along with an examination of the ERC-3643 standard tailored for this purpose. @leonidcryptonstudio Read More. Bitcoin's Surge Past $60,000: Rally or Short-Lived Surge? By [ 4 Min read ]
Bitcoin has crossed the $60,000 mark, and it looks like this is not the limit yet. How sustainable is this growth and what to expect further? 🧑‍💻 What happened in your world this week? It's been said that , , . Feeling stuck? We got you covered ⬇️⬇️⬇️ We hope you enjoy this worth of free reading material. Feel free to forward this email to a nerdy friend who'll love you for it.
See you on Planet Internet! With love,
The HackerNoon Team ✌️ @dshishov Read More. writing can help consolidate technical knowledge establish credibility and contribute to emerging community standards ANSWER THESE GREATEST INTERVIEW QUESTIONS OF ALL TIME

The TechBeat: To Focus or Not Focus: How to Find the Right Product-Market Fit (3/8/2024)

Generate and Pray: Using SALLMS to Evaluate the Security: Conclusion & References

This paper is available on arxiv under CC 4.0 license. Authors: (1) Mohammed Latif Siddiq, Department of Computer Science and Engineering, University of Notre Dame, Notre Dame; (2) Joanna C. S. Santos, Department of Computer Science and Engineering, University of Notre Dame, Notre Dame. Table of Links Abstract & Introduction Background and Motivation Our Framework: SALLM Experiments Results Limitations and Threats to the Validity Related Work Conclusion & References 5 Results The next subsections describe the results and provide an answer to each of our RQs. 5.1 RQ1 Results Table 1 contrasts each dataset, including our framework’s dataset (denoted by SALLM on this table). CWE Coverage As shown in this table, our dataset covers 2.5 times more CWEs (45 CWEs) than LLMSecEval [67], which covers only 18 CWEs (a subset of the CWE top 25 [42]. In contrast, SecurityEval [63] covers 69 CWEs, whereas SALLM’s dataset has a slightly less amount of CWEs. Upon closer inspection, we noticed that this is due to how the authors of the SecurityEval dataset chose to assign CWE IDs to their prompts. The CWE list includes hierarchical relationships (e.g., CWE-89: SQL Injection is a child of CWE943: Improper Neutralization of Special Elements in Data Query Logic). In our dataset, we deliberately chose to map to CWE IDs that were at the lowest level of the CWE hierarchy (as more specialized as possible), unlike SecurityEval, which would have prompts tagged with higher-level abstraction CWE when a more specific one was available. Dataset Size As shown in this table, LLMSecEval has prompts instructing an LLM to generate C code and Python code. Out of their 150 prompts, only 83 of them are for Python. Unlike our dataset, their prompts are natural language prompts in the form of “Generate [language] code for the following: [coding problem description]”. Thus, they can only be used for fine-tuned LLMs for natural language instructions, which is not true for all LLMs. For example, StarCoder [35] is an LLM that was not trained for natural language prompts and, as a result, is unable to understand prompts in the form of "Write a Python code that parses a CSV file.". It is also important to highlight that although SecurityEval has more prompts than SALLM’s dataset, its dataset size in the number of tokens is smaller than ours. SALLM’s dataset prompts have an average of 265 tokens, whereas SecurityEval has 157 tokens on average. Moreover, we also found several prompts that were not compilable because they required external libraries or were single scripts part of a codebase (e.g., a Django application). 5.2 RQ2 Results In this section we report the results of running our assessment techniques for the code generated by the studied LLMs. Table 2 presents the vulnerable@k and secure@k computed based on the outcomes from SALLM’s assessment technique. The numbers in dark green are those that had the best performance for a given metric; the numbers in dark red are those in which the model had the worst performance. Recall that for the vulnerable@k metric, a lower value is better. As shown in this table, the vulnerable@k varied from 16% to 59%. For temperature 0, all models had the same value for their vulnerable@1, vulnerable@3, and vulnerable@5 as well as their secure@1, secure@3, and secure@5. The explanation for this observation is that the temperature 0 makes the results more predictable, i.e., the generated output has less variance. From these results, we also found that, on one hand, StarCoder was the best performing LLM with respect to secure code generation. It had the lowest vulnerable@k across all temperatures. On the other hand, CodeGen-2B and CodeGen2.5-7B had a worse performance, on average, than the other LLMs. For the GPT-style models, GPT-4 performed better than GPT-3.5-Turbo. 5.3 RQ3 Results We collected 423 compilable Python samples from the ChatGPT-generated code using Developers’ conversationstyle prompts. We run CodeQL to check vulnerable APIs and taint analysis on the generated code. In table 3, we presented the CWEs CodeQL found and the number of vulnerabilities in each CWE. CodeQL found 10 types of CWEs across 12 Python samples. CWE 312: Cleartext Storage of Sensitive Information is the most common occurrence in the generated Python codes. Out of 10 types of CWEs, four CWEs are in the top 25 CWE ranks in 2023 of these 10 CWEs. There is also noticeable no injection based CWE i.e., OS, Path, or SQL Injection. Upon further inspection of the CodeQL output, we found that ChatGPT uses a pseudo-random generator to generate security-sensitive values. This random generator can limit the search space and generate duplicate values, which the hackers can exploit. Another common issue we found was flask applications running in debug mode. Though it is helpful for the preproduction phase, debug information can leak sensitive information, and ChatGPT generates code where debugging is on for the Flask application. We also found that ChatGPT generates a logging code where the sensitive information is not encrypted or hashed. This sensitive information can be used to exploit an application. It also provides hard-coded credentials in the code. Users should modify them before using the code in their application.

Generate and Pray: Using SALLMS to Evaluate the Security of LLM Generated Code: Results

