superbio

If you check popular benchmarks used for measuring LLM performance, then you probably feel like AI is oh sooo clever. This is a fair surface-level impression; however, is AI truly better than an average human for cognitive tasks? Frontier models like o1 from OpenAI and Claude 3.5 Sonnet from Anthropic do perform better than expert humans in a number of fields, including law, coding, and math. Why, then, can’t ChatGPT solve some simple spatial reasoning tasks or some silly trick questions? Well, we are still talking about “large language models"—they take in a ton of letters and try to predict what letters to spit out for a given query. Notice, that nowhere in this equation the actual “thinking” is mentioned. The models are stochastic parrots of a sort, as they try to retrieve the right information from their training dataset instead of actually considering your question. At least, this was the case until OpenAI released o1-preview, but more on this later. Among those who started questioning the existing LLM benchmarks for relevance is the author of “AI Explained” a popular YouTube channel of which I’m a big fan. Phillip (the YouTuber’s name) noticed that the industry standard benchmarks have a clear style of questions that are also mostly publicly available. This means that not only can those exact questions be part of the training dataset, but due to standardization, it is easier for models to spot & apply patterns from the overall training data. Put simply, AI researchers who create groundbreaking complex tech can surely find a way to give their model the relevant questions and answers to “remember” before benchmarking. Looking at the results of the top model out there, o1 from OpenAI, one can imply that it scores above average in a lot of professional domains. And this is true, but this result relies on the availability of relevant training data and past examples from those specific fields. Don’t get me wrong, the models are now amazing at giving textbook answers to textbook questions, and this in itself is insanely impressive. The term “artificial intelligence," though, implies a bit more than just information retrieval; there should be some actual thinking involved. So a logical follow-up to all of the impressive numbers above is whether such “AI” can answer a tricky reasoning question. Does it have any spatial intelligence? Or can it navigate well in common social scenarios? The answer is - sometimes. Unlike field-specific questions with predefined answers, problems that humans solve on a daily basis often require understanding the context beyond natural language (which is the only thing LLMs have). Above are the top scorers on the SIMPLE benchmark, which gives LLMs questions that an average person would deem trivial but the models can’t necessarily answer yet. We are used to seeing AI doing so much better than an average human in exams or specialized benchmarks, but here, the top model performance is actually just 41.7% (o1-preview) vs. 83.7% of an average human. This benchmark uses 200 multiple-choice text questions focused on spatio-temporal reasoning, social intelligence, and trick questions. The most important feature of the benchmark is that those questions are not publicly available, so AI labs cannot just add them to their training data. You can learn more about this benchmark here. This new approach to measuring LLM performance shows just how far all the models still are from an average human reasoning capability. The quicker this gap closes in the coming months, the more definitive the answer “yes” to our headline becomes. An interesting new metric to look out for if you are enthusiastic but cautious about AI. If you check popular benchmarks used for measuring LLM performance, then you probably feel like AI is oh sooo clever. This is a fair surface-level impression; however, is AI truly better than an average human for cognitive tasks? Frontier models like o1 from OpenAI and Claude 3.5 Sonnet from Anthropic do perform better than expert humans in a number of fields, including law, coding, and math. Why, then, can’t ChatGPT solve some simple spatial reasoning tasks or some silly trick questions? Well, we are still talking about “large language models"—they take in a ton of letters and try to predict what letters to spit out for a given query. Notice, that nowhere in this equation the actual “thinking” is mentioned. The models are stochastic parrots of a sort, as they try to retrieve the right information from their training dataset instead of actually considering your question. At least, this was the case until OpenAI released o1-preview, but more on this later. stochastic parrots Among those who started questioning the existing LLM benchmarks for relevance is the author of “AI Explained” a popular YouTube channel of which I’m a big fan. Phillip (the YouTuber’s name) noticed that the industry standard benchmarks have a clear style of questions that are also mostly publicly available. This means that not only can those exact questions be part of the training dataset, but due to standardization, it is easier for models to spot & apply patterns from the overall training data. Put simply, AI researchers who create groundbreaking complex tech can surely find a way to give their model the relevant questions and answers to “remember” before benchmarking. Looking at the results of the top model out there, o1 from OpenAI, one can imply that it scores above average in a lot of professional domains. And this is true, but this result relies on the availability of relevant training data and past examples from those specific fields. Don’t get me wrong, the models are now amazing at giving textbook answers to textbook questions, and this in itself is insanely impressive. The term “artificial intelligence," though, implies a bit more than just information retrieval; there should be some actual thinking involved. So a logical follow-up to all of the impressive numbers above is whether such “AI” can answer a tricky reasoning question. Does it have any spatial intelligence? Or can it navigate well in common social scenarios? The answer is - sometimes. Unlike field-specific questions with predefined answers, problems that humans solve on a daily basis often require understanding the context beyond natural language (which is the only thing LLMs have). Above are the top scorers on the SIMPLE benchmark, which gives LLMs questions that an average person would deem trivial but the models can’t necessarily answer yet. We are used to seeing AI doing so much better than an average human in exams or specialized benchmarks, but here, the top model performance is actually just 41.7% (o1-preview) vs. 83.7% of an average human. This benchmark uses 200 multiple-choice text questions focused on spatio-temporal reasoning, social intelligence, and trick questions. The most important feature of the benchmark is that those questions are not publicly available, so AI labs cannot just add them to their training data. You can learn more about this benchmark here . here This new approach to measuring LLM performance shows just how far all the models still are from an average human reasoning capability. The quicker this gap closes in the coming months, the more definitive the answer “yes” to our headline becomes. An interesting new metric to look out for if you are enthusiastic but cautious about AI.

AI vs Human - Is the Machine Already Superior? 

الذكاء الاصطناعي مقابل الإنسان - هل الآلة متفوقة بالفعل؟

KI vs Mensch – Ist die Maschine schon überlegen?

IA vs. humanos: ¿la máquina ya es superior?

AI protiv čovjeka - je li stroj već superioran? 

Intelligenza artificiale contro uomo: la macchina è già superiore?

AI vs 人間 - 機械はすでに優れているのか?

AI prieš žmogų – ar mašina jau pranašesnė? 

Sztuczna inteligencja kontra człowiek – czy maszyna jest już lepsza?

AI بمقابله انسان - ایا ماشین دمخه غوره دی؟ 

AI vs Aadanaha - Mashiinku mar hore ma ka sarreeyaa? 

AI vs மனித - இயந்திரம் ஏற்கனவே உயர்ந்ததா? 

AI กับ มนุษย์ - เครื่องจักรเหนือกว่าแล้วจริงหรือ?

