Spinzo, a leading provider of customized ticket sales solutions for sports teams, venues, and events, has made a major donation to Shriners Children's, an organization committed to improving the lives of children every day by providing world-class pediatric specialty care,

Spinzo Makes Major Donation to Shriners Children's

With the new Colorado Avalanche and Denver Nuggets campaigns having already launched, spectators should be aware of the updated bag guidelines at Ball Arena. Those familiar with the previous bag policy will need to downsize. Previously, bags were limited ...

Ball Arena's updated bag policy: What you need to know

With the new Colorado Avalanche and Denver Nuggets campaigns having already launched, spectators should be aware of the updated bag guidelines at Ball Arena.

DETROIT, Sept. 22, 2023 /PRNewswire/ - Spinzo, a leading promotional sales platform driving ticket sales programs for sports teams, venues, and events, nationwide, has made a major gift of support ...

Spinzo Rallies to Support Special Spectators VIP Game Days Nationwide

DETROIT, Sept. 22, 2023 /CNW/ - Spinzo, a leading promotional sales platform driving ticket sales programs for sports teams, venues, and events, nationwide, has made a major gift of support to ...

Get the most recent info and news about Spinzo on HackerNoon, where 10k+ technologists publish stories for 4M+ monthly readers.

Spinzo is the social group sales platform used by major sports leagues, colleges, and venues to sell tickets and packages with speed, flexibility, and style. Group Experience Tracking drives groups to hit big milestones. No-Hassle Seat Selector gives buyers the choice they need to commit. Ticketmaster Integration automatically reserves seats from open or pre-held inventory. Competitive Fundraising with live leaderboards makes group fundraising a breeze. Self-service dashboard puts you in control – Create offers in seconds. Track in real-time. #CrowdPricing option drives more buyers – the MORE who buy, the LESS everyone pays. ***** Your brand. Our technology. Everyone wins. ***** Get in touch with us today: hello@spinzo.com