This paper is available on arxiv under CC 4.0 license. Authors: (1) Mohammed Latif Siddiq, Department of Computer Science and Engineering, University of Notre Dame, Notre Dame; (2) Joanna C. S. Santos, Department of Computer Science and Engineering, University of Notre Dame, Notre Dame. Table of Links Abstract & Introduction Background and Motivation Our Framework: SALLM Experiments Results Limitations and Threats to the Validity Related Work Conclusion & References 4 Experiments This section describes the research questions we address in our experiments (§ 4.1) as well as the methodology to answer each of these questions (§ 4.2–4.4). 4.1 Research Questions We aim to answer the following questions: RQ1 How does SALLM compare to existing datasets? First, we demonstrate the value of our manually curated dataset of prompts by comparing it to two existing datasets: LLMSecEval [67] and SecurityEval [63]. The goal is to contrast the coverage of vulnerability types (CWEs) and dataset size. RQ2 How do LLMs perform with security-centric prompts compared to the evaluation setting used in the original studies? As explained in Section 2.3, LLMs are evaluated with respect to their ability to generate functional code (not necessarily secure). Thus, in this question, we evaluate the models’ performance on the datasets they originally used and compare them to their performance in our dataset. RQ3 How can we use SALLM’s assessment techniques to prevent vulnerable generated code from being integrated into the code base? This research question explores the usage of our assessment techniques to detect the vulnerable code generated by the model integrated into the code base. To answer this question, we obtain a dataset [71] of code snippets generated by ChatGPT that were publicly shared on GitHub commits or inside a source code comment. 4.2 RQ1 Methodology To answer our first research question, we compare SALLM’s dataset to two prior datasets of prompts used to evaluate the security of LLM generated code: [63]: It is a prompt-based dataset covering 69 CWEs, including the MITRE’s Top 25 CWEs with 121 Python prompts from a diverse source. The prompts are signatures of Python functions along with their docstrings and import statements. • SecurityEval dataset l dataset [67]: it is a natural language (NL) prompt-to-code dataset crafted from Pearce et al. [51]. This dataset covers MITRE’s top 25 CWEs and contains 150 NL prompts to benchmark the code generation model. • LLMSecEva We compare these datasets according to two dimensions: (I) (CWEs); and (II) number of supported vulnerability types dataset size (number of prompts). 4.3 RQ2 Methodology We investigate in RQ2 the performance of existing LLMs when evaluated using SALLM, our framework. To answer this question, we provide each of the 100 prompts in our dataset as inputs to four models from three LLM families: [47] is an LLM for code generation trained on three large code datasets. This model has three variants: CODEGEN-NL, CODEGEN-MULTI, and CODEGEN-MONO. CODEGEN-NL is trained with the Pile dataset [17] is focused on text generation. The CODEGEN-MULTI is built on top of CODEGEN-NL but further trained with a large scale-dataset of code snippets in six different languages (i.e., C, C++, Go, Java, JavaScript, and Python) [27]. The CODEGEN-MONO is built from CODEGEN-MULTI and further trained with a dataset [47] of only Python code snippets. They also released another version called CODEGEN2.5 [46] which is trained on the StarCoder data from BigCode [31]. It has a mono and multi version. Since the latter variant is focused on Python-only generation, we use CODEGEN-2B-MONO and CODEGEN-2.5-7B-MONO to generate Python code. • CODEGEN [35] is an LLM with 15.5B parameters trained with over 80 different programming languages. This model is focused on fill-in-the-middle objectives and can complete code given a code-based prompt. • STARCODER [8] is a family of transformer-based [68] and task-agnostic models capable of both understanding and generating natural language. We used the latest OpenAI’s GPT models, i.e., GPT-3.5-TURBO and GPT-4, which are tuned for chat-style conversation and powers a popular chat-based question-answering tool, ChatGPT [2] and its paid variant (ChatGPT plus). • The GENERATIVE PRE-TRAINED MODEL (GPT) For each model, we generate code solutions for each prompt with new tokens and varying temperature from 0 to 1 by increasing by 0.2 (i.e., 0.0, 0.2, 0.4, 0.6, 0.8, and 1.0). We selected 256 as the token size to generate because we observed that the insecure code in our dataset has an average of 54 tokens and a maximum of 245 tokens. Thus, a 256 token size would be sufficient for the models. In the case of the GPT models, however, we made the token size double this value (i.e., 512) because they can generate an explanation along with the code (which would consume tokens). 10 256 After obtaining the generated code solutions from each model, we measure and contrast the performance of these models with respect to three metrics: pass@k [10], vulnerable@k and secure@k (the last two, are our novel metrics, as defined in Section 3.3.2). In our experiments, we choose k to be equal to 1, 3, and 5. This is because our goal is to evaluate these models for typical use scenarios, where developers will likely inspect only the first few generated code snippets by a model. 4.4 RQ3 Methodology In this question, we investigate to what extent the assessment techniques in SALLM could help engineers identify code generated with vulnerabilities. To answer this RQ, we collect code snippets generated by ChatGPT from the DevGPT dataset [71]. This dataset contains over 17,000 prompts written by engineers that were publicly shared on GitHub or HackerNews. This dataset was constructed by finding ChatGPT links (i.e., URLs in the format https:\\chat.openai.com/share/) from these different sources. The search was performed in July and August 2023. Once their web crawler identifies a ChatGPT sharing link, it extracts the code generated by ChatGPT and the corresponding prompt used by the developer to generate it. After collecting these sharing links, we analyzed their metadata to identify which links are for prompts that request the generation of python code. As a result, we obtained a total 437 Python code samples generated by ChatGPT. For each of these 437, we performed a filtering step where we disregarded samples with compilation errors. Since we found 14 samples that were not compilable,. we excluded those, obtaining a total of Python code samples. 423 After extracting these Python codes generated by ChatGPT, we run our static analyzer-based assessment technique for each. In our study, we investigate to what extent our techniques can identify which code snippets are vulnerable / not vulnerable.

Generate and Pray: Using SALLMS to Evaluate the Security of LLM Generated Code: Experiments