Authors:
(1) Dustin Podell, Stability AI, Applied Research;
(2) Zion English, Stability AI, Applied Research;
(3) Kyle Lacey, Stability AI, Applied Research;
(4) Andreas Blattmann, Stability AI, Applied Research;
(5) Tim Dockhorn, Stability AI, Applied Research;
(6) Jonas Müller, Stability AI, Applied Research;
(7) Joe Penna, Stability AI, Applied Research;
(8) Robin Rombach, Stability AI, Applied Research. Table of Links Abstract and 1 Introduction 2 Improving Stable Diffusion 2.1 Architecture & Scale 2.2 Micro-Conditioning 2.3 Multi-Aspect Training 2.4 Improved Autoencoder and 2.5 Putting Everything Together 3 Future Work Appendix A Acknowledgements B Limitations C Diffusion Models D Comparison to the State of the Art E Comparison to Midjourney v5.1 F On FID Assessment of Generative Text-Image Foundation Models G Additional Comparison between Single- and Two-Stage SDXL pipeline References 2.2 Micro-Conditioning Conditioning the Model on Image Size A notorious shortcoming of the LDM paradigm [38] is the fact that training a model requires a minimal image size, due to its two-stage architecture. The two main approaches to tackle this problem are either to discard all training images below a certain minimal resolution (for example, Stable Diffusion 1.4/1.5 discarded all images with any size below 512 pixels), or, alternatively, upscale images that are too small. However, depending on the desired image resolution, the former method can lead to significant portions of the training data being discarded, what will likely lead to a loss in performance and hurt generalization. We visualize such effects in Fig. 2 for the dataset on which SDXL was pretrained. For this particular choice of data, discarding all samples below our pretraining resolution of 256 2 pixels would lead to a significant 39% of discarded data. The second method, on the other hand, usually introduces upscaling artifacts which may leak into the final model outputs, causing, for example, blurry samples. Conditioning the Model on Cropping Parameters The first two rows of Fig. 4 illustrate a typical failure mode of previous SD models: Synthesized objects can be cropped, such as the cut-off head of the cat in the left examples for SD 1-5 and SD 2-1. An intuitive explanation for this behavior is the use of random cropping during training of the model: As collating a batch in DL frameworks such as PyTorch [32] requires tensors of the same size, a typical processing pipeline is to (i) resize an image such that the shortest size matches the desired target size, followed by (ii) randomly cropping the image along the longer axis. While random cropping is a natural form of data augmentation, it can leak into the generated samples, causing the malicious effects shown above. While other methods like data bucketing [31] successfully tackle the same task, we still benefit from cropping-induced data augmentation, while making sure that it does not leak into the generation process - we actually use it to our advantage to gain more control over the image synthesis process. Furthermore, it is easy to implement and can be applied in an online fashion during training, without additional data preprocessing. This paper is available on arxiv under CC BY 4.0 DEED license. Authors: (1) Dustin Podell, Stability AI, Applied Research; (2) Zion English, Stability AI, Applied Research; (3) Kyle Lacey, Stability AI, Applied Research; (4) Andreas Blattmann, Stability AI, Applied Research; (5) Tim Dockhorn, Stability AI, Applied Research; (6) Jonas Müller, Stability AI, Applied Research; (7) Joe Penna, Stability AI, Applied Research; (8) Robin Rombach, Stability AI, Applied Research. Authors: Authors: (1) Dustin Podell, Stability AI, Applied Research; (2) Zion English, Stability AI, Applied Research; (3) Kyle Lacey, Stability AI, Applied Research; (4) Andreas Blattmann, Stability AI, Applied Research; (5) Tim Dockhorn, Stability AI, Applied Research; (6) Jonas Müller, Stability AI, Applied Research; (7) Joe Penna, Stability AI, Applied Research; (8) Robin Rombach, Stability AI, Applied Research. Table of Links Abstract and 1 Introduction Abstract and 1 Introduction 2 Improving Stable Diffusion 2 Improving Stable Diffusion 2.1 Architecture & Scale 2.1 Architecture & Scale 2.2 Micro-Conditioning 2.2 Micro-Conditioning 2.3 Multi-Aspect Training 2.3 Multi-Aspect Training 2.4 Improved Autoencoder and 2.5 Putting Everything Together 2.4 Improved Autoencoder and 2.5 Putting Everything Together 3 Future Work 3 Future Work Appendix Appendix A Acknowledgements A Acknowledgements B Limitations B Limitations C Diffusion Models C Diffusion Models D Comparison to the State of the Art D Comparison to the State of the Art E Comparison to Midjourney v5.1 E Comparison to Midjourney v5.1 F On FID Assessment of Generative Text-Image Foundation Models F On FID Assessment of Generative Text-Image Foundation Models G Additional Comparison between Single- and Two-Stage SDXL pipeline G Additional Comparison between Single- and Two-Stage SDXL pipeline References References 2.2 Micro-Conditioning Conditioning the Model on Image Size A notorious shortcoming of the LDM paradigm [38] is the fact that training a model requires a minimal image size, due to its two-stage architecture. The two main approaches to tackle this problem are either to discard all training images below a certain minimal resolution (for example, Stable Diffusion 1.4/1.5 discarded all images with any size below 512 pixels), or, alternatively, upscale images that are too small. However, depending on the desired image resolution, the former method can lead to significant portions of the training data being discarded, what will likely lead to a loss in performance and hurt generalization. We visualize such effects in Fig. 2 for the dataset on which SDXL was pretrained. For this particular choice of data, discarding all samples below our pretraining resolution of 256 2 pixels would lead to a significant 39% of discarded data. The second method, on the other hand, usually introduces upscaling artifacts which may leak into the final model outputs, causing, for example, blurry samples. Conditioning the Model on Cropping Parameters The first two rows of Fig. 4 illustrate a typical failure mode of previous SD models: Synthesized objects can be cropped, such as the cut-off head of the cat in the left examples for SD 1-5 and SD 2-1. An intuitive explanation for this behavior is the use of random cropping during training of the model: As collating a batch in DL frameworks such as Conditioning the Model on Cropping Parameters PyTorch [32] requires tensors of the same size, a typical processing pipeline is to (i) resize an image such that the shortest size matches the desired target size, followed by (ii) randomly cropping the image along the longer axis. While random cropping is a natural form of data augmentation, it can leak into the generated samples, causing the malicious effects shown above. While other methods like data bucketing [31] successfully tackle the same task, we still benefit from cropping-induced data augmentation, while making sure that it does not leak into the generation process - we actually use it to our advantage to gain more control over the image synthesis process. Furthermore, it is easy to implement and can be applied in an online fashion during training, without additional data preprocessing. This paper is available on arxiv under CC BY 4.0 DEED license. This paper is available on arxiv under CC BY 4.0 DEED license. available on arxiv

Micro-Conditioning Strategies for Superior Image Generation in SDXL

A brief understanding of the difference between C# and JavaScript web scraping As a compiled language, C# provides a wealth of libraries and frameworks, such as HtmlAgilityPack, HttpClient, etc., which facilitate the implementation of complex web crawling logic, and the code is concise and efficient, with strong debugging and error handling capabilities. At the same time, C# has good cross-platform support and is suitable for a variety of operating systems. However, the learning curve of C# may be relatively steep and requires a certain programming foundation. In contrast, JavaScript, as a scripting language, is more flexible in web crawling and can be run directly in the browser without the need for additional installation environment. JavaScript has a rich DOM operation API, which is convenient for direct operation of web page elements. In addition, JavaScript is also supported by a large number of third-party libraries and frameworks, such as Puppeteer, Cheerio, etc., which further simplifies the implementation of web crawling. However, JavaScript’s asynchronous programming model may be relatively complex and requires a certain learning cost. Summary of C# vs JavaScript for web scraping Differences in language and environment‌ C#:‌ Requires .NET environment,‌ suitable for desktop or server-side applications. ‌
JavaScript:‌ Built-in in the browser,‌ suitable for front-end and Node.js environment. ‌ Crawl tools and libraries‌:‌ C#: Commonly used HttpClient,‌ combined with HtmlAgilityPack parsing. ‌
JavaScript:‌ Libraries such as Axios can be used,‌ with Cheerio parsing. ‌ Execution environment and restrictions‌ C#:‌ Executed on the server or desktop,‌ less restricted by browsers. ‌
JavaScript:‌ Executed in the browser,‌ restricted by the same-origin policy, etc. ‌ Processing dynamic content‌ Both require additional processing,‌ such as Selenium assistance. ‌ JavaScript has a natural advantage in the browser environment. ‌ Summary‌ Choose based on project requirements,‌ development environment and resources. ‌ Which one is better for crawling complex dynamic web pages, C# or JavaScript? For crawling complex dynamic web pages, C# and JavaScript each have their own advantages, but C# combined with tools such as Selenium is usually more suitable. ‌ JavaScript‌:‌ As a front-end scripting language, JavaScript is executed in a browser environment and naturally supports processing dynamic content. ‌However, when JavaScript is executed on the server side or in desktop applications, it requires the help of tools such as Node.js, and may be limited by the browser’s homology policy, etc. ‌ ‌