Spinzo

Meet the Winners of Startups in North and Central America

The ideology serves as a powerful mechanism not only for uniting individuals with similar interests but also for exchanging experiences and ideas, embodying the principle that sharing is indeed caring. is a great example of how a community of enthusiasts can change the world. Its are highly impressive and we can claim that most of the Internet works on Linux’s horsepower. open source Linux usage statistics Bad programmers worry about the code. Good programmers worry about data structures and their relationships. Linus Torvalds Managing an open-source project carries significant weight, particularly as its popularity grows: it brings forth the spotlight on technical expertise and architectural choices. Popular projects could be a treasure trove of knowledge for anyone who wants to boost their skills. Let's delve deeply into a few notable projects. Why focus on open source? First of all, it embodies top-tier coding standards, integrating best practices championed by industry leaders. My firm belief lies in the notion that genuine programmers are fueled by passion and curiosity. Why not learn something new and integrate it into our daily practice? Secondly, hacking code could take your skills to a whole new level if you are consistent and tenacious. Last but not least, it could help you jump right into the project you like: find new buddies, brand yourself as a skilled developer, and get the job of your dreams. Help the community to solve their problems, and the reward will not keep you waiting. Take your cup of coffee, let’s get straight to work. Linux Kernel The very heart of the Linux system — its kernel. Indeed, this is not merely a singular project, but rather a constellation of diverse projects unified by a central idea: tying up different parts of hardware and providing a sustainable, secure API for the end users. The Linux codebase is huge and contains a lot of vendor-specific code, but it is well-organized and structured. I strongly recommend everyone interested to read . It might be a bit outdated but it will foster a solid understanding of how the kernel internals work. The kernel version I have on the table is 6.3. this Radix tree for memory management At the very beginning of the main.c we can spot definitions of some functions like . It is responsible for initializing the data structure that is used in memory management processes. radix_tree_init In the codebase, the radix_tree_root struct is an alias for a wrapper above the struct. It a lock (pthread mutex); xa_flags — the enum that tells how memory allocations will be performed; and the pointer to xa_head. XArray holds gfp_t struct xarray {
	spinlock_t	xa_lock;
/* private: The rest of the data structure is not to be used directly. */
	gfp_t		xa_flags;
	void __rcu *	xa_head;
}; typedef enum {
	GFP_KERNEL,
	GFP_ATOMIC,
	__GFP_HIGHMEM,
	__GFP_HIGH
} gfp_t; If all of the entries in the array are NULL, xa_head is a NULL pointer. If the only non-NULL entry in the array is at index 0, xa_head is that entry. If any other entry in the array is non-NULL, xa_head points to a xa_node — the internal struct that handles offset, counts, slots, and other fields. struct xa_node {
	unsigned char	shift;		/* Bits remaining in each slot */
	unsigned char	offset;		/* Slot offset in parent */
	unsigned char	count;		/* Total entry count */
	unsigned char	nr_values;	/* Value entry count */
	struct xa_node __rcu *parent;	/* NULL at top of tree */
	struct xarray	*array;		/* The array we belong to */
	union {
		struct list_head private_list;	/* For tree user */
		struct rcu_head	rcu_head;	/* Used when freeing node */
	};
	void __rcu	*slots[XA_CHUNK_SIZE];
	union {
		unsigned long	tags[XA_MAX_MARKS][XA_MARK_LONGS];
		unsigned long	marks[XA_MAX_MARKS][XA_MARK_LONGS];
	};
}; The radix tree data structure contains “nodes” that link with each other based on the int key (index). The node contains several slots, each of which can handle a pointer. The main idea here is to keep a tree wide, not deep — it is effective for fast addressing. When the function is called by a user (it can be a device driver), the following actions occur: radix_tree_insert __radix_create_tree creates a slot in the tree if it is necessary, based on the passed index which is used to check the size of the tree. Or raise an error if there is no available memory; insert_entities pass an item to the slot; some asserts check that offsets are applied correctly to the data structure. Wherever a user of the tree is going to look up a specific page, they need to look up it with an index. The lookup phase is quite efficient because of the broad nature of the structure. It is based on the key that is modified on the user side. Take a look at the for more insight into how it works. article Users of this data structure include GPU and network device drivers, which maintain memory page pointers in a structured format rather than juggling them. It's like having a pocket dimension! This is where Linus's earlier quote comes into play. Consider the essence of the data you handle; aim for efficiency, compactness, and cleanliness in its usage, regardless of the language or project you're involved in. In 2021 Liam Howlett (no, another Liam) and Matthew Wilcox introduced data structure. I encourage you to read more about the intentions, motivation, and realization details and . the maple tree here here Kubernetes Let's direct our steps to the point of the architectural design. One of the most famous orchestrators, invented by Google and spread around the world at rocket pace — Kubernetes. There are a lot of fairly new resources out there covering the internal details, for example here is a that you can start exploring as you delve deeper into the details. repository The whole point of this article is to get some useful insights that answer not just “how” but “why”. Understanding the