This paper is available on arxiv under CC 4.0 license. Authors: (1) Mohammed Latif Siddiq, Department of Computer Science and Engineering, University of Notre Dame, Notre Dame; (2) Joanna C. S. Santos, Department of Computer Science and Engineering, University of Notre Dame, Notre Dame. Table of Links Abstract & Introduction Background and Motivation Our Framework: SALLM Experiments Results Limitations and Threats to the Validity Related Work Conclusion & References Abstract With the growing popularity of Large Language Models (e.g., GitHub Copilot, ChatGPT, etc.) in software engineers’ daily practices, it is important to ensure that the code generated by these tools is not only functionally correct but also free of vulnerabilities. Although LLMs can help developers to be more productive, prior empirical studies have shown that LLMs can generate insecure code. There are two contributing factors to the insecure code generation. First, existing datasets used to evaluate Large Language Models (LLMs) do not adequately represent genuine software engineering tasks sensitive to security. Instead, they are often based on competitive programming challenges or classroom-type coding tasks. In real-world applications, the code produced is integrated into larger codebases, introducing potential security risks. There’s a clear absence of benchmarks that focus on evaluating the security of the generated code. Second, existing evaluation metrics primarily focus on the functional correctness of the generated code while ignoring security considerations. Metrics such as pass@k gauge the probability of obtaining the correct code in the top k suggestions. Other popular metrics like BLEU, CodeBLEU, ROUGE, and METEOR similarly emphasize functional accuracy, neglecting security implications. In light of these research gaps, in this paper, we described SALLM, a framework to benchmark LLMs’ abilities to generate secure code systematically. This framework has three major components: a novel dataset of security-centric Python prompts, an evaluation environment to test the generated code, and novel metrics to evaluate the models’ performance from the perspective of secure code generation. 1 Introduction A is a Large Language Model (LLM) that has been trained on a large dataset consisting of both text and code. As a result, code LLMs can generate code written in a specific programming language from a given prompt. These prompts provide a high-level specification of a developer’s intent [34]. code LLM Prompts can include single/multi-line code comments, code expressions (e.g., a function definition), text, or a combination of these. etc. Given a prompt as input, the LLM generates new tokens, one by one, until it reaches a stop sequence (i.e., a pre-configured sequence of tokens) or the maximum number of tokens is reached. With the recent releases of GitHub Copilot [25] and ChatGPT [2], LLM-based source code generation tools are increasingly being used by developers in order to reduce software development efforts [77]. A recent survey with 500 US-based developers who work for large-sized companies showed that of them are using LLMs to generate code for work and personal use [60]. Part of this fast widespread adoption is due to the increased productivity perceived by developers; LLMs help them to automate repetitive tasks so that they can focus on higher-level challenging tasks [77]. 92% Although LLM-based code generation techniques may produce functionally correct code, prior works showed that they can also generate code with vulnerabilities and security smells [51, 52, 58]. A prior study has also demonstrated that training sets commonly used to train and/or fine-tune LLMs contain harmful coding patterns, which leak to the generated code [62]. Moreover, a recent study [52] with 47 participants showed that individuals who used the codex-davinci-002 LLM wrote code that was less secure compared to those who did not use it. Even worse, participants who used the LLM were more likely to believe that their code was secure, unlike their peers who did not use the LLM to write code. There are two major factors contributing to this unsafe code generation. First, code LLMs are evaluated using benchmarks, which do not include constructs to evaluate the security of the generated code [63, 75]. Second, existing evaluation metrics (e.g., pass@k [11], CodeBLEU [56], etc.) assess models’ performance with respect to their ability to produce functionally correct code while ignoring security concerns. Therefore, the performance reported for these models overly focuses on improving the precision of the generated code with respect to passing the test cases of these benchmarks without evaluating the of the produced code. functional security The contributions of this paper are: • A novel framework to ; systematically and automatically evaluate the security of LLM generated code • A publicly available dataset of Python prompts [1]; • Two novel metrics (secure@k and vulnerability@k) and a demonstration of how to compute it statically and dynamically. • A benchmarking of five LLMs (CodeGen-2B-mono, CodeGen-2.5-7B-mono, StarCoder, GPT-3.5, and GPT4) using our framework. The rest of this paper is organized as follows: Section 2 introduces the core concepts necessary to understand this paper. Section 3 describes our framework in detail. Section 4 describes the empirical investigation we performed to benchmark LLMs. Section 5 presents the results of our experiments. Section 6 explains SALLM’s limitations. Section 7 presents related work. Finally, Section 8 concludes this paper while describing plans for future work. [1] The dataset will be made public on GitHub upon acceptance and submitted to the artifact evaluation track.

Generate and Pray: Using SALLMS to Evaluate the Security of LLM Generated Code: Abstract & Intro