C#‌:‌ By combining libraries such as Selenium WebDriver, C# can simulate browser behavior and process JavaScript-rendered content, including login, click, scroll, and other operations. ‌This method can more comprehensively crawl dynamic web page data, and C#’s strong typing characteristics and rich library support also improve development efficiency and stability. ‌ Therefore,‌ in scenarios where complex dynamic web pages need to be crawled,‌ it is recommended to use C# combined with tools such as Selenium for development‌ What technologies and tools are needed for web scraping with C#? Web scraping with C# requires the following technologies and tools: ‌ ‌ HttpClient or WebClient class‌:‌ used to send HTTP requests and obtain web page content. ‌HttpClient provides more flexible functions and is suitable for handling complex HTTP requests. ‌ ‌
HTML parsing library‌:‌ such as HtmlAgilityPack,‌ used to parse the obtained HTML document and extract the required data from it. ‌HtmlAgilityPack supports XPath and CSS selectors, which is convenient for locating HTML elements. ‌ ‌
Regular expression‌:‌ used to match and extract specific text content in HTML documents, but attention should be paid to the accuracy and efficiency of regular expressions. ‌ ‌
Selenium WebDriver‌:‌ For scenarios that need to simulate browser behavior (such as logging in,‌ processing JavaScript rendered content),‌ Selenium WebDriver can be used to simulate user operations. ‌
JSON parsing library‌:‌ such as Json.NET,‌ used to parse JSON formatted data,‌ which is very useful when processing data returned by API. ‌
Exception handling and multithreading: In order to improve the stability and efficiency of the program, you need to write exception handling code and consider using multithreading technology to process multiple requests concurrently. ‌
Proxy and User-Agent settings: In order to bypass the anti-crawling mechanism of the website, you may need to set the proxy and custom User-Agent to simulate different access environments. ‌ The combination of these technologies and tools can efficiently implement the C# web crawling function. ‌ How to crawl dynamic web pages with C# combined with Selenium? ‌ How to use C# combined with Selenium to crawl dynamic web pages? ‌C# combined with Selenium to crawl dynamic web pages‌ Environment preparation‌:‌ Make sure that the C# development environment is installed. ‌
Install Selenium WebDriver, which is used to simulate browser behavior. ‌
Download and set up the browser driver, such as ChromeDriver, to ensure that it is consistent with the browser version. ‌ Usage steps‌:‌ Import Selenium-related external libraries, such as WebDriver, WebDriverWait, etc. ‌
Initialize WebDriver, set up the browser driver, and open the target web page. ‌
Use the methods provided by Selenium to simulate user behaviors, such as clicking, input, scrolling, etc., to handle operations such as dynamically loading content or logging in. ‌
Parse the web page source code and extract the required data. ‌
Close the browser and WebDriver instance. ‌ By combining C# with Selenium, you can effectively crawl dynamic web page content, handle complex interactions, and avoid being blocked by website detection. ‌‌ Conclusion In summary, C# and JavaScript each have their own advantages and disadvantages in web crawling. The choice of language depends on specific needs and development environment. A brief understanding of the difference between C# and JavaScript web scraping As a compiled language, C# provides a wealth of libraries and frameworks, such as HtmlAgilityPack, HttpClient, etc., which facilitate the implementation of complex web crawling logic, and the code is concise and efficient, with strong debugging and error handling capabilities. At the same time, C# has good cross-platform support and is suitable for a variety of operating systems. However, the learning curve of C# may be relatively steep and requires a certain programming foundation. In contrast, JavaScript, as a scripting language, is more flexible in web crawling and can be run directly in the browser without the need for additional installation environment. JavaScript has a rich DOM operation API, which is convenient for direct operation of web page elements. In addition, JavaScript is also supported by a large number of third-party libraries and frameworks, such as Puppeteer, Cheerio, etc., which further simplifies the implementation of web crawling. However, JavaScript’s asynchronous programming model may be relatively complex and requires a certain learning cost. Summary of C# vs JavaScript for web scraping Differences in language and environment‌ C#:‌ Requires .NET environment,‌ suitable for desktop or server-side applications. ‌ JavaScript:‌ Built-in in the browser,‌ suitable for front-end and Node.js environment. ‌ C#:‌ Requires .NET environment,‌ suitable for desktop or server-side applications. ‌ JavaScript:‌ Built-in in the browser,‌ suitable for front-end and Node.js environment. ‌ Crawl tools and libraries‌:‌ C#: Commonly used HttpClient,‌ combined with HtmlAgilityPack parsing. ‌ JavaScript:‌ Libraries such as Axios can be used,‌ with Cheerio parsing. ‌ C#: Commonly used HttpClient,‌ combined with HtmlAgilityPack parsing. ‌ JavaScript:‌ Libraries such as Axios can be used,‌ with Cheerio parsing. ‌ Execution environment and restrictions‌ C#:‌ Executed on the server or desktop,‌ less restricted by browsers. ‌ JavaScript:‌ Executed in the browser,‌ restricted by the same-origin policy, etc. ‌ C#:‌ Executed on the server or desktop,‌ less restricted by browsers. ‌ JavaScript:‌ Executed in the browser,‌ restricted by the same-origin policy, etc. ‌ Processing dynamic content‌ Both require additional processing,‌ such as Selenium assistance. ‌ JavaScript has a natural advantage in the browser environment. ‌ Summary‌ Choose based on project requirements,‌ development environment and resources. ‌ Which one is better for crawling complex dynamic web pages, C# or JavaScript? For crawling complex dynamic web pages, C# and JavaScript each have their own advantages, but C# combined with tools such as Selenium is usually more suitable. ‌ JavaScript‌:‌ As a front-end scripting language, JavaScript is executed in a browser environment and naturally supports processing dynamic content. ‌However, when JavaScript is executed on the server side or in desktop applications, it requires the help of tools such as Node.js, and may be limited by the browser’s homology policy, etc. ‌ ‌ C#‌:‌ By combining libraries such as Selenium WebDriver, C# can simulate browser behavior and process JavaScript-rendered content, including login, click, scroll, and other operations. ‌This method can more comprehensively crawl dynamic web page data, and C#’s strong typing characteristics and rich library support also improve development efficiency and stability. ‌ JavaScript‌:‌ As a front-end scripting language, JavaScript is executed in a browser environment and naturally supports processing dynamic content. ‌However, when JavaScript is executed on the server side or in desktop applications, it requires the help of tools such as Node.js, and may be limited by the browser’s homology policy, etc. ‌ ‌ JavaScript‌:‌ As a front-end scripting language, JavaScript is executed in a browser environment and naturally supports processing dynamic content. ‌However, when JavaScript is executed on the server side or in desktop applications, it requires the help of tools such as Node.js, and may be limited by the browser’s homology policy, etc. ‌ ‌ C#‌:‌ By combining libraries such as Selenium WebDriver, C# can simulate browser behavior and process JavaScript-rendered content, including login, click, scroll, and other operations. ‌This method can more comprehensively crawl dynamic web page data, and C#’s strong typing characteristics and rich library support also improve development efficiency and stability. ‌ C#‌:‌ By combining libraries such as Selenium WebDriver, C# can simulate browser behavior and process JavaScript-rendered content, including login, click, scroll, and other operations. ‌This method can more comprehensively crawl dynamic web page data, and C#’s strong typing characteristics and rich library support also improve development efficiency and stability. ‌ Therefore,‌ in scenarios where complex dynamic web pages need to be crawled,‌ it is recommended to use C# combined with tools such as Selenium for development‌ What technologies and tools are needed for web scraping with C#? Web scraping with C# requires the following technologies and tools: ‌ ‌ HttpClient or WebClient class‌:‌ used to send HTTP requests and obtain web page content. ‌HttpClient provides more flexible functions and is suitable for handling complex HTTP requests. ‌ ‌ HTML parsing library‌:‌ such as HtmlAgilityPack,‌ used to parse the obtained HTML document and extract the required data from it. ‌HtmlAgilityPack supports XPath and CSS selectors, which is convenient for locating HTML elements. ‌ ‌ Regular expression‌:‌ used to match and extract specific text content in HTML documents, but attention should be paid to the accuracy and efficiency of regular expressions. ‌ ‌ Selenium WebDriver‌:‌ For scenarios that need to simulate browser behavior (such as logging in,‌ processing JavaScript rendered content),‌ Selenium WebDriver can be used to simulate user operations. ‌ JSON parsing library‌:‌ such as Json.NET,‌ used to parse JSON formatted data,‌ which is very useful when processing data returned by API. ‌ Exception handling and multithreading: In order to improve the stability and efficiency of the program, you need to write exception handling code and consider using multithreading technology to process multiple requests concurrently. ‌ Proxy and User-Agent settings: In order to bypass the anti-crawling mechanism of the website, you may need to set the proxy and custom User-Agent to simulate different access environments. ‌ HttpClient or WebClient class‌:‌ used to send HTTP requests and obtain web page content. ‌HttpClient provides more flexible functions and is suitable for handling complex HTTP requests. ‌ ‌ HTML parsing library‌:‌ such as HtmlAgilityPack,‌ used to parse the obtained HTML document and extract the required data from it. ‌HtmlAgilityPack supports XPath and CSS selectors, which is convenient for locating HTML elements. ‌ ‌ Regular expression‌:‌ used to match and extract specific text content in HTML documents, but attention should be paid to the accuracy and efficiency of regular expressions. ‌ ‌ Selenium WebDriver‌:‌ For scenarios that need to simulate browser behavior (such as logging in,‌ processing JavaScript rendered content),‌ Selenium WebDriver can be used to simulate user operations. ‌ JSON parsing library‌:‌ such as Json.NET ,‌ used to parse JSON formatted data,‌ which is very useful when processing data returned by API. ‌ Json.NET Exception handling and multithreading: In order to improve the stability and efficiency of the program, you need to write exception handling code and consider using multithreading technology to process multiple requests concurrently. ‌ Proxy and User-Agent settings: In order to bypass the anti-crawling mechanism of the website, you may need to set the proxy and custom User-Agent to simulate different access environments. ‌ The combination of these technologies and tools can efficiently implement the C# web crawling function. ‌ How to crawl dynamic web pages with C# combined with Selenium? ‌ How to use C# combined with Selenium to crawl dynamic web pages? ‌C# combined with Selenium to crawl dynamic web pages‌ Environment preparation‌:‌ Make sure that the C# development environment is installed. ‌ Install Selenium WebDriver, which is used to simulate browser behavior. ‌ Download and set up the browser driver, such as ChromeDriver, to ensure that it is consistent with the browser version. ‌ Make sure that the C# development environment is installed. ‌ Install Selenium WebDriver, which is used to simulate browser behavior. ‌ Download and set up the browser driver, such as ChromeDriver, to ensure that it is consistent with the browser version. ‌ Usage steps‌:‌ Import Selenium-related external libraries, such as WebDriver, WebDriverWait, etc. ‌ Initialize WebDriver, set up the browser driver, and open the target web page. ‌ Use the methods provided by Selenium to simulate user behaviors, such as clicking, input, scrolling, etc., to handle operations such as dynamically loading content or logging in. ‌ Parse the web page source code and extract the required data. ‌ Close the browser and WebDriver instance. ‌ Import Selenium-related external libraries, such as WebDriver, WebDriverWait, etc. ‌ Initialize WebDriver, set up the browser driver, and open the target web page. ‌ Use the methods provided by Selenium to simulate user behaviors, such as clicking, input, scrolling, etc., to handle operations such as dynamically loading content or logging in. ‌ Parse the web page source code and extract the required data. ‌ Close the browser and WebDriver instance. ‌ By combining C# with Selenium, you can effectively crawl dynamic web page content, handle complex interactions, and avoid being blocked by website detection. ‌‌ Conclusion In summary, C# and JavaScript each have their own advantages and disadvantages in web crawling. The choice of language depends on specific needs and development environment.

Web Scraping: Is C# or JavaScript the Superior Choice?