motivation is crucial, in my opinion, that is how we can “catch a mindset” to re-implement all those cool staff in our daily job. Let’s briefly recall what the Kubernetes is. This is the system that manages containers within the applications and provides orchestration features such as scaling and high availability. Kubernetes’ main components are the API service, the scheduler, etcd — the key-value storage, the controller service, and the agent service — “kubelete” on each node to manage containers. Guys from developed it to utilize the compute resources of their cloud . Their main intention was a “great container management system”. Google as efficiently as possible Architectural highlights The very first thing I’d like to highlight is the distributed nature of execution. To synchronize all of the user’s entities states the orchestrator uses etcd — the key-value storage which information in a tree data structure. It is a common pattern that presents a to manage different states between multiple entities, with one slight difference: the storage is also used to “instruct” all of the components about the final cluster’s state. keeps So the whole system uses the storage as a source of truth to keep maintaining the desired state of a cluster. source of truth One more nice architectural decision is the — container runtime interface. CRI It allows Kubernetes to work not just with Docker but many others (CRI-O, Dockerd) by defining the , just like one from RFC. The kubelet agent utilizes GRPC to connect with a runtime, so in case one would like to implement its own runtime, they should implement the exact interface on their side. The concept aims to loosen the connection between the kubelet agent and a runtime which manages containers. interaction protocol At the end let’s briefly look at the “sidecar” pattern that was presented by third-party services like and has recently been within the core Kubernetes functionality. Istio implemented The whole idea is to embed a service to the main application that helps to handle some specific cases such as proxy or load-balancing. Kubernetes manages this container as a whole different entity of a pod: the container can the entire pod status; restarting and updating separately of the main application; resources with the main app. influence share The main point here is that the main application may know nothing about the sidecar, while the behavior of the entire service can be changed. Bless Open Source, we how new functionality was implemented by the community. can trace PostgreSQL PostgreSQL is not just a database but a whole platform. After the client connects to the server a new is established. The query goes through many while executed: parsing, the rewriting system, planning and executing. backend process stages The PSQL server has a bunch of that allow developers to interact with different internal components. For example, we can our scan provider aimed, for some optimizations at each stage. Or create an . Even though the overall process required for specific knowledge and the documentation could be a bit knotty, interfaces write index add-on the loosely coupled architecture empowers developers to vastly expand functionality. There are many that we can use within the platform and the “bloom index” is one of them. extensions realization Bloom index I faced a problem of growing indexes back in the day. The task was to implement the custom search filter: every column in the database and the combination of it could be a key for a search. Building indexes for each and every column could be a bad decision because of the complexity and size. For example, if we are sure that item==2 is False there is no reason to check all other fields. I remembered the — the probabilistic data structure that could return a false-positive result (because of hash collision) and be quite sure when the data is absent. As I’ve mentioned earlier PSQL has an extension for this data structure. The very first idea that comes to mind is to reduce the dimension of possible variations removing the query values that we are sure are not in the database. Bloom filter Bloom filter in a nutshell — it’s just a bit array. Every item would be hashed with a certain amount of different functions that translates it to some index. That index will set the exact bit in the filter’s array. In the case of 3 different functions, an item would set 3 different bits. At the lookup phase, the query would be also hashed and compared within the bits in the array. If at least one of the bits is not set, we can be sure the item is absent. If all of them were set, it is still possible that the item was not placed — that is because each bit we have found could refer to other items — the collision takes place. Check out this great of how it works within the examples. explanation To start using the bloom index within the PSQL we need to create it properly: CREATE INDEX bloom_idx_bar ON foo.bar USING bloom (id,dept_id,zipcode) WITH (length=80, col1=4, col2=2, col3=4); where “length” — is the length of the array of bits for each row; and the “col” numbers — describe how many bits would be occupied by the column (how many functions would be applied). By increasing these numbers we can reduce false-positive shots. Of index creation in PSQL is a bit complicated, we should understand internal data structures and protocols; let’s try to make the observation short and sweet. The function promotes the data structure’s functions as an IndexAmRoutine pointer — a pointer to the , that represents an API for the index access methods. blhandler struct The function creates the new index: allocate memory pages, scan the table for tuples that will be indexed and in the structure. blbuild BloomState typedef struct BloomState
{
	FmgrInfo	hashFn[INDEX_MAX_KEYS];
	Oid			collations[INDEX_MAX_KEYS];
	BloomOptions opts;			/* copy of options on index's metapage */
	int32		nColumns;