Authors: (1) Savvas Petridis, Google Research, New York, New York, USA; (2) Ben Wedin, Google Research, Cambridge, Massachusetts, USA; (3) James Wexler, Google Research, Cambridge, Massachusetts, USA; (4) Aaron Donsbach, Google Research, Seattle, Washington, USA; (5) Mahima Pushkarna, Google Research, Cambridge, Massachusetts, USA; (6) Nitesh Goyal, Google Research, New York, New York, USA; (7) Carrie J. Cai, Google Research, Mountain View, California, USA; (8) Michael Terry, Google Research, Cambridge, Massachusetts, USA. Table Of Links Abstract & Introduction Related Work Formative Study Constitution Maker Implementation User Study Findings Discussion Conclusion and References 9 CONCLUSION This paper presents ConstitutionMaker, a tool for interactively refining LLM outputs by converting users’ intuitive feedback into principles. ConstitutionMaker’s design is informed by a formative study, where we also collected and classified the types of principles users wanted to write. ConstitutionMaker incorporates three principle-elicitation features: kudos, critique, and rewrite. In a user study with 14 industry professionals, participants felt that ConstitutionMaker helped them (1) write principles that effectively guided the chatbot, (2) convert their feedback into principles more easily, and (3) write principles more efficiently, with (4) less mental demand than the baseline. This was due to ConstitutionMaker supporting their thought processes, including helping them to: identify ways to improve the bot’s responses, convert their intuition into verbal feedback, and phrase their feedback as specific principles. There are many avenues of future work, including supporting users in iterating on and clarifying their principles, organizing larger sets of principles and supporting multiple writers, and helping users test chatbots across multiple user journeys. Together, these findings inform future tools that support interactively customizing LLM outputs. REFERENCES [1] Yuntao Bai, Saurav Kadavath, Sandipan Kundu, Amanda Askell, Jackson Kernion, Andy Jones, Anna Chen, Anna Goldie, Azalia Mirhoseini, Cameron McKinnon, Carol Chen, Catherine Olsson, Christopher Olah, Danny Hernandez, Dawn Drain, Deep Ganguli, Dustin Li, Eli Tran-Johnson, Ethan Perez, Jamie Kerr, Jared Mueller, Jeffrey Ladish, Joshua Landau, Kamal Ndousse, Kamile Lukosuite, Liane Lovitt, Michael Sellitto, Nelson Elhage, Nicholas Schiefer, Noemi Mercado, Nova DasSarma, Robert Lasenby, Robin Larson, Sam Ringer, Scott Johnston, Shauna Kravec, Sheer El Showk, Stanislav Fort, Tamera Lanham, Timothy TelleenLawton, Tom Conerly, Tom Henighan, Tristan Hume, Samuel R. Bowman, Zac Hatfield-Dodds, Ben Mann, Dario Amodei, Nicholas Joseph, Sam McCandlish, Tom Brown, and Jared Kaplan. 2022. Constitutional AI: Harmlessness from AI Feedback. arXiv:2212.08073 [cs.CL] [2] Svetlin Bostandjiev, John O’Donovan, and Tobias Höllerer. 2012. TasteWeights: A Visual Interactive Hybrid Recommender System. In Proceedings of the Sixth ACM Conference on Recommender Systems (Dublin, Ireland) (RecSys ’12). Association for Computing Machinery, New York, NY, USA, 35–42. https://doi.org/10.1145/ 2365952.2365964 [3] Stephen Brade, Bryan Wang, Mauricio Sousa, Sageev Oore, and Tovi Grossman. 2023. Promptify: Text-to-Image Generation through Interactive Prompt Exploration with Large Language Models. In Proceedings of the 35th Annual ACM Symposium on User Interface Software and Technology (UIST). https: //doi.org/10.1145/3586183.3606725 arXiv:2304.09337 [cs.HC] [4] Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert-Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel Ziegler, Jeffrey Wu, Clemens Winter, Chris Hesse, Mark Chen, Eric Sigler, Mateusz Litwin, Scott Gray, Benjamin Chess, Jack Clark, Christopher Berner, Sam McCandlish, Alec Radford, Ilya Sutskever, and Dario Amodei. 2020. Language Models are Few-Shot Learners. In Advances in Neural Information Processing Systems, H. Larochelle, M. Ranzato, R. Hadsell, M.F. Balcan, and H. Lin (Eds.), Vol. 33. Curran Associates, Inc., 1877–1901. https://proceedings.neurips.cc/paper_files/paper/2020/file/ 1457c0d6bfcb4967418bfb8ac142f64a-Paper.pdf [5] Sébastien Bubeck, Varun Chandrasekaran, Ronen Eldan, Johannes Gehrke, Eric Horvitz, Ece Kamar, Peter Lee, Yin Tat Lee, Yuanzhi Li, Scott Lundberg, Harsha Nori, Hamid Palangi, Marco Tulio Ribeiro, and Yi Zhang. 2023. Sparks of Artificial General Intelligence: Early experiments with GPT-4. arXiv:2303.12712 [cs.CL] [6] Ernie Chang, Xiaoyu Shen, Hui-Syuan Yeh, and Vera Demberg. 2021. On Training Instance Selection for Few-Shot Neural Text Generation. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 2: Short Papers). Association for Computational Linguistics, Online, 8–13. https: //doi.org/10.18653/v1/2021.acl-short.2 [7] Qiaochu Chen, Xinyu Wang, Xi Ye, Greg Durrett, and Isil Dillig. 2020. MultiModal Synthesis of Regular Expressions. In Proceedings of the 41st ACM SIGPLAN Conference on Programming Language Design and Implementation (London, UK) (PLDI 2020). Association for Computing Machinery, New York, NY, USA, 487–502. https://doi.org/10.1145/3385412.3385988 [8] Katy Ilonka Gero, Vivian Liu, and Lydia Chilton. 2022. Sparks: Inspiration for Science Writing Using Language Models. In Proceedings of the 2022 ACM Designing Interactive Systems Conference (Virtual Event, Australia) (DIS ’22). Association for Computing Machinery, New York, NY, USA, 1002–1019. https: //doi.org/10.1145/3532106.3533533 [9] Sandra G. Hart and Lowell E. Staveland. 1988. Development of NASA-TLX (Task Load Index): Results of Empirical and Theoretical Research. In Human Mental Workload, Peter A. Hancock and Najmedin Meshkati (Eds.). Advances in Psychology, Vol. 52. North-Holland, 139–183. https://doi.org/10.1016/S0166- 4115(08)62386-9 [10] Jeremy Howard and Sebastian Ruder. 