Los Angeles, CA – July 21, 2024 – The Los Angeles Superior Court experienced an unprecedented ransomware attack on Friday, July 19, 2024, leading to the shutdown of all 36 courthouse locations across the county. The breach, which occurred in the early hours of the morning, necessitated the immediate disabling of nearly all network systems to contain the damage and protect sensitive information. Incident Details The ransomware attack on the Los Angeles Superior Court began early on Friday, July 19, targeting both internal and external systems, including the MyJuryDuty portal and case management systems. The court's cybersecurity division promptly detected the attack and shut down the network to prevent further damage. This swift action, while necessary, significantly disrupted services, making critical systems inaccessible and affecting juror communications. In response, the court collaborated with the California Governor's Office of Emergency Services and various law enforcement agencies to investigate the breach and assess the damage. Preliminary findings indicate no compromised user data, though the investigation is ongoing. Despite the immediate response, many systems remained offline, leading court officials to close all courthouses on Monday, July 22, to allow IT experts additional time to safely restore operations. Presiding Judge Samantha P. Jessner emphasized the court's heavy investment in cybersecurity, which facilitated the rapid containment of the attack. The court expects to resume normal operations by Tuesday, July 23, after ensuring all systems are secure. This incident underscores the persistent threat of cyberattacks and the need for robust cybersecurity measures. The court clarified that this attack is unrelated to the recent global disruption caused by a faulty CrowdStrike update. Ongoing Response and Recovery Throughout the weekend, efforts to restore and recover the affected systems were underway. As of Sunday evening, many critical systems remained offline. However, court officials are optimistic about reopening by Tuesday, July 23, 2024. The additional time is intended to allow the court’s team of experts to focus exclusively on bringing systems back online as quickly and safely as possible. The Los Angeles Superior Court is working closely with the California Governor's Office of Emergency Services, as well as local, state, and federal law enforcement agencies, to investigate the incident and assess its impact. Preliminary investigations indicate no evidence that court user data has been compromised. Impact on Court Operations The shutdown affected all 36 court facilities, halting court operations and services temporarily. The court’s website, including the MyJuryDuty portal, displayed error messages indicating the extent of the disruption. Court officials emphasized their commitment to restoring services and resuming operations as expeditiously as possible. Not Related to CrowdStrike Issue Officials clarified that the ransomware attack on the Los Angeles Superior Court is unrelated to the recent global disruption caused by a faulty CrowdStrike update, which affected Windows systems worldwide around the same time. Cybersecurity Measures The court credited its swift detection and immediate response to the attack to its significant investment in cybersecurity and the expansion of its cybersecurity team. This incident highlights the critical importance of robust cybersecurity measures and ongoing vigilance against evolving cyber threats. Tools like NAKIVO Backup and Replication play a vital role in ensuring data protection and rapid recovery, further emphasizing the need for comprehensive cybersecurity strategies in today’s digital landscape. About Los Angeles Superior Court The Los Angeles Superior Court is the largest trial court in the United States, with over 4,800 employees operating in 41 court facilities across 26 cities in Los Angeles County. The court serves a population of over 10 million people. Los Angeles, CA – July 21, 2024 – The Los Angeles Superior Court experienced an unprecedented ransomware attack on Friday, July 19, 2024, leading to the shutdown of all 36 courthouse locations across the county. The breach, which occurred in the early hours of the morning, necessitated the immediate disabling of nearly all network systems to contain the damage and protect sensitive information. Incident Details The ransomware attack on the Los Angeles Superior Court began early on Friday, July 19, targeting both internal and external systems, including the MyJuryDuty portal and case management systems. The court's cybersecurity division promptly detected the attack and shut down the network to prevent further damage. This swift action, while necessary, significantly disrupted services, making critical systems inaccessible and affecting juror communications. Los Angeles Superior Court Los Angeles Superior Court In response, the court collaborated with the California Governor's Office of Emergency Services and various law enforcement agencies to investigate the breach and assess the damage. Preliminary findings indicate no compromised user data, though the investigation is ongoing. Despite the immediate response, many systems remained offline, leading court officials to close all courthouses on Monday, July 22, to allow IT experts additional time to safely restore operations. Presiding Judge Samantha P. Jessner emphasized the court's heavy investment in cybersecurity, which facilitated the rapid containment of the attack. The court expects to resume normal operations by Tuesday, July 23, after ensuring all systems are secure. This incident underscores the persistent threat of cyberattacks and the need for robust cybersecurity measures. The court clarified that this attack is unrelated to the recent global disruption caused by a faulty CrowdStrike update . CrowdStrike update CrowdStrike update Ongoing Response and Recovery Throughout the weekend, efforts to restore and recover the affected systems were underway. As of Sunday evening, many critical systems remained offline. However, court officials are optimistic about reopening by Tuesday, July 23, 2024. The additional time is intended to allow the court’s team of experts to focus exclusively on bringing systems back online as quickly and safely as possible. The Los Angeles Superior Court is working closely with the California Governor's Office of Emergency Services, as well as local, state, and federal law enforcement agencies, to investigate the incident and assess its impact. Preliminary investigations indicate no evidence that court user data has been compromised. Impact on Court Operations The shutdown affected all 36 court facilities, halting court operations and services temporarily. The court’s website, including the MyJuryDuty portal, displayed error messages indicating the extent of the disruption. Court officials emphasized their commitment to restoring services and resuming operations as expeditiously as possible. Not Related to CrowdStrike Issue Officials clarified that the ransomware attack on the Los Angeles Superior Court is unrelated to the recent global disruption caused by a faulty CrowdStrike update, which affected Windows systems worldwide around the same time. ransomware attack on the Los Angeles Superior Court ransomware attack on the Los Angeles Superior Court Cybersecurity Measures The court credited its swift detection and immediate response to the attack to its significant investment in cybersecurity and the expansion of its cybersecurity team. This incident highlights the critical importance of robust cybersecurity measures and ongoing vigilance against evolving cyber threats. Tools like NAKIVO Backup and Replication play a vital role in ensuring data protection and rapid recovery, further emphasizing the need for comprehensive cybersecurity strategies in today’s digital landscape. NAKIVO Backup and Replication NAKIVO Backup and Replication About Los Angeles Superior Court The Los Angeles Superior Court is the largest trial court in the United States, with over 4,800 employees operating in 41 court facilities across 26 cities in Los Angeles County. The court serves a population of over 10 million people.

Cybersecurity Breach Shuts Down LA Superior Court Network Systems

How To Resolve Conflicts With Your Superiors At Work

**LUXEMBOURG, Luxembourg, November 21st, 2024/Chainwire/--**DegenLayer, a newly launched memecoin focused blockchain & trading terminal app suite, has announced its testnet release, marking a key step toward its upcoming mainnet launch. The platform aims to facilitate zero setup memecoin trading and creation, leveraging the $20 billion liquidity within the Optimism Superchain ecosystem. The project's developers project daily revenues of $1 million in ETH, assuming a daily DEX trading volume of $200 million. With low transaction fees and a streamlined user interface, DegenLayer seeks to provide a gateway for mainstream users to engage in blockchain-based trading and creation. The project's native token $DELAY was fair launched on Uniswap last week, and is set to be listed on one of the top 15 CoinGecko-ranked exchanges next week, providing access to the token to their 10 million+ user community. The project is powered by a 60+ person team behind notable successes including PunksClub.io the CryptoPunk social network, Music.com (developed with Pharrell Williams), and AAA games like The Witcher 3 and Dying Light 2. The founding team previously achieved remarkable success with SuperBid, driving token value from $0.01 to $12 in 2021. "Our proven track record in both Web3 and gaming demonstrates our ability to deliver compelling user experiences," says Jacob Rylko, Co-Founder & CEO. "With DegenLayer, we're leveraging our existing reach of 3 million+ users through our Telegram mini-app, Firecoin, to accelerate adoption." Key Features And Projections: Innovative "Pump Technology" with 50% of revenue allocated to viral user rewards, $DELAY and memecoin buybacks
Viral referral program projecting $100,000 daily reward distributions
Seamless integration with Telegram's 1B+ user base via a mini app
One-click memecoin creation and trading interface for non-crypto users
Transaction fees below $0.01 with 2,000 TPS capacity DegenLayer's launch represents a significant milestone in making memecoin trading accessible to mainstream users while leveraging established Optimism infrastructure, the same that is used by Coinbase's BASE Layer 2. About DegenLayer DegenLayer is a trading terminal and Ethereum Layer 2 blockchain designed to make memecoin trading accessible to mainstream users. Built on Optimism technology, the platform offers low-cost, high-speed transactions and a zero setup interface for memecoin trading and creation. Backed by a team with expertise in Web3, gaming, and entertainment, DegenLayer aims to bring innovative blockchain solutions to the global market. For more information about DegenLayer and its revolutionary approach to memecoin trading, users can visit https://degenlayer.wtf. Media inquiries can be directed to contact@degenlayer.wtf. Contact CEO Jacob Rylko DegenLayer contact@degenlayer.wtf This story was distributed as a release by Chainwire under HackerNoon’s Business Blogging Program. Learn more about the program here **LUXEMBOURG, Luxembourg, November 21st, 2024/Chainwire/--**DegenLayer, a newly launched memecoin focused blockchain & trading terminal app suite, has announced its testnet release, marking a key step toward its upcoming mainnet launch. The platform aims to facilitate zero setup memecoin trading and creation, leveraging the $20 billion liquidity within the Optimism Superchain ecosystem. The project's developers project daily revenues of $1 million in ETH, assuming a daily DEX trading volume of $200 million. With low transaction fees and a streamlined user interface, DegenLayer seeks to provide a gateway for mainstream users to engage in blockchain-based trading and creation. The project's native token $DELAY was fair launched on Uniswap last week, and is set to be listed on one of the top 15 CoinGecko-ranked exchanges next week, providing access to the token to their 10 million+ user community. The project is powered by a 60+ person team behind notable successes including PunksClub.io the CryptoPunk social network, Music.com (developed with Pharrell Williams), and AAA games like The Witcher 3 and Dying Light 2. The founding team previously achieved remarkable success with SuperBid, driving token value from $0.01 to $12 in 2021. PunksClub.io PunksClub.io Music.com Music.com "Our proven track record in both Web3 and gaming demonstrates our ability to deliver compelling user experiences," says Jacob Rylko, Co-Founder & CEO. "With DegenLayer, we're leveraging our existing reach of 3 million+ users through our Telegram mini-app, Firecoin, to accelerate adoption." "Our proven track record in both Web3 and gaming demonstrates our ability to deliver compelling user experiences," says Jacob Rylko, Co-Founder & CEO. "With DegenLayer, we're leveraging our existing reach of 3 million+ users through our Telegram mini-app, Firecoin, to accelerate adoption." Key Features And Projections: Innovative "Pump Technology" with 50% of revenue allocated to viral user rewards, $DELAY and memecoin buybacks Viral referral program projecting $100,000 daily reward distributions Seamless integration with Telegram's 1B+ user base via a mini app One-click memecoin creation and trading interface for non-crypto users Transaction fees below $0.01 with 2,000 TPS capacity Innovative "Pump Technology" with 50% of revenue allocated to viral user rewards, $DELAY and memecoin buybacks Viral referral program projecting $100,000 daily reward distributions Seamless integration with Telegram's 1B+ user base via a mini app One-click memecoin creation and trading interface for non-crypto users Transaction fees below $0.01 with 2,000 TPS capacity DegenLayer's launch represents a significant milestone in making memecoin trading accessible to mainstream users while leveraging established Optimism infrastructure, the same that is used by Coinbase's BASE Layer 2. About DegenLayer DegenLayer is a trading terminal and Ethereum Layer 2 blockchain designed to make memecoin trading accessible to mainstream users. Built on Optimism technology, the platform offers low-cost, high-speed transactions and a zero setup interface for memecoin trading and creation. Backed by a team with expertise in Web3, gaming, and entertainment, DegenLayer aims to bring innovative blockchain solutions to the global market. DegenLayer DegenLayer For more information about DegenLayer and its revolutionary approach to memecoin trading, users can visit https://degenlayer.wtf . Media inquiries can be directed to contact@degenlayer.wtf . https://degenlayer.wtf https://degenlayer.wtf contact@degenlayer.wtf contact@degenlayer.wtf Contact CEO Jacob Rylko DegenLayer contact@degenlayer.wtf This story was distributed as a release by Chainwire under HackerNoon’s Business Blogging Program. Learn more about the program here This story was distributed as a release by Chainwire under HackerNoon’s Business Blogging Program. Learn more about the program here here here