	/*
	 * sizeOfBloomTuple is index-specific, and it depends on reloptions, so
	 * precompute it
	 */
	Size		sizeOfBloomTuple;
} BloomState; When we insert data to the index, we pass it in the form of an array within the array of booleans that marks NULL values to skip it at the indexing phase. The BloomFormTuple function forms the BloomTuple struct from the data — this is the part where the data would be signed. The tuple would be passed to a memory page after that phase. BloomTuple *
BloomFormTuple(BloomState *state, ItemPointer iptr, Datum *values, bool *isnull)
{
	int			i;
	BloomTuple *res = (BloomTuple *) palloc0(state->sizeOfBloomTuple);

	res->heapPtr = *iptr;

	/* Blooming each column */
	for (i = 0; i < state->nColumns; i++)
	{
		/* skip nulls */
		if (isnull[i])
			continue;

		signValue(state, res->sign, values[i], i);
	}

	return res;
} At the scan phase the forms the struct which contains the index’s data and the makes the TIDBitmap struct from it — the internal data structure that holds item pointers — reading BloomState page by page. blbeginscan IndexScanDesc blgetbitmap I hope the explanation isn't too confusing, PSQL internals are quite specific. Bloom filter is a pretty common data structure, PSQL uses the same approach for some purposes. It is also broadly used in many different situations: from cache-hitting cases to implementations. internal kernel packet filter More to know In the article, I’ve just lifted the veil of the shiny open-source world. There are many more great implementations you could meet in the wild: and — if you’d like to catch the magic of the “fair” resource allocation; — to understand the simultaneous editing in Google Docs. CFS CFQ operational transform and CRDT The whole software world is in motion: sidecars and operators to be the new model for software delivery; eBPF is in a to change the service-mesh post. claimed hurry Take a look at the elegance of the great things’ internals.

Algorithms in the Wild: Reusable Insights and Strategies from the Open Source Community

Some people may not know it, but I actually studied to be a teacher. At a certain point, I realized I didn’t want to work as a teacher. By that time, I had already started my career in the customer service and support industry, so the decision to leave teaching behind came naturally to me. That doesn’t mean my time in the university was wasted. On the contrary — today, in my position as a CEO and Managing Partner of a , I find the knowledge I got while I was studying more and more. Support-as-a-Service company Here are the three best approaches I brought from teaching to being a CEO. Giving advice and steering in the right direction Almost every day, I find that my role as a and Managing Partner is evolving. Today, it has much more to do with mentoring than with managing actual workflows and processes. Being a mentor involves giving my team advice and inspiring them for new ideas and future achievements. This is not a management technique I have only just picked up. This is something my team and I have been practicing for years. CEO The best example of this approach is the fact that about 60% of SupportYourApp’s management team have started their careers as customer support consultants, and today, they work in our Technical, Cybersecurity, Integrations, and other company departments. Seeing the potential, giving the advice to develop it, and mentoring have also helped me spot the right candidates for the position of CEO of SupportYourApp’s spinoff companies — and . Today, these are independent projects with their own KPIs, goals, and teams, so I know I made the right choice. Label Your Data Outstaff Your Team Steering talents in the right direction and showing them the best ways to use their skills for their professional and personal growth is one of the most pleasant tasks of any manager. At least, in my book. Cheering the growth of others What do teachers find most rewarding in their job? I would say seeing the growth and development of their students. It is the same for me. Earlier, when a newbie joined SupportYourApp, I personally tried to oversee the development of their potential and career. Today, with over 1400 talents on our team, it has become an impossible task. What helps me here is delegating. Even so, hearing about a new promotion or seeing someone grow from a customer support consultant to a Service Delivery Manager or an HR Specialist still brings me an indescribable amount of joy. I know — like a good teacher, we are helping people find their calling and change their lives for the better, fulfilling “People First” — one of the corporate values. SupportYourApp’s When I think about it, it just brings a smile to my face. Implementing active listening during almost every conversation “It’s not what you said, it’s the way you said it” — say what you will, but Joey Tribbiani clearly was the master of active listening. Just like for a good teacher, active listening is one of the cornerstones of our internal communication. It involves taking in not only the information but also paying attention to the mood and the tone of the dialogue. How does that help us in our everyday processes? First, we are able to pick up on the smallest clues and spot those members of our team who are on the verge of burnout. It enables us to suggest they take a rest or, if possible, change their project or position. Second, active listening promotes empathy and respect, which helps us build trust and create a supportive atmosphere where everyone is open to communicating and suggesting their ideas. I personally love to engage this skill during our brainstorming sessions because it allows us to get to the most creative and efficient solution. Active listening is the key skill behind our effectiveness and mutual understanding and respect. Without it, I would not be able to build a team I trust so deeply. Today, I truly believe that my choice of a profession of a teacher was not an accident. The knowledge I gained during my time at the university has helped me build the team I have today. A team of reliable, trustworthy, and skillful professionals whom I believe in wholeheartedly. I can apply the knowledge I got in the past to change our common present and build a better future. And isn’t it what teaching is all about?