2018. Universal Language Model Finetuning for Text Classification. In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). Association for Computational Linguistics, Melbourne, Australia, 328–339. https://doi.org/ 10.18653/v1/P18-1031 [11] Jiyou Jia. 2009. CSIEC: A computer assisted English learning chatbot based on textual knowledge and reasoning. Knowledge-Based Systems 22, 4 (2009), 249–255. https://doi.org/10.1016/j.knosys.2008.09.001 Artificial Intelligence (AI) in Blended Learning. [12] Ellen Jiang, Kristen Olson, Edwin Toh, Alejandra Molina, Aaron Donsbach, Michael Terry, and Carrie J Cai. 2022. PromptMaker: Prompt-Based Prototyping with Large Language Models. In Extended Abstracts of the 2022 CHI Conference on Human Factors in Computing Systems (New Orleans, LA, USA) (CHI EA ’22). Association for Computing Machinery, New York, NY, USA, Article 35, 8 pages. https://doi.org/10.1145/3491101.3503564 [13] Ellen Jiang, Edwin Toh, Alejandra Molina, Aaron Donsbach, Carrie J Cai, and Michael Terry. 2021. GenLine and GenForm: Two Tools for Interacting with Generative Language Models in a Code Editor. In Adjunct Proceedings of the 34th Annual ACM Symposium on User Interface Software and Technology (Virtual Event, USA) (UIST ’21 Adjunct). Association for Computing Machinery, New York, NY, USA, 145–147. https://doi.org/10.1145/3474349.3480209 [14] Ellen Jiang, Edwin Toh, Alejandra Molina, Kristen Olson, Claire Kayacik, Aaron Donsbach, Carrie J Cai, and Michael Terry. 2022. Discovering the Syntax and Strategies of Natural Language Programming with Generative Language Models. In Proceedings of the 2022 CHI Conference on Human Factors in Computing Systems (New Orleans, LA, USA) (CHI ’22). Association for Computing Machinery, New York, NY, USA, Article 386, 19 pages. https://doi.org/10.1145/3491102.3501870 [15] Daniel Kahneman. 2011. Thinking, fast and slow. macmillan. [16] Tae Soo Kim, Yoonjoo Lee, Minsuk Chang, and Juho Kim. 2023. Cells, Generators, and Lenses: Design Framework for Object-Oriented Interaction with Large Language Models. In Proceedings of the 36th Annual ACM Symposium on User Interface Software and Technology. ACM. 1–18. [17] Johannes Kunkel, Benedikt Loepp, and Jürgen Ziegler. 2017. A 3D Item Space Visualization for Presenting and Manipulating User Preferences in Collaborative Filtering. In Proceedings of the 22nd International Conference on Intelligent User Interfaces (Limassol, Cyprus) (IUI ’17). Association for Computing Machinery, New York, NY, USA, 3–15. https://doi.org/10.1145/3025171.3025189 [18] Peter Lee, Sebastien Bubeck, and Joseph Petro. 2023. Benefits, Limits, and Risks of GPT-4 as an AI Chatbot for Medicine. New England Journal of Medicine 388, 13 (2023), 1233–1239. https://doi.org/10.1056/NEJMsr2214184 arXiv:https://doi.org/10.1056/NEJMsr2214184 PMID: 36988602. [19] Brian Lester, Rami Al-Rfou, and Noah Constant. 2021. The Power of Scale for Parameter-Efficient Prompt Tuning. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics, Online and Punta Cana, Dominican Republic, 3045–3059. https: //doi.org/10.18653/v1/2021.emnlp-main.243 [20] Michael Xieyang Liu, Advait Sarkar, Carina Negreanu, Benjamin Zorn, Jack Williams, Neil Toronto, and Andrew D. Gordon. 2023. “What It Wants Me To Say”: Bridging the Abstraction Gap Between End-User Programmers and CodeGenerating Large Language Models. In Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems (Hamburg, Germany) (CHI ’23). Association for Computing Machinery, New York, NY, USA, Article 598, 31 pages. https: //doi.org/10.1145/3544548.3580817 [21] Vivian Liu, Han Qiao, and Lydia Chilton. 2022. Opal: Multimodal Image Generation for News Illustration. In Proceedings of the 35th Annual ACM Symposium on User Interface Software and Technology (Bend, OR, USA) (UIST ’22). Association for Computing Machinery, New York, NY, USA, Article 73, 17 pages. https://doi.org/10.1145/3526113.3545621 [22] Vivian Liu, Jo Vermeulen, George Fitzmaurice, and Justin Matejka. 2023. 3DALLE: Integrating Text-to-Image AI in 3D Design Workflows. In Proceedings of the 2023 ACM Designing Interactive Systems Conference (Pittsburgh, PA, USA) (DIS ’23). Association for Computing Machinery, New York, NY, USA, 1955–1977. https://doi.org/10.1145/3563657.3596098 [23] Aditi Mishra, Utkarsh Soni, Anjana Arunkumar, Jinbin Huang, Bum Chul Kwon, and Chris Bryan. 2023. PromptAid: Prompt Exploration, Perturbation, Testing and Iteration using Visual Analytics for Large Language Models. arXiv:2304.01964 [cs.HC] [24] Joon Sung Park, Joseph C. O’Brien, Carrie J. Cai, Meredith Ringel Morris, Percy Liang, and Michael S. Bernstein. 2023. Generative Agents: Interactive Simulacra of Human Behavior. arXiv:2304.03442 [cs.HC] [25] Savvas Petridis, Nediyana Daskalova, Sarah Mennicken, Samuel F Way, Paul Lamere, and Jennifer Thom. 2022. TastePaths: Enabling Deeper Exploration and Understanding of Personal Preferences in Recommender Systems. In 27th International Conference on Intelligent User Interfaces (Helsinki, Finland) (IUI ’22). Association for Computing Machinery, New York, NY, USA, 120–133. https: //doi.org/10.1145/3490099.3511156 [26] Savvas Petridis, Nicholas Diakopoulos, Kevin Crowston, Mark Hansen, Keren Henderson, Stan Jastrzebski, Jeffrey V Nickerson, and Lydia B Chilton. 2023. AngleKindling: Supporting Journalistic Angle Ideation with Large Language Models. In Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems (Hamburg, Germany) (CHI ’23). Association for Computing Machinery, New York, NY, USA, Article 225, 16 pages. https://doi.org/10.1145/3544548. 3580907 [27] Reid Pryzant, Dan Iter, Jerry Li, Yin Tat Lee, Chenguang Zhu, and Michael Zeng. 2023. Automatic Prompt Optimization with "Gradient Descent" and Beam Search. arXiv:2305.03495 [cs.CL] [28] Alec Radford, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei, Ilya Sutskever, et al. 2019. Language models are unsupervised multitask learners. OpenAI blog 1, 8 (2019), 9. [29] Kiran Ramesh, Surya Ravishankaran, Abhishek Joshi, and K. Chandrasekaran. 2017. A Survey of Design Techniques for Conversational Agents. In Information, Communication and Computing Technology, Saroj Kaushik, Daya Gupta, Latika Kharb, and Deepak Chahal (Eds.). Springer Singapore, Singapore, 336–350. [30] Emily Reif, Minsuk Kahng, and Savvas Petridis. 2023. Visualizing Linguistic Diversity of Text Datasets Synthesized by Large Language Models. arXiv:2305.11364 [cs.CL] [31] Laria Reynolds and Kyle McDonell. 