DegenLayer Introduces The First Memecoin Focused Blockchain

DegenLayer የመጀመሪያውን Memecoin ተኮር Blockchainን አስተዋውቋል

DegenLayer İlk Memecoin Focused Blockchain təqdim edir

DegenLayer presenta la primera blockchain centrada en Memecoin

DegenLayer اولین بلاک چین متمرکز بر Memecoin را معرفی می کند

DegenLayer מציגה את הבלוקצ'יין הראשון ממוקד Memecoin

डेजेनलेयर ने पहला मेमेकॉइन केंद्रित ब्लॉकचेन पेश किया

DegenLayerが初のミームコインに特化したブロックチェーンを発表

DegenLayer წარმოგიდგენთ პირველ Memecoin-ზე ორიენტირებულ ბლოკჩეინს

DegenLayer ແນະນໍາ Memecoin ທໍາອິດທີ່ສຸມໃສ່ Blockchain

DegenLayer introduceert de eerste op memecoins gerichte blockchain

DegenLayer аввалин Blockchain Memecoin-ро муаррифӣ мекунад

DegenLayer birinchi Memecoin yo'naltirilgan blokcheynni taqdim etadi

Amazon’s new Alexa model featuring Anthropic’s Claude AI could redefine voice assistants, but it comes with a price tag and data privacy concerns for consumers. Amazon is working towards a significant revamp for its flagship voice assistant, Alexa. After facing developmental challenges with its in-house generative AI (genA.I.) models, the tech giant has now turned to California-based artificial intelligence (A.I.) startup Anthropic’s generative model Claude AI to upgrade Alexa into a more capable and intuitive assistant. Amazonannounced in September that Claude AI will now serve as the core technology for Alexa’s upcoming upgrade. It is expected to enhance the voice assistant’s capabilities of handling more complex tasks and engaging in richer, multi-step conversations. From planning vacations to curating personalized news briefings, Amazon aims to significantly elevate speed, accuracy, and user experience for Alexa’s consumer base.“Anthropic’s visionary work with generative AI, most recently the introduction of its state-of-the-art Claude 3 family of models, combined with Amazon’s best-in-class infrastructure like AWS Trainium and managed services like Amazon Bedrockfurther unlocks exciting opportunities for customers to quickly, securely, and responsibly innovate with generative AI,” Dr. Swami Sivasubramanian, vice president of Data and AI at AWS said in ablog post. As with any advancement in A.I., the integration of Claude AI into Alexa also raises critical privacy concerns. Anthropic was recently found to be quietly injecting hidden instructions to user prompts submitted through its API to prevent copyright issues. Users discovered that prompts like "Respond as helpfully as possible" and "Do not comply with complex instructions" are hardcoded into the system. Although these hidden instructions aren't visible to users, they can affect how the AI responds, highlighting that Anthropic’s large language models (LLMs) may work differently in practice than expected. “In our recent evaluations of the latest LLMs, including those from Anthropic, we found that Claude demonstrates substantial implicit biases,” Sahil Agarwal, co-founder & CEO at Enkrypt AI, told me. “We have proactively reached out to the Anthropic team to address and rectify these issues. For Amazon, ensuring safety and security will be the critical challenge in dominating the AI market.” Likewise, Amazon’s $4 billion investment in Anthropic has caught the attention of the UK’s Competition and Markets Authority (CMA). The regulator is looking into whether this deal might significantly reduce competition in the UK market. The tech giant’s decision to integrate another genAI model introduces a complex interplay of innovation, user experience, and privacy concerns that could redefine how consumers interact with smart home technology. Claude's Potential Impact on Alexa: What to Expect Amazon’s journey to revamp Alexa hasn’t been without its hurdles. A recent Reuters report asserts that Amazon's initial tests with its new A.I. revealed major issues, such as slow responses and trouble understanding user prompts. With rivals like ChatGPT and Google Assistant setting high A.I. speed and accuracy standards, Amazon decided to switch to Anthropic’s Claude AI. The genA.I. model was found to be superior due to its better contextual understanding and more human-like conversation skills. While the basic "Classic" version of Alexa will still be free, users will need to pay for the extra features offered in the upgraded version. Amazon will charge a subscription fee for its "Remarkable Alexa" services, which will cost between $5 and $10 per month. Introducing a paid model for Alexa seems like a calculated risk for Amazon. The timing of this move is also strategic, transitioning Alexa towards a revenue-generating A.I. service like other tech conglomerates. Yet, this progress comes with potential risks. The enhanced AI capabilities could lead to increased data collection, raising concerns about how much personal information users are willing to share. “The new AI capabilities depend heavily on large datasets to refine and improve their responses,” Nate MacLeitch, CEO and founder of API platform QuickBlox, told me. “Amazon’s premium services will need access to the user's data to function effectively and understand user habits, health data, or their preferences. This means a user’s interactions with Alexa are not just private conversations but are also part of the data that trains the AI. ” Apple recently implemented Private Cloud Compute for its Apple Intelligence, ensuring user data remains secure, even from Apple itself. Although Amazon possesses similar capabilities, the specifics of how the integration between Alexa and Anthropic will uphold user privacy remain unclear. Will users be comfortable knowing their data contributes to the ongoing improvement of Alexa AI, especially when they are paying a subscription fee for the service? The company has been tight-lipped about specific measures to ensure data privacy, leaving users and privacy advocates to speculate and watch closely. “The key is to only train aggregate AI models using consented data that has been anonymized. Amazon can do the same by integrating a consent API and CDP system in the backend, and use Alexa to ask for consent before recording any data from the user and utilizing it,” Xun Wang, Chief Technology Officer of Bloomreach and former VP of NVIDIA, told me. “A.I. will soon be able to generate an audio text that zeroes in on the information you care about the most as opposed to a generic dump of all the information related to any specific piece of content.” Amazon’s new Alexa model featuring Anthropic’s Claude AI could redefine voice assistants, but it comes with a price tag and data privacy concerns for consumers. Amazon’s new Alexa model featuring Anthropic’s Claude AI could redefine voice assistants, but it comes with a price tag and data privacy concerns for consumers. Amazon’s new Alexa model featuring Anthropic’s Claude AI could redefine voice assistants, but it comes with a price tag and data privacy concerns for consumers. Amazon is working towards a significant revamp for its flagship voice assistant, Alexa. After facing developmental challenges with its in-house generative AI (genA.I.) models, the tech giant has now turned to California-based artificial intelligence (A.I.) startup Anthropic’s generative model Claude AI to upgrade Alexa into a more capable and intuitive assistant. Amazon announced in September that Claude AI will now serve as the core technology for Alexa’s upcoming upgrade. It is expected to enhance the voice assistant’s capabilities of handling more complex tasks and engaging in richer, multi-step conversations. From planning vacations to curating personalized news briefings, Amazon aims to significantly elevate speed, accuracy, and user experience for Alexa’s consumer base. “Anthropic’s visionary work with generative AI, most recently the introduction of its state-of-the-art Claude 3 family of models, combined with Amazon’s best-in-class infrastructure like AWS Trainium and managed services like Amazon Bedrockfurther unlocks exciting opportunities for customers to quickly, securely, and responsibly innovate with generative AI,” Dr. Swami Sivasubramanian, vice president of Data and AI at AWS said in a blog post . Amazon blog post blog post As with any advancement in A.I., the integration of Claude AI into Alexa also raises critical privacy concerns. Anthropic was recently found to be quietly injecting hidden instructions to user prompts submitted through its API to prevent copyright issues. Users discovered that prompts like "Respond as helpfully as possible" and "Do not comply with complex instructions" are hardcoded into the system. Although these hidden instructions aren't visible to users, they can affect how the AI responds, highlighting that Anthropic’s large language models (LLMs) may work differently in practice than expected. injecting hidden instructions injecting hidden instructions “In our recent evaluations of the latest LLMs, including those from Anthropic, we found that Claude demonstrates substantial implicit biases,” Sahil Agarwal, co-founder & CEO at Enkrypt AI, told me. “We have proactively reached out to the Anthropic team to address and rectify these issues. For Amazon, ensuring safety and security will be the critical challenge in dominating the AI market.” Likewise, Amazon’s $4 billion investment in Anthropic has caught the attention of the UK’s Competition and Markets Authority ( CMA ). The regulator is looking into whether this deal might significantly reduce competition in the UK market. The tech giant’s decision to integrate another genAI model introduces a complex interplay of innovation, user experience, and privacy concerns that could redefine how consumers interact with smart home technology. CMA CMA Claude's Potential Impact on Alexa: What to Expect Amazon’s journey to revamp Alexa hasn’t been without its hurdles. A recent Reuters report asserts that Amazon's initial tests with its new A.I. revealed major issues, such as slow responses and trouble understanding user prompts. With rivals like ChatGPT and Google Assistant setting high A.I. speed and accuracy standards, Amazon decided to switch to Anthropic’s Claude AI. The genA.I. model was found to be superior due to its better contextual understanding and more human-like conversation skills. Reuters report Reuters report While the basic "Classic" version of Alexa will still be free, users will need to pay for the extra features offered in the upgraded version. Amazon will charge a subscription fee for its "Remarkable Alexa" services, which will cost between $5 and $10 per month. Introducing a paid model for Alexa seems like a calculated risk for Amazon. The timing of this move is also strategic, transitioning Alexa towards a revenue-generating A.I. service like other tech conglomerates. Yet, this progress comes with potential risks. The enhanced AI capabilities could lead to increased data collection, raising concerns about how much personal information users are willing to share. “The new AI capabilities depend heavily on large datasets to refine and improve their responses,” Nate MacLeitch, CEO and founder of API platform QuickBlox, told me. “Amazon’s premium services will need access to the user's data to function effectively and understand user habits, health data, or their preferences. This means a user’s interactions with Alexa are not just private conversations but are also part of the data that trains the AI. ” Apple recently implemented Private Cloud Compute for its Apple Intelligence, ensuring user data remains secure, even from Apple itself. Although Amazon possesses similar capabilities, the specifics of how the integration between Alexa and Anthropic will uphold user privacy remain unclear. Will users be comfortable knowing their data contributes to the ongoing improvement of Alexa AI, especially when they are paying a subscription fee for the service? The company has been tight-lipped about specific measures to ensure data privacy, leaving users and privacy advocates to speculate and watch closely. “The key is to only train aggregate AI models using consented data that has been anonymized. Amazon can do the same by integrating a consent API and CDP system in the backend, and use Alexa to ask for consent before recording any data from the user and utilizing it,” Xun Wang, Chief Technology Officer of Bloomreach and former VP of NVIDIA, told me. “A.I. will soon be able to generate an audio text that zeroes in on the information you care about the most as opposed to a generic dump of all the information related to any specific piece of content.”