How My Teacher Education Helped Me in My Job as a CEO

So, let’s start! Tell us a bit about yourself. For example, name, profession, and personal interests. My name is Daria Leshchenko; I am the CEO and Managing Partner at SupportYourApp — a Support-as-a-Service company that provides technical and customer support to global startups and businesses. I am also a co-founder of our two spinoff companies — Label Your Data — which deals with data labeling for AI and ML, and Outstaff Your Team, which takes care of the outstaffing services for global tech businesses. In whatever time I can spare, I love doing sports (squash, for example, is my latest activity of choice), studying (management, coaching, latest technologies — whatever it is, I’m in), and traveling. For example, I have just returned from Lisbon, where I attended this year’s Web Summit. Interesting! What was your latest Hackernoon Top story about? The latest one that made it to the HackerNoon TOP list was “Founders, Do You Have a Plan for the End of the World?”. My team and I kept on working and developing through COVID-19, lockdowns, and every other crisis global businesses faced over the last decade. In the article, I wanted to share my thoughts and experiences on withstanding even the biggest shakes and turmoils. I think it is especially important now when the world around us is so unstable and uncertain. Do you usually write on similar topics? If not, what do you usually write about? I usually write about the challenges that I have as a founder, mistakes I have made earlier in my career, advice I would love to share with others, and so on. Something that is near and dear to my heart. I believe this is the best approach for any writer. Great! What is your usual writing routine like (if you have one?) As soon as I narrow the topic down, I just start writing and perfecting the text to make sure it is as engaging as it can be. I also make it as short as I can because I like to be on point and precise. Why waste time writing 10 words, when a thought can be expressed in 5? This is my golden rule of writing. Being a writer in tech can be a challenge. It’s not often our main role, but an addition to another one. What is the biggest challenge you have when it comes to writing? I would say time is my biggest constraint. Writing helps me structure my thoughts and experiences. Almost relive it. And I hope every piece I write helps someone make the correct decision or find the right solutions. As every top-tier and C-suite manager knows, our schedules can get quite hectic, so time is a crucial factor in our ability to share our knowledge with our peers and teams. What is the next thing you hope to achieve in your career? The next step for SupportYourApp and me (because, in my mind, we are inseparable) is implementing more AI solutions into our support services. We also keep on adding more value to our core services and offering complex solutions. The progress we are seeing now is not going to slow down, so it is important for businesses and leaders to not follow it, but actually get ahead of it. I would love to concentrate my attention on creating an ecosystem that will unite people and technology. That is something we have already been doing, but as time and progress go by, I think we need to dive deeper and deeper into the process. The business world sees a slowdown, so I would like to keep the pace up even despite the economic crisis the world is facing right now. I want us to keep on growing despite it all. Wow, that’s admirable. Now, something more casual: What is your guilty pleasure of choice? Interesting question. Now that I think about it, if something brings me pleasure, I do not feel guilty about it. It is as simple as that. So, I guess I can say I don’t have one. Do you have a non-tech-related hobby? If yes, what is it? Sports and reading. These are something my team and I share — a lot of us love running, doing yoga, and generally taking care of ourselves. Only in 2023, we took part in two races in our hometown of Kyiv and our Colombian team took part in the Intercontinental Cali 10K De La Luz run. I run and play squash. It keeps my head clear and relaxes me. As for reading — my team and I are on our 11th book of the year right now. In 2023, we tackled books like The Remains of the Day by Kazuo Ishiguro (which is my favorite this year) and Limits to Growth: The 30-Year Update by Donella H. Meadows, Jorgen Randers, and Dennis Meadows. What can the Hacker Noon community expect to read from you next? It will depend on the challenges we will be facing at that particular moment and the evolution of my interests. What’s your opinion on HackerNoon as a platform for writers? It gives us the possibility to open up about things that worry us, as well as find people who share our interests and can discuss these topics with us. It helps us create and grow a community of peers. HackerNoon is also a good resource for those who are seeking knowledge and want to learn more about complicated matters, but in simple words. Thanks for taking time to join our “ ” series. It was a pleasure. Do you have any closing words? Meet the writer Stay curious and follow the flow, everybody.