2021. Prompt Programming for Large Language Models: Beyond the Few-Shot Paradigm. In Extended Abstracts of the 2021 CHI Conference on Human Factors in Computing Systems (Yokohama, Japan) (CHI EA ’21). Association for Computing Machinery, New York, NY, USA, Article 314, 7 pages. https://doi.org/10.1145/3411763.3451760 [32] Taylor Shin, Yasaman Razeghi, Robert L. Logan IV, Eric Wallace, and Sameer Singh. 2020. AutoPrompt: Eliciting Knowledge from Language Models with Automatically Generated Prompts. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP). Association for Computational Linguistics, Online, 4222–4235. https://doi.org/10.18653/v1/2020. emnlp-main.346 [33] Hendrik Strobelt, Albert Webson, Victor Sanh, Benjamin Hoover, Johanna Beyer, Hanspeter Pfister, and Alexander M. Rush. 2023. Interactive and Visual Prompt Engineering for Ad-hoc Task Adaptation with Large Language Models. IEEE Transactions on Visualization and Computer Graphics 29, 1 (2023), 1146–1156. https://doi.org/10.1109/TVCG.2022.3209479 [34] Hongjin Su, Jungo Kasai, Chen Henry Wu, Weijia Shi, Tianlu Wang, Jiayi Xin, Rui Zhang, Mari Ostendorf, Luke Zettlemoyer, Noah A. Smith, and Tao Yu. 2022. Selective Annotation Makes Language Models Better Few-Shot Learners. arXiv:2209.01975 [cs.CL] [35] Gust Verbruggen, Vu Le, and Sumit Gulwani. 2021. Semantic Programming by Example with Pre-Trained Models. Proc. ACM Program. Lang. 5, OOPSLA, Article 100 (oct 2021), 25 pages. https://doi.org/10.1145/3485477 [36] Sitong Wang, Savvas Petridis, Taeahn Kwon, Xiaojuan Ma, and Lydia B Chilton. 2023. PopBlends: Strategies for Conceptual Blending with Large Language Models. In Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems (Hamburg, Germany) (CHI ’23). Association for Computing Machinery, New York, NY, USA, Article 435, 19 pages. https://doi.org/10.1145/3544548. [37] Yunlong Wang, Shuyuan Shen, and Brian Y Lim. 2023. RePrompt: Automatic Prompt Editing to Refine AI-Generative Art Towards Precise Expressions. In Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems (Hamburg, Germany) (CHI ’23). Association for Computing Machinery, New York, NY, USA, Article 22, 29 pages. https://doi.org/10.1145/3544548.3581402 [38] Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, brian ichter, Fei Xia, Ed Chi, Quoc V Le, and Denny Zhou. 2022. Chain-of-Thought Prompting Elicits Reasoning in Large Language Models. In Advances in Neural Information Processing Systems, S. Koyejo, S. Mohamed, A. Agarwal, D. Belgrave, K. Cho, and A. Oh (Eds.), Vol. 35. Curran Associates, Inc., 24824–24837. https://proceedings.neurips.cc/paper_files/paper/2022/file/ 9d5609613524ecf4f15af0f7b31abca4-Paper-Conference.pdf [39] Sherry Wu, Hua Shen, Daniel S Weld, Jeffrey Heer, and Marco Tulio Ribeiro. 2023. ScatterShot: Interactive In-Context Example Curation for Text Transformation. In Proceedings of the 28th International Conference on Intelligent User Interfaces (Sydney, NSW, Australia) (IUI ’23). Association for Computing Machinery, New York, NY, USA, 353–367. https://doi.org/10.1145/3581641.3584059 [40] Tongshuang Wu, Ellen Jiang, Aaron Donsbach, Jeff Gray, Alejandra Molina, Michael Terry, and Carrie J Cai. 2022. PromptChainer: Chaining Large Language Model Prompts through Visual Programming. In Extended Abstracts of the 2022 CHI Conference on Human Factors in Computing Systems (New Orleans, LA, USA) (CHI EA ’22). Association for Computing Machinery, New York, NY, USA, Article 359, 10 pages. https://doi.org/10.1145/3491101.3519729 [41] Tongshuang Wu, Michael Terry, and Carrie Jun Cai. 2022. AI Chains: Transparent and Controllable Human-AI Interaction by Chaining Large Language Model Prompts. In Proceedings of the 2022 CHI Conference on Human Factors in Computing Systems (New Orleans, LA, USA) (CHI ’22). Association for Computing Machinery, New York, NY, USA, Article 385, 22 pages. https: //doi.org/10.1145/3491102.3517582 [42] Yu Wu, Wei Wu, Chen Xing, Ming Zhou, and Zhoujun Li. 2017. Sequential Matching Network: A New Architecture for Multi-turn Response Selection in RetrievalBased Chatbots. In Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). Association for Computational Linguistics, Vancouver, Canada, 496–505. https://doi.org/10.18653/v1/P17-1046 [43] Anbang Xu, Zhe Liu, Yufan Guo, Vibha Sinha, and Rama Akkiraju. 2017. A New Chatbot for Customer Service on Social Media. In Proceedings of the 2017 CHI Conference on Human Factors in Computing Systems (Denver, Colorado, USA) (CHI ’17). Association for Computing Machinery, New York, NY, USA, 3506–3510. https://doi.org/10.1145/3025453.3025496 [44] Ann Yuan, Andy Coenen, Emily Reif, and Daphne Ippolito. 2022. Wordcraft: Story Writing With Large Language Models. In 27th International Conference on Intelligent User Interfaces (Helsinki, Finland) (IUI ’22). Association for Computing Machinery, New York, NY, USA, 841–852. https://doi.org/10.1145/3490099.3511105 [45] J.D. Zamfirescu-Pereira, Heather Wei, Amy Xiao, Kitty Gu, Grace Jung, Matthew G Lee, Bjoern Hartmann, and Qian Yang. 2023. Herding AI Cats: Lessons from Designing a Chatbot by Prompting GPT-3. In Proceedings of the 2023 ACM Designing Interactive Systems Conference (Pittsburgh, PA, USA) (DIS ’23). Association for Computing Machinery, New York, NY, USA, 2206–2220. https://doi.org/10.1145/3563657.3596138 [46] J.D. Zamfirescu-Pereira, Richmond Y. Wong, Bjoern Hartmann, and Qian Yang. 2023. Why Johnny Can’t Prompt: How Non-AI Experts Try (and Fail) to Design LLM Prompts. In Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems (Hamburg, Germany) (CHI ’23). Association for Computing Machinery, New York, NY, USA, Article 437, 21 pages. https://doi.org/10.1145/ 3544548.3581388 [47] Tianyi Zhang, London Lowmanstone, Xinyu Wang, and Elena L. Glassman. 2020. Interactive Program Synthesis by Augmented Examples. In Proceedings of the 33rd Annual ACM Symposium on User Interface Software and Technology (Virtual Event, USA) (UIST ’20). Association for Computing Machinery, New York, NY, USA, 627–648. https://doi.org/10.1145/3379337.3415900 [48] Hao Zhou, Minlie Huang, Tianyang Zhang, Xiaoyan Zhu, and Bing Liu. 2018. Emotional Chatting Machine: Emotional Conversation Generation with Internal and External Memory. arXiv:1704.01074 [cs.CL] This paper is under CC 4.0 license. available on arxiv

Interactively Critiquing LLMs by Converting Feedback into Principle— Conclusion and References

90 Stories To Learn About Startup Strategy

197 Stories To Learn About Startup Marketing

Brace yourselves folks, AI is taking over the world! Well, maybe it is not quite there yet. But still, you can’t deny the level of impact it has had over the past 12 months. The weird part of it all is that “ ” as a concept and as a field of study is not that new, but it “feels” new with all the chatbots and AI-powered tools that have come up this year. artificial intelligence More important than that is the fact that by this time last year, the only widely known AI tool was GPT-3 and then ChatGPT. But now, you can have a hard time tracking all the different AI tools, chatbots, and LLMs available. All of which fall into 2 categories: Closed source (like ChatGPT and Claude) Open source (like Falcon or Mistral) And here is where it gets interesting. As popular and highly polished as proprietary AI models can be, there are open source models making waves in the AI space and punching above their weight class. That’s what we’ll be looking at in this article. With technology as revolutionary as AI, is proprietary, black-box-like, software the way, or is open source a better option? This and other questions will be answered in today’s episode. Open or Closed? That is the question To start off, the whole scientific process has been built on principles of honesty, integrity, and transparency. It involves openness, collaboration, and peer reviews to validate findings. Many of the world’s biggest scientific advancements, like pasteurization, penicillin, and fertilizers, were possible thanks to the collaborative work of many scientists over the years. Often, they tackled a big problem where they lacked the resources at that time. They published their findings and scientists many years later used that as the foundation to develop a solution to the original problem for the benefit of mankind. And this also applies to open-source technology. The world changed when computers went from being huge machines that occupied entire rooms to devices that every home can have. And after the internet came, that was another step forward in allowing many people access to technology instead of just the privileged few. Tim Berners-Lee invented the World Wide Web in 1989 and made it freely available to everyone without any patent or royalties. This fueled the rapid growth of the internet and the many innovations that came in the next decade. A similar story happens with operating systems, think Windows vs Linux. And the same has happened with web technologies. With all these previous examples, it stands to reason that a technology as transformative as AI, can (and should) follow a similar path. So, let’s look at how both sides (closed & open-source AI) have progressed this year. The State of Proprietary AI By now, it is no news for anyone the impact that had when it was released last November. And for the rest of the year, proprietary AI became the talk of the town. ChatGPT In March of 2023, the successor to GPT-3, GPT-4 was released. That event sparked the AI race. Soon enough, Google joined the fray with . Then came Anthropic, founded by former OpenAI researchers, releasing , a contender to the popular ChatGPT. Bard Claude OpenAI is, as of now, the company with the most “hits” in the market. Those are the GPT models, different Dall-E versions, and Whisper. Microsoft is also up there with its new and improved Bing Chat (which is based on OpenAI’s technology) and the soon-to-be-included-everywhere Copilot. Google joined the race with the early research project Bard which underwhelmed everyone at first and made us pay more attention to Microsoft and its initiatives. But after that “science fair” project, Google stepped up its game and released offerings like Vertex AI, PaLM (and PaLM2), Imagen, and Codey. And then there’s Anthropic with different versions of its powerful Claude (Claude-instant, Claude 2). The interesting part is the approach they've used to train Claude, what they call "Constitutional AI". This approach puts safety at the forefront and helps create AI that is aligned with human interests and values. Those are great advancements in the AI field that are more widely known thanks to the fact that they are developed by companies that have several employees, extensive resources, and great marketing departments. Now, let’s look at the other side of the coin. The State of Open Source AI Since the release of GPT-4, not only tech giants have jumped in to take part in the AI race, but other independent projects have also emerged. Made possible by open-source ML frameworks like TensorFlow and PyTorch. Stability AI released , an alternative to Dall-E, and many tech enthusiasts have experimented extensively with its capabilities to the point that it brought up ethical concerns regarding the nature of art and creativity. Stable Diffusion Meta announced the release of a quasi-open large language model called (with several model sizes and then a second version). LLaMA That model along with Hugging Face services (like Gradio, Spaces, Transformers) sparked a revolution because for the first time, people all around the world had access to open-source technology that rivaled the likes of ChatGPT or PaLM. And do you know what happens when a group of techies, hackers, and tech enthusiasts have enough time and resources? Yeah, they can go crazy building stuff. The niche internet forums and IRC channels from the 90’s with the advent of the internet were replaced by Hugging Face discussions, GitHub issues, and Discord servers. Something else that contributed to open-source growth was the Pile dataset from EleutherAI. This initiative helped progress unsupervised and self-supervised learning, reducing the need for large labeled datasets. With the large language models, the datasets to train/finetune them, and the reduced requirements for computing, a whole ecosystem of products and services soon emerged. (When I say reduced requirements for computing, I mean that LLMs don’t need a ton of parameters to produce the quality of results generated by proprietary models, this is shown by models like and ) LLaMA 13B Mistral 7B There are a ton of projects, pre-trained and fine-tuned models alike, datasets, and tools in this space available for everyone who wants to get in and collaborate with others. We now have different types of chatbots that don’t rely on GPT-3/GPT-4 to work like Zephyr-chat, LLaMA2-chat, Mistral-instruct, and Falcon-chat. LLMs fine-tuned for code generation and assistance like , , and . Code-LLaMA CodeGen StarCoder An open-access multilingual language model called . Bloom Multimodal LLMs (that are not just text) like and . LLaVA Fuyu A Hugging Face leaderboard that evaluates and ranks all the existing open-source models. Several datasets for pre-training and fine-tuning LLMs like or . RedPajama OpenOrca And most recently we have more autonomous models called “AI agents”. The most popular ones are powered by GPT-3.5 but there are others based on LLaMA. And it seems we’re racing to build agents that don’t get stuck in loops or can finish tasks independently without spewing a bunch of text that looks convincing but it’s either inaccurate or plain wrong. There’s been a ton of progress in the past 6 months alone and you can be sure that no one front is showing signs of slowing down. Going Forward Even with all the head-spinning, fast-paced progress we’ve seen in the past year, we’re still early in AI’s development. There are several things that we need to figure out, different aspects to consider like AI privacy, ethics, inbuilt biases, and so on. As with everything in life, no one side is completely wrong and the other is right. Both proprietary and open-source AI have their pros and cons. Proprietary AI can leverage a greater amount of resources to train new and more powerful models, while providing access to people at a wider scale. But they operate like a black box, lack observability, and their interests might be more aligned to big players with money than the regular consumer. Open-source AI, on the other hand, benefits from worldwide collaboration, transparency, and open innovation. But it lacks organization, resources for more ambitious initiatives, and is at risk if stricter regulations are established. The question now is how we can keep the progress in AI going in a hybrid manner. A way in which we can collaborate jointly with some of the brightest minds in the space and with the necessary resources to drive this innovation forward in a responsible manner that puts safety and privacy at the forefront. A way in which the interests and benefits of a few don’t trump the ones of the rest of us. A way in which such revolutionary technology like AI doesn’t get privatized, restricted, or weaponized against groups of people deemed “enemies” of the greater powers. We are in a unique moment in history where the decisions that we make and the way we handle technology will determine how the future will take shape, for the better or for worse. Thanks for reading. Don’t forget to and don’t miss the coming articles. subscribe on Hackernoon

The Battle Between Proprietary and Open Source AI

মালিকানা এবং ওপেন সোর্স এআই এর মধ্যে যুদ্ধ

Der Kampf zwischen proprietärer und Open-Source-KI

La batalla entre la IA propietaria y la de código abierto

La bataille entre l’IA propriétaire et l’IA open source

मालिकाना और ओपन सोर्स एआई के बीच लड़ाई

プロプライエタリ AI とオープンソース AI の間の戦い

독점 AI와 오픈 소스 AI 간의 전투

A batalha entre IA proprietária e de código aberto

Битва между проприетарным и открытым исходным кодом ИИ

Tescilli ve Açık Kaynak Yapay Zeka Arasındaki Savaş

Cuộc chiến giữa AI độc quyền và nguồn mở

专有人工智能与开源人工智能之间的战斗

The Journey Toward Streamlined CV Management Professionals often find themselves wearing multiple hats, juggling roles, projects, or courses all at once. This bustling activity could lead to significant milestones achieved in quick succession. Hence, the need for frequent CV updates emerges, sometimes on a monthly or even weekly basis. The Maze of Multiple Versions Stumbling upon an old CV and wondering if it's the latest version is a common scenario. The presence of multiple versions can lead to confusion and sometimes, sending out an outdated CV to potential employers or collaborators. Uniformity Across Various Platforms Professionals showcase their profiles on numerous platforms like LinkedIn, personal blogs, and job portals. Ensuring a consistent portrayal across these platforms becomes a mammoth task. Any inconsistency could send mixed signals to potential employers or collaborators. Investment of Time and Effort Creating a CV isn't merely about listing achievements; it's about doing so in an engaging manner. The effort in revising wordings, realigning formats, or updating metrics can be significant and sometimes, daunting. The Need for Adaptability Tailoring CVs to specific roles or industries is advisable. However, this customization can further complicate the updating process, especially when trying to recall which skills or experiences were highlighted for which opportunity. These challenges underscore the need for a solution to simplify the CV maintenance process while ensuring professionals are aptly represented. Building an Online Presence With , , and React Gatsby Hygraph CRM The journey towards an impactful online presence commenced with the creation of . A personal website isn't merely a digital card; it's a platform narrating one's professional journey, achievements, and insights. teimurjan.dev Through design inspirations from Dribbble, a vision for the site crystallized - a digital platform that's both aesthetic and highly functional. The choice of technology was crucial. React, with its component-driven architecture, provided the required modularity and scalability. Gatsby ensured a high-performance, engaging user experience. The true game-changer, however, was the integration with Hygraph CRM, becoming the central hub for professional data like skills, projects, and accomplishments. This integration made updating the website's data a breeze, ensuring real-time accuracy. Moreover, the synergies of GraphQL, along with Hygraph, allowed for fetching and manipulating the data with incredible precision. Additionally, ensured type safety, making the data interactions dynamic, robust, and error-free. GraphQL codegen Automating CV Creation With react-pdf The traditional approach to CV maintenance is manual, involving various word processors and PDF editors. However, the modern solution harnessed by teimurjan.dev propels CV creation into the realm of automation, ensuring dynamism, accuracy, and efficiency. React Components Meet PDFs is a remarkable library, allowing the rendering of React components directly into PDF documents. This fusion enables developers to leverage the power and flexibility of React's component-based structure while producing output in the universally accepted PDF format. React-pdf It simplifies the design and layout process, providing a more intuitive, code-driven approach to document creation. Exploring Dynamic Rendering Possibilities React-pdf offers versatile rendering capabilities. Developers can render the PDF directly to a browser URL on the fly or save it as a file. This flexibility offers immense advantages: : Beneficial during the development phase, on-the-fly rendering allows developers to witness changes in real-time, streamlining the iteration process and ensuring the CV design aligns perfectly with expectations. On-the-Fly Rendering : For production purposes, file rendering ensures the CV is readily available for download, providing a static PDF file that's consistent and reliable. File Rendering Ensuring Data Consistency The CV generator and the Gatsby website both draw from the same data source, ensuring uniformity. Whether a visitor is browsing the website or downloading the CV, they receive the most up-to-date and consistent information, eliminating discrepancies and reinforcing credibility. A Glimpse Into the Structured Development Approach The teimurjan.dev repository tends to have clean and structured development. A well-organized repository facilitates easier maintenance and updates, allowing for modular development, where different components or features can be developed, tested, and integrated seamlessly. Here's a breakdown of the repository's structure: : This is the heart of the repository, housing the primary website. It's here that visitors get their first impression, navigating through the digital content and getting a feel for the professional journey, skills, and achievements. Built with Gatsby, it offers a blazing-fast user experience. @teimurjan/landing : Functioning as a distinct package, the folder is dedicated entirely to CV generation. Harnessing the power of react-pdf, this section transforms professional data into a structured, visually appealing document ready for download or on-the-fly viewing. @teimurjan/resume resume : Web development often involves ensuring consistent data types across different sections or components. The folder serves as a central repository for TypeScript typings, ensuring both the website and the CV adhere to the same data structure, eliminating discrepancies, and reinforcing data integrity. @teimurjan/gql-types gql-types : A toolkit for developers, the folder houses utility functions that streamline various tasks, ensuring code reusability and consistency. @teimurjan/utils utils Behind the Scenes: Automated Deployment and Real-time Updates The infrastructure supporting teimurjan.dev demonstrates the potency of automation and integrated systems. This setup can be broken down into a few key components: GitHub Actions: Automates the build and deployment process right from the main branch. Webhook Activation: Upon any change on Hygraph, a webhook is activated. Amazon Lambda Function: This webhook consequently triggers an Amazon Lambda function.
Automated Deployment: The Lambda function then initiates the GitHub Action deployment, guaranteeing the website is automatically updated with the latest data. For a clearer visual representation, please refer to the diagram below which illustrates this architecture. In Summary In today's fast-paced professional world, it's essential to keep one's CV updated, consistent, and ready for opportunities that may arise. The traditional approach of manual updates, inconsistent versions, and multiple platforms can be cumbersome, leading to potential discrepancies and missed opportunities. By harnessing the potential of modern technologies like React, Gatsby, and Hygraph CRM, and incorporating tools like react-pdf, professionals can simplify and automate the CV maintenance process, ensuring consistency and adaptability. The approach I’ve chosen offers a modern solution to the age-old challenge of CV maintenance and online presence enhancement. By intertwining automation with modern development techniques, it turns the contemporary challenges into a 1-minute CRM update. For the technically curious, the offers a deeper dive into the project, promising insights, and a path to replicating such automation feats. GitHub repository

Crafting Automation Magic: A Look Into Modern CV Development Techniques

Artificial intelligence has proliferated across industries and sectors. Be it education, astrology, food and beverages, engineering, or medicine — AI applications are found in all and sundry. Software engineers, programmers, and developers can utilize Artificial Intelligence tools to 10X their productivity with enhanced efficiency. Writing code is an exciting yet tedious job. With the help of AI tools, error-less secured codes can be written fast. AI is here to make the life of our computer programmers conveniently easier with the below-mentioned five free AI tools. What Are AI Tools in Programming or Coding? AI coding tools act as assistants and assist programmers and developers in writing error-free and accurate codes. They are also efficient at completing codes with initial prompts and providing suggestions in real time for betterment. Amazon CodeWhisperer The coding companion, the Amazon CodeWhisperer, helps developers build applications faster with extreme security. More than a billion lines of code have been used to train CodeWhisperer which enables it to garner a wide range of intelligent code suggestions in real-time. Be it snippets or completely functional functions based on your prompts or pre-existing codes — CodeWhisperer helps developers avoid tedious and time-consuming coding jobs, eventually expediting the building of APIs. ChatGPT The most popular AI content generation tool that helps in content moderation as well. ChatGPT intriguingly writes amazing codes. The only know-how you need while using ChatGPT to make optimum utilization is the art of writing prompts. CodeGeeX This is your AI-based assistant that completes your coding tasks before you can imagine. Powered with the pre-training of 20+ programming languages, trained on 850+ billion tokens on a cluster of 1,536 Ascend 910 AI Processors, with 13 billion parameters. CodeGeeX offers Multilingual code generation, cross-lingual translation, customizable programming, and open-source and cross-platform task completion. GitHub Copilot One of the most popular and widely adopted AI developer tools, GitHub Copilot has been trained on billions of lines of code. It transforms natural language into coding suggestions in more than 12 languages. prompts Bugasura An AI-powered bug-tracking tool that streamlines the tedious bug management process seamlessly. The ability to automatically detect bugs and provide insightful pattern detection assists Bugasura in monitoring and controlling bugs. Also, it provides detailed analysis for better bug tracking for the future. It mitigates bugs as per the severity and tags each bug found with a relevant description. It offers a single platform where you can connect, write about each bug, and discuss for future prevention. Is It Possible To Develop Codes Even if You Have Negligible Knowledge of Coding? Well, if this question had been asked two years back, the answer would have been no. However, with the emergence of flawless AI tools that have been trained with accurate and error-free data, today, anyone can write codes. The use of machine learning algorithms enables these AI tools to generate code as per the asked-for user functionalities and specifications. Does That Mean Efficient Does Will Lose Their Job? The answer is no! AI is not here to take the job of coders or developers but to make coding easier for them. The thought of AI replacing humans in totality is a far-sighted cry, and it is not going to happen. Remember, AI cannot use itself. Humans will use AI. Wrapping It Up Simply put, AI tools for programming are a boon for developers as they help them complete their coding tasks quickly and with precision. There are many AI tools for programming available, and five of the most popular ones have been mentioned here. The advent of AI tools has simplified coding, and anyone with minimum or negligible know-how can write code. Also published here

Stories, News & Mentions about CODEGEM

Codegem’s Stories on HackerNoon