Can Claude AI Solve Alexa’s Data Privacy Issues?

ক্লড এআই কি আলেক্সার ডেটা গোপনীয়তার সমস্যাগুলি সমাধান করতে পারে?

Kann Claude AI die Datenschutzprobleme von Alexa lösen?

¿Puede Claude AI resolver los problemas de privacidad de datos de Alexa?

Claude AI peut-il résoudre les problèmes de confidentialité des données d’Alexa ?

क्या क्लाउड एआई एलेक्सा की डेटा गोपनीयता समस्याओं को हल कर सकता है?

Claude AI は Alexa のデータ プライバシー問題を解決できるか?

Claude AI가 Alexa의 데이터 프라이버시 문제를 해결할 수 있을까?

A Claude AI pode resolver os problemas de privacidade de dados da Alexa?

Сможет ли Claude AI решить проблемы конфиденциальности данных Alexa?

Claude AI, Alexa'nın Veri Gizliliği Sorunlarını Çözebilir mi?

Liệu Claude AI có thể giải quyết được vấn đề về quyền riêng tư dữ liệu của Alexa không?

Claude AI 能解决 Alexa 的数据隐私问题吗？

Authors:
(1) Yuwei Guo, The Chinese University of Hong Kong;
(2) Ceyuan Yang, Shanghai Artificial Intelligence Laboratory with Corresponding Author;
(3) Anyi Rao, Stanford University;
(4) Zhengyang Liang, Shanghai Artificial Intelligence Laboratory;
(5) Yaohui Wang, Shanghai Artificial Intelligence Laboratory;
(6) Yu Qiao, Shanghai Artificial Intelligence Laboratory;
(7) Maneesh Agrawala, Stanford University;
(8) Dahua Lin, Shanghai Artificial Intelligence Laboratory;
(9) Bo Dai, The Chinese University of Hong Kong and The Chinese University of Hong Kong. Table of Links Abstract and 1 Introduction 2 Work Related 3 Preliminary AnimateDiff 4.1 Alleviate Negative Effects from Training Data with Domain Adapter 4.2 Learn Motion Priors with Motion Module 4.3 Adapt to New Motion Patterns with MotionLora 4.4 AnimateDiff in Practice 5 Experiments and 5.1 Qualitative Results 5.2 Qualitative Comparison 5.3 Ablative Study 5.4 Controllable Generation 6 Conclusion 7 Ethics Statement 8 Reproducibility Statement, Acknowledgement and References 5.2 QUANTITATIVE COMPARISON We conduct the quantitative comparison through user study and CLIP metrics. The comparison focuses on three key aspects: text alignment, domain similarity, and motion smoothness. The results are shown in Table 1. Detailed implementations can be found in supplementary materials. User study. In the user study, we generate animations using all three methods based on the same personalized T2I models. Participants are then asked to individually rank the results based on the above three aspects. We use the Average User Ranking (AUR) as a preference metric where a higher score indicates superior performance. Note that the corresponding prompts and images are provided for reference for text alignment and domain similarity evaluation. CLIP metric. We also employed the CLIP (Radford et al., 2021) metric, following the approach taken by previous studies (Wu et al., 2023; Khachatryan et al., 2023). When evaluating domain similarity, it is important to note that the CLIP score was computed between the animation frames and the reference images generated using the personalized T2Is. This paper is available on arxiv under CC BY 4.0 DEED license. Authors: (1) Yuwei Guo, The Chinese University of Hong Kong; (2) Ceyuan Yang, Shanghai Artificial Intelligence Laboratory with Corresponding Author; (3) Anyi Rao, Stanford University; (4) Zhengyang Liang, Shanghai Artificial Intelligence Laboratory; (5) Yaohui Wang, Shanghai Artificial Intelligence Laboratory; (6) Yu Qiao, Shanghai Artificial Intelligence Laboratory; (7) Maneesh Agrawala, Stanford University; (8) Dahua Lin, Shanghai Artificial Intelligence Laboratory; (9) Bo Dai, The Chinese University of Hong Kong and The Chinese University of Hong Kong. Authors: Authors: (1) Yuwei Guo, The Chinese University of Hong Kong; (2) Ceyuan Yang, Shanghai Artificial Intelligence Laboratory with Corresponding Author; (3) Anyi Rao, Stanford University; (4) Zhengyang Liang, Shanghai Artificial Intelligence Laboratory; (5) Yaohui Wang, Shanghai Artificial Intelligence Laboratory; (6) Yu Qiao, Shanghai Artificial Intelligence Laboratory; (7) Maneesh Agrawala, Stanford University; (8) Dahua Lin, Shanghai Artificial Intelligence Laboratory; (9) Bo Dai, The Chinese University of Hong Kong and The Chinese University of Hong Kong. Table of Links Abstract and 1 Introduction Abstract and 1 Introduction 2 Work Related 2 Work Related 3 Preliminary 3 Preliminary AnimateDiff AnimateDiff 4.1 Alleviate Negative Effects from Training Data with Domain Adapter 4.1 Alleviate Negative Effects from Training Data with Domain Adapter 4.2 Learn Motion Priors with Motion Module 4.2 Learn Motion Priors with Motion Module 4.3 Adapt to New Motion Patterns with MotionLora 4.3 Adapt to New Motion Patterns with MotionLora 4.4 AnimateDiff in Practice 4.4 AnimateDiff in Practice 5 Experiments and 5.1 Qualitative Results 5 Experiments and 5.1 Qualitative Results 5.2 Qualitative Comparison 5.2 Qualitative Comparison 5.3 Ablative Study 5.3 Ablative Study 5.4 Controllable Generation 5.4 Controllable Generation 6 Conclusion 6 Conclusion 7 Ethics Statement 7 Ethics Statement 8 Reproducibility Statement, Acknowledgement and References 8 Reproducibility Statement, Acknowledgement and References 5.2 QUANTITATIVE COMPARISON We conduct the quantitative comparison through user study and CLIP metrics. The comparison focuses on three key aspects: text alignment, domain similarity, and motion smoothness. The results are shown in Table 1. Detailed implementations can be found in supplementary materials. User study. In the user study, we generate animations using all three methods based on the same personalized T2I models. Participants are then asked to individually rank the results based on the above three aspects. We use the Average User Ranking (AUR) as a preference metric where a higher score indicates superior performance. Note that the corresponding prompts and images are provided for reference for text alignment and domain similarity evaluation. User study. CLIP metric . We also employed the CLIP (Radford et al., 2021) metric, following the approach taken by previous studies (Wu et al., 2023; Khachatryan et al., 2023). When evaluating domain similarity, it is important to note that the CLIP score was computed between the animation frames and the reference images generated using the personalized T2Is. CLIP metric This paper is available on arxiv under CC BY 4.0 DEED license. This paper is available on arxiv under CC BY 4.0 DEED license. available on arxiv

User Preferences and CLIP Metrics: Results of AnimateDiff’s Performance in Video Generation

Many crypto enthusiasts are on the hunt for the next “100x gem.” While it’s possible to find one, most people lose funds long before they do. Maybe if they opened their eyes and saw the behemoth right in front of them, they’d still have skin in the game. In 2024, well over a decade after its initial launch, Bitcoin is still untouched in terms of its influence and market position in the cryptocurrency space. Since 2016, other infrastructures like Ethereum and Solana have gained momentum in recent times through the introduction of smart contracts, NFTs, and more. However, the Bitcoin ecosystem has followed suit wherever it could thanks to additions such as the lightning network, runes, and ordinals. In a recent in-depth interview with Markus Bopp, the CEO of Tap Protocol, I explored the nuances, benefits, and challenges of these latest developments to unravel what the future truly holds for the BTC ecosystem. Let’s get to it. Understanding native bitcoin assets In simple terms, native bitcoin assets refer to applications and digital assets built directly on the so-called Layer 1 network of Bitcoin. As such, these don’t include third party networks such as Wormhole, where only wrapped BTC are used. These assets are inherently tied to Bitcoin's infrastructure, benefiting from its decentralization and immutability. Examples include using Bitcoin's Script language to create simple smart contracts or leveraging existing transaction types to develop new asset classes. In contrast, building alongside Bitcoin involves creating Layer 2 solutions, sidechains, or even forking Bitcoin's codebase to develop new, independent blockchains (which for example has previously happened with Bitcoin Cash, BSV, and Bitcoin Gold). Layer 2 solutions can provide more flexibility and functionalities, such as complex smart contracts and faster transaction times. However, they also come with unique challenges, particularly regarding security and asset traceability. “In terms of assets, the origin matters. Doesn’t matter if it’s for collectibles or NFTs. For assets that originate on an L2, tracing their origin can be challenging if the L2 ceases to exist. While you could argue that bridging these assets back to Bitcoin could solve the problem, you'd still need to recover the L2 protocol that created these assets. With multiple L2s, the likelihood of losing track of their origins becomes pretty high,” according to Bopp. The benefits of building on the Bitcoin blockchain Bitcoin's blockchain has historically been a hotbed of innovation. By building on Bitcoin, developers can create entirely new asset classes that are virtually unparalleled in security and exposure. After all, the biggest benefit of the Bitcoin network is its enormous size and security –its extensive decentralization ensures that assets remain virtually unhackable (e.g. through double spending attacks), making it an ideal platform for long-term projects. “This approach opens up tons of opportunities for consumers and miners. Many of them are finding new ways to take advantage of these options,” stated Bopp. The drawbacks Despite its advantages, building on Bitcoin does come with quite a few challenges. Many essential tools and products are yet to be developed, which can slow adoption and limit possibilities compared to more established platforms like Ethereum. Bitcoin's relatively simple scripting language lacks the capabilities of more advanced smart contract platforms. This can make developing complex dApps directly on Bitcoin more difficult. Moreover, unlike Ethereum, which generates a new block every 10-20 seconds, Bitcoin's slower block time of roughly 10 minutes can be a major drawback for time-sensitive use cases. While this ensures greater security, it may not be a dealbreaker for some developers. However, the Bopp said there is a flip side. “This might not be a real negative, as it allows new ecosystems to develop and new discoveries to be made.” The case for native Bitcoin assets Native Bitcoin assets offer superior security and traceability compared to those built on sidechains or L2 solutions. This makes them more reliable for long-term value storage and regulatory compliance. Developers can leverage its established security, decentralization, and regulatory standing by building directly on Bitcoin. This reduces risks and enhances the credibility of the assets created. “Bitcoin lacks native smart contract support. I can’t argue that. And yes, it has slower block finality. But building on Bitcoin as a platform remains advantageous. Building on the native chain is important since many L1-based products should inherit the regulatory characteristics already applied to Bitcoin,” according to Bopp. Given Bitcoin's established status and ongoing development, assets built on its blockchain are more likely to withstand the test of time, offering a safer investment for the future. The Growing Popularity of Bitcoin L2 Solutions Layer 2 solutions are gaining traction for their ability to address Bitcoin's scalability and speed issues. Technologies like the Lightning Network enable faster transactions and lower fees, making Bitcoin more practical for everyday use. But with Tap Protocol, it’s unnecessary. Tap’s solution enables smart contract development directly on the Bitcoin ecosystem without needing a Layer 2 built on top of it. “Tap Protocol created smart contract capability to take something DeFi users love and make it possible on the Bitcoin chain,” said Bopp. Bitcoin Is Still the Future of Finance and Innovation Bitcoin remains a foundational pillar in the evolving landscape of blockchain technology and digital finance. Its network offers unparalleled security and traceability for assets built directly on its blockchain. Tap Protocol’s developers have found the answer to one of the ecosystem’s biggest challenges with their product: enabling the use of smart contracts directly on Bitcoin’s blockchain. No sidechain or L2 is needed. The emerging tech and teams working on these solutions underscore Bitcoin's flexibility and ongoing potential to adapt and grow. As native Bitcoin assets and alternative blockchain solutions continue to develop, Bitcoin's established regulatory framework and foundational strengths ensure its continued relevance and prominence in the financial industry. Many crypto enthusiasts are on the hunt for the next “100x gem.” While it’s possible to find one, most people lose funds long before they do. Maybe if they opened their eyes and saw the behemoth right in front of them, they’d still have skin in the game. Many crypto enthusiasts are on the hunt for the next “100x gem.” While it’s possible to find one, most people lose funds long before they do. Maybe if they opened their eyes and saw the behemoth right in front of them , they’d still have skin in the game. behemoth right in front of them behemoth right in front of them In 2024, well over a decade after its initial launch, Bitcoin is still untouched in terms of its influence and market position in the cryptocurrency space. Since 2016, other infrastructures like Ethereum and Solana have gained momentum in recent times through the introduction of smart contracts, NFTs, and more. However, the Bitcoin ecosystem has followed suit wherever it could thanks to additions such as the lightning network, runes, and ordinals. In a recent in-depth interview with Markus Bopp, the CEO of Tap Protocol, I explored the nuances, benefits, and challenges of these latest developments to unravel what the future truly holds for the BTC ecosystem. Let’s get to it. Understanding native bitcoin assets Understanding native bitcoin assets In simple terms, native bitcoin assets refer to applications and digital assets built directly on the so-called Layer 1 network of Bitcoin. As such, these don’t include third party networks such as Wormhole, where only wrapped BTC are used. These assets are inherently tied to Bitcoin's infrastructure, benefiting from its decentralization and immutability. Examples include using Bitcoin's Script language to create simple smart contracts or leveraging existing transaction types to develop new asset classes. In contrast, building alongside Bitcoin involves creating Layer 2 solutions , sidechains, or even forking Bitcoin's codebase to develop new, independent blockchains (which for example has previously happened with Bitcoin Cash, BSV, and Bitcoin Gold). creating Layer 2 solutions creating Layer 2 solutions Layer 2 solutions can provide more flexibility and functionalities, such as complex smart contracts and faster transaction times. However, they also come with unique challenges, particularly regarding security and asset traceability. “In terms of assets, the origin matters. Doesn’t matter if it’s for collectibles or NFTs. For assets that originate on an L2, tracing their origin can be challenging if the L2 ceases to exist. While you could argue that bridging these assets back to Bitcoin could solve the problem, you'd still need to recover the L2 protocol that created these assets. With multiple L2s, the likelihood of losing track of their origins becomes pretty high,” according to Bopp. The benefits of building on the Bitcoin blockchain The benefits of building on the Bitcoin blockchain Bitcoin's blockchain has historically been a hotbed of innovation. By building on Bitcoin, developers can create entirely new asset classes that are virtually unparalleled in security and exposure. After all, the biggest benefit of the Bitcoin network is its enormous size and security –its extensive decentralization ensures that assets remain virtually unhackable (e.g. through double spending attacks), making it an ideal platform for long-term projects. “This approach opens up tons of opportunities for consumers and miners. Many of them are finding new ways to take advantage of these options,” stated Bopp. The drawbacks The drawbacks Despite its advantages, building on Bitcoin does come with quite a few challenges. Many essential tools and products are yet to be developed, which can slow adoption and limit possibilities compared to more established platforms like Ethereum. Bitcoin's relatively simple scripting language lacks the capabilities of more advanced smart contract platforms. This can make developing complex dApps directly on Bitcoin more difficult. Bitcoin's relatively simple scripting language Bitcoin's relatively simple scripting language Moreover, unlike Ethereum, which generates a new block every 10-20 seconds, Bitcoin's slower block time of roughly 10 minutes can be a major drawback for time-sensitive use cases. While this ensures greater security, it may not be a dealbreaker for some developers. However, the Bopp said there is a flip side. “This might not be a real negative, as it allows new ecosystems to develop and new discoveries to be made.” The case for native Bitcoin assets The case for native Bitcoin assets Native Bitcoin assets offer superior security and traceability compared to those built on sidechains or L2 solutions. This makes them more reliable for long-term value storage and regulatory compliance. Developers can leverage its established security, decentralization, and regulatory standing by building directly on Bitcoin. This reduces risks and enhances the credibility of the assets created. “Bitcoin lacks native smart contract support. I can’t argue that. And yes, it has slower block finality. But building on Bitcoin as a platform remains advantageous. Building on the native chain is important since many L1-based products should inherit the regulatory characteristics already applied to Bitcoin,” according to Bopp. Given Bitcoin's established status and ongoing development, assets built on its blockchain are more likely to withstand the test of time, offering a safer investment for the future. The Growing Popularity of Bitcoin L2 Solutions The Growing Popularity of Bitcoin L2 Solutions Layer 2 solutions are gaining traction for their ability to address Bitcoin's scalability and speed issues. Technologies like the Lightning Network enable faster transactions and lower fees, making Bitcoin more practical for everyday use. But with Tap Protocol, it’s unnecessary. Tap’s solution enables smart contract development directly on the Bitcoin ecosystem without needing a Layer 2 built on top of it. “Tap Protocol created smart contract capability to take something DeFi users love and make it possible on the Bitcoin chain,” said Bopp. Bitcoin Is Still the Future of Finance and Innovation Bitcoin Is Still the Future of Finance and Innovation Bitcoin remains a foundational pillar in the evolving landscape of blockchain technology and digital finance. Its network offers unparalleled security and traceability for assets built directly on its blockchain. Tap Protocol’s developers have found the answer to one of the ecosystem’s biggest challenges with their product: enabling the use of smart contracts directly on Bitcoin’s blockchain. No sidechain or L2 is needed. The emerging tech and teams working on these solutions underscore Bitcoin's flexibility and ongoing potential to adapt and grow. As native Bitcoin assets and alternative blockchain solutions continue to develop, Bitcoin's established regulatory framework and foundational strengths ensure its continued relevance and prominence in the financial industry.

Exploring Native Bitcoin Ecosystem Assets vs Alternatives 

Authors:
(1) Nora Schneider, Computer Science Department, ETH Zurich, Zurich, Switzerland (nschneide@student.ethz.ch);
(2) Shirin Goshtasbpour, Computer Science Department, ETH Zurich, Zurich, Switzerland and Swiss Data Science Center, Zurich, Switzerland (shirin.goshtasbpour@inf.ethz.ch);
(3) Fernando Perez-Cruz, Computer Science Department, ETH Zurich, Zurich, Switzerland and Swiss Data Science Center, Zurich, Switzerland (fernando.perezcruz@sdsc.ethz.ch). Table of Links Abstract and 1 Introduction 2 Background 2.1 Data Augmentation 2.2 Anchor Regression 3 Anchor Data Augmentation 3.1 Comparison to C-Mixup and 3.2 Preserving nonlinear data structure 3.3 Algorithm 4 Experiments and 4.1 Linear synthetic data 4.2 Housing nonlinear regression 4.3 In-distribution Generalization 4.4 Out-of-distribution Robustness 5 Conclusion, Broader Impact, and References A Additional information for Anchor Data Augmentation B Experiments 4.4 Out-of-distribution Robustness In this section, we evaluate the performance of ADA and compare it to prior approaches on tasks involving out-of-distribution robustness. We use the same datasets as [49] and closely follow their experimental setup. Data: We use four of the five out-of-distribution datasets used in [49]. First, we use RCFashionMNIST (RCF-MNIST) [49], which is a synthetic modification of Fashion-MNIST that models subpopulation shifts. Second, we investigate domain shifts using Communities and Crime (Crime) [12], SkillCraft1 Master Table (SkillCraft) [12] and Drug-target Interactions (DTI) [17] all of which are tabular datasets. For Crime, we use state identification, in SkillCraft we use "League Index", which corresponds to different levels of competitors, and in DTI we use year, as domain information. We split the datasets into train-, validation- and test data based on the domain information resulting in domain-distinct datasets. We provide a detailed description of datasets in Appendix B.4. Due to computational complexity, we could not establish a fair comparison on the satellite image regression dataset [23] (the fifth dataset in [49]), so we report some exploratory results in Appendix B.4. Models and comparisons: As detailed in Section 4.3. Additionally, we use a ResNet-18 [15] for RCF-MNIST and DeepDTA [38] for DTI, as proposed in [49]. Results: We report the RMSE and the "worst" domain RMSE, which corresponds to the worst within-domain RMSE for out-of-domain test sets in Table 2. Similar to [49], we report the R value for the DTI dataset (higher values suggest a better fit of the regression model). For full transparency, in the last row, we copy the results from [49]. We can assess that ADA is competitive with C-Mixup and the other data augmentation strategies. Under the same conditions (split of data and Neural network structure), ADA is superior to C-Mixup. But, the degree of improvement is marginal on some datasets and as we show in the last row, we could not fully replicate their results. ADA is significantly better than C-Mixup and other strategies on the SkillCraft data, in which ADA reduces the error by around 15% compared to the ERM solution. This paper is available on arxiv under CC0 1.0 DEED license. Authors: (1) Nora Schneider, Computer Science Department, ETH Zurich, Zurich, Switzerland (nschneide@student.ethz.ch); (2) Shirin Goshtasbpour, Computer Science Department, ETH Zurich, Zurich, Switzerland and Swiss Data Science Center, Zurich, Switzerland (shirin.goshtasbpour@inf.ethz.ch); (3) Fernando Perez-Cruz, Computer Science Department, ETH Zurich, Zurich, Switzerland and Swiss Data Science Center, Zurich, Switzerland (fernando.perezcruz@sdsc.ethz.ch). Authors: Authors: (1) Nora Schneider, Computer Science Department, ETH Zurich, Zurich, Switzerland (nschneide@student.ethz.ch); (2) Shirin Goshtasbpour, Computer Science Department, ETH Zurich, Zurich, Switzerland and Swiss Data Science Center, Zurich, Switzerland (shirin.goshtasbpour@inf.ethz.ch); (3) Fernando Perez-Cruz, Computer Science Department, ETH Zurich, Zurich, Switzerland and Swiss Data Science Center, Zurich, Switzerland (fernando.perezcruz@sdsc.ethz.ch). Table of Links Abstract and 1 Introduction Abstract and 1 Introduction 2 Background 2.1 Data Augmentation 2.1 Data Augmentation 2.2 Anchor Regression 2.2 Anchor Regression 3 Anchor Data Augmentation 3 Anchor Data Augmentation 3.1 Comparison to C-Mixup and 3.2 Preserving nonlinear data structure 3.1 Comparison to C-Mixup and 3.2 Preserving nonlinear data structure 3.3 Algorithm 3.3 Algorithm 4 Experiments and 4.1 Linear synthetic data 4 Experiments and 4.1 Linear synthetic data 4.2 Housing nonlinear regression 4.2 Housing nonlinear regression 4.3 In-distribution Generalization 4.3 In-distribution Generalization 4.4 Out-of-distribution Robustness 4.4 Out-of-distribution Robustness 5 Conclusion, Broader Impact, and References 5 Conclusion, Broader Impact, and References A Additional information for Anchor Data Augmentation A Additional information for Anchor Data Augmentation B Experiments B Experiments 4.4 Out-of-distribution Robustness In this section, we evaluate the performance of ADA and compare it to prior approaches on tasks involving out-of-distribution robustness. We use the same datasets as [49] and closely follow their experimental setup. Data : We use four of the five out-of-distribution datasets used in [49]. First, we use RCFashionMNIST (RCF-MNIST) [49], which is a synthetic modification of Fashion-MNIST that models subpopulation shifts. Second, we investigate domain shifts using Communities and Crime (Crime) [12], SkillCraft1 Master Table (SkillCraft) [12] and Drug-target Interactions (DTI) [17] all of which are tabular datasets. For Crime, we use state identification, in SkillCraft we use "League Index", which corresponds to different levels of competitors, and in DTI we use year, as domain information. We split the datasets into train-, validation- and test data based on the domain information resulting in domain-distinct datasets. We provide a detailed description of datasets in Appendix B.4. Due to computational complexity, we could not establish a fair comparison on the satellite image regression dataset [23] (the fifth dataset in [49]), so we report some exploratory results in Appendix B.4. Data Models and comparisons: As detailed in Section 4.3. Additionally, we use a ResNet-18 [15] for RCF-MNIST and DeepDTA [38] for DTI, as proposed in [49]. Models and comparisons: Results : We report the RMSE and the "worst" domain RMSE, which corresponds to the worst within-domain RMSE for out-of-domain test sets in Table 2. Similar to [49], we report the R value for the DTI dataset (higher values suggest a better fit of the regression model). For full transparency, in the last row, we copy the results from [49]. We can assess that ADA is competitive with C-Mixup and the other data augmentation strategies. Under the same conditions (split of data and Neural network structure), ADA is superior to C-Mixup. But, the degree of improvement is marginal on some datasets and as we show in the last row, we could not fully replicate their results. ADA is significantly better than C-Mixup and other strategies on the SkillCraft data, in which ADA reduces the error by around 15% compared to the ERM solution. Results This paper is available on arxiv under CC0 1.0 DEED license. This paper is available on arxiv under CC0 1.0 DEED license. available on arxiv

ADA's Impact on Out-of-Distribution Robustness

Stories, News & Mentions about SUPERBIO

Superbio’s Stories on HackerNoon