Meet the Writer: HackerNoon's Contributor Daria Leshchenko, the CEO of SupportYourApp

In recent years, we witnessed a curious phenomenon—digital humans and deepfakes—that appeared at the intersection of artificial intelligence (AI) and media technology. China, known for its technological advances, was one of the first countries to adopt and nurture the development and deployment of these technologies. In this article, I will demonstrate the use of AI-driven digital humans and deepfakes in various industries and media channels. In the example of China, we can see what potential these technologies can bring and what ethical considerations they may raise. First of all, let’s find out what digital humans are. They are images and videos of digitally created people, so-called AI avatars, that are able to communicate with users and viewers in a very human-like way—in real-time. They are created using a number of the latest AI technologies, such as computer vision, machine learning, and natural language processing. Computer vision is used to capture images and videos of real people on which the AI avatars will be based. Machine learning algorithms train a model to generate realistic images and videos of the person. For some algorithms, only a few photos or a short video of a real person is enough to create a convincing AI clone—a digital persona that would move and interact in a realistic way. Natural language processing algorithms and speech-generating tools enable the digital avatars to communicate like real people. Combined, these tools produce an exceptionally realistic picture that often can be mistaken for a real person. AI Avatar Generators Answering the growing demand for AI avatars, Chinese companies produce more and more affordable AI generative tools that can be used for digitizing real humans as well as creating unique digital personas. Meishe, a solution provider that focuses on independent development of video and audio SDK for mobile devices, has recently revealed three AI avatar generators: they allow creating an avatar using just a photo, video, or 3D model (GLB file) in a few minutes time—for a very reasonable price. This makes the technology more accessible both for businesses and individuals. Concurrently, , a leading Chinese tech company, is starting the internal testing of its platform for interaction with digital personas. Their solution is based on an advanced language model that allows cross-format generation. Qihoo 360 Unlike Meishe, which bases its AI avatars on real humans, Quihoo 360 focuses on creating a wide range of personalities such as digital employees for marketing, operations, and programming purposes, as well as an interactive platform that includes classic Chinese literary figures. Deepfake Technology Another Chinese Tech giant, Tencent, put its effort into democratizing deepfake technologies. The company recently presented a platform that allows the creation of a deepfake digital human for an astonishingly . affordable price of $145 This new service allows anyone to make a high-definition digital human based on just three minutes of live-action video of a real person and 100 sentences of voice material. Digital Humans in Business In China, companies have already started to employ virtual humans for various tasks. “Hiring” a virtual person would cost approximately $14,000 a year, which can be considered pretty cost-efficient. AI-powered digital humans can help streamline operations and also open up new possibilities for engaging with customers and stakeholders 24/7. Digitized humans can answer questions, interact on a live stream, and even sing. This means that their use is not limited to just chatbots, automated technical support, and customer support—even though it’s becoming very common in many business areas such as financial services and tourism. AI avatars can also act as social media influencers and even become broadcasters on the state media. According to , the use of virtual humans in various business areas is expected to grow by 50% annually through 2025—thanks to the rapid development of the technology, its increasing availability, and new applications in various fields. CNBC AI News Anchors Recently, Chinese media has introduced who would engage with viewers and answer questions during important social events, particularly, the “Two Sessions”—the Chinese government's annual plenary sessions held every March in Beijing. AI news anchors These digital news anchors are able to provide real-time responses and engage with the audience. It is evident that the Chinese government views AI avatars as a significant asset that can help advance and propagate Chinese culture and values, investing in the development of the technology. Influencer Live Streaming Deepfakes, which are the AI avatars of real people, have become a significant part of Chinese e-commerce, particularly among livestream influencers. Live streaming has transformed into a billion-dollar industry in just a few years and is expected to grow by at least 20% a year in the future. Deepfakes take on the appearances and voices of and seamlessly integrate into livestreams, taking content creation to a whole new level. popular influencers The integration of deepfake technology into live streams is a powerful tool that may redefine influencer marketing for e-commerce purposes. Chinese companies are actively integrating this technology to boost sales and brand reach, and it proved to be highly beneficial for the overall viewing experience and elevates the revenue from livestream shopping. However, this advancement also raises ethical concerns. Many people see the need for regulations that would ensure transparency and protect consumers from fraud and misinformation. AI Cloning To create highly interactive and personalized AI-driven avatars, some Chinese tech companies have introduced AI cloning programs. The idea is to use volunteers and testers who will participate in recording their voices and appearances for the creation of more advanced and realistic AI personas based on that data. For example, Microsoft spinoff Xiaoice in China and Japan, seeking testers who can “lend” the AI avatars their personas, voices, and appearances in order to further refine and develop virtual beings. If the first testing group appears successful, the company is planning to expand the number of AI avatars to a hundred thousand. has launched an AI clone program Other China's tech giant, SenseTime, based on this method. In addition to a chatbot, the company showcased a digital avatar creation platform and a pair of complementary 3D modeling tools, showcasing the integration of AI-driven virtual assistants into various applications and industries. has unveiled new AI products These initiatives can help integrate highly personalized AI clones and digital humans into our daily lives, however, this approach poses a risk of unauthorized data collection. Many AI tools over the world have already been accused of collecting users’ data without consent. With the growing amount of video and audio content on the web, chances are high that anyone’s social media posts can become training material for an AI cloning tool. Market Perspectives China is one of the world leaders in AI development, amassing 19% of the total AI funding deals globally. It is also one of the major producers and consumers of the digital avatar industry. According to , the global digital human avatar market is expected to reach USD 527.58 billion by 2030, with the Asia-Pacific region accounting for the largest share. a report by Emergen Research The key factors that facilitate explosive growth of the AI avatar market in China are the following: Social media propelled demand for virtual content and surging metaverse trends. AI avatars help users engage in interactive virtual experiences and allow the creation of high-quality content 24/7. Dropping costs of developing and managing AI avatars make the technology largely accessible for Chinese businesses and individuals, helping it gain a broader audience. Government support and encouragement play an indispensable role in the country’s economy, amplifying the potential of Chinese AI solutions on the global market.

The Rise of Digital Humans and Deepfakes in China

Making new mistakes I’ve been a software engineer for about 15 years now. Throughout my career I learned a lot and applied these learnings to design and implement (and occasionally phase out or leave as is) many distributed systems. Along the way I made numerous mistakes and still keep making them. But as my main focus was reliability I’ve been looking back at my experience and the community to find ways to minimize the error frequency. My motto is: we must absolutely try making new mistakes (less obvious, more sophisticated). Making a mistake is fine - that is how we learn, repeating - is sad and discouraging. That’s probably what has always fascinated me about mathematics. Not only because it is elegant and concise, but because its logical rigor prevents mistakes. It forces you to think about your current context, what postulates and theorems you can rely on. Following these rules proves to be fruitful, you get the correct result. It is true, that computer science is a branch of mathematics. But what we usually practice is software engineering, a very distinct thing. We apply computer science achievements and discoveries to practice, accounting for time constraints and business needs. This blog is an attempt to apply semi-mathematical reasoning onto design and implementation of computer programs. We’ll put forward a model of different execution provides a framework to avoids many programming errors. regimes From humble beginnings When we learn to program and make our first tentative (or daring) steps we usually start with something simple: programming loops, doing basic arithmetic and printing the results in a terminal solving math problems, probably in some specialized environment as MathCAD or Mathematica We acquire muscle memory, learn language syntax and most importantly we change the way we think and reason. We learn to read the code, make assumptions about how it is being executed. We almost never start by reading a language standard and closely read through its “Memory model” section - because we are not yet equipped to fully appreciate and make use of those. We practice trial and error: we do introduce logical and arithmetic bugs in our first programs. These mistakes teach us to check our assumptions: is this loop invariant correct, can we compare array element’s index and length this way (where do you put this -1)? But if we don’t see some kind of errors, oftentimes implicitly we internalize some the system enforces and provides us. invariants Namely this one: Lines of code are always evaluation in the same order (serialized). This postulate allows us to assume that next propositions are true (we are not going to prove them): evaluation order does not change between executions function calls always return Mathematical axioms allow to derive and build larger structures on a solid basis. In mathematics, we have with 4+1 postulates. The last one says: Euclidean geometry parallel lines stay parallel, they do not intersect or diverge For millennia mathematicians tried to prove it, and derive it from the first four. Turns out that is not possible. We can replace this “parallel lines” postulate with alternatives and get different kinds of geometries (namely, hyperbolic and elliptic), which open up new prospects and turn out to be applicable and useful. After all our own planet’s surface is not flat and we have to account for that e.g. in GPS software and airplane routes. The need for change But before that let’s stop and ask the most engineering questions: why bother? If the program does its job, it is easy to support, maintain and evolve, why should we give up this cozy invariant of predictable sequential execution in the first place? I see two answers. First one is . If we can make our program run twice as fast or similarly - require half of the hardware - this is an engineering achievement. If using the same amount of computational resources we can grind through 2x (or 3, 4, 5, 10x) of data - it may open completely new applications of the same program. It may run on a mobile phone in your pocket instead of a server. Sometimes we can achieve speeds up by applying clever algorithms or rewriting in a more performant language. These are our first options to explore, yes. But they have a limit. Architecture almost always beats the implementation. Moor’s law has not been doing that well lately, a single CPU’s performance is growing slowly, RAM performance (latency, mainly) is lagging behind. So, naturally, engineers started to looking for other options. performance Second consideration is . Nature is chaotic, second law of thermodynamics constantly works against anything precise, sequential and repeatable. Bits flip, materials degrade, power browns out, wires get cut preventing our programs execution. Keeping sequential and repeatable abstraction becomes a hard job. If our programs could outlive software and hardware failures we could deliver services that have a competitive business advantage - that’s another engineering task we can start addressing. reliability Equipped with the goal, we can start experiments with non-serialized approaches. Threads of execution Let’s look at this chunk of pseudo code: ``` def fetch_coordinates(poi: str) -> Point: … def find_pois(center: Point, distance: int) -> List[str]: … def get_my_location() -> Point: … def fetch_coordinates(p) - Point: … def main(): me = get_my_location() for point in find_pois(me, 500): loc = fetch_coordinates(point) sys.stdout.write(f“Name: {point} is at x={loc.x} y={loc.y}”) We can read the code top-to-buttom and reasonably assume that `find_pois` function will be called `get_my_location`. And we’ll fetch and return coordinates of the first POI after fetching the next one. Those assumptions are correct and allow building a mental model, reason about the program. after Let’s imagine we can make our code to execute non-sequentially. There are many ways we can do that syntactically. We’ll skip experiments with statement reordering (that’s what modern compilers and CPUs do) and extend our language so that we’re able to express a new execution regime: or even with respect to other functions. Rephrasing, we need to introduce multiple of execution. Our program functions execute in a specific environment (crafted and maintained by OS), right now we’re interested in addressable virtual memory and a thread - a scheduling unit, something that can be executed by a CPU. function concurrently in parallel threads Threads come in different flavours: a POSIX thread, green thread,  coroutine, goroutine. Details differ greatly, but it boils down to something that can be executed. If several functions can run concurrently each needs its own scheduling unit. That is were multi-threading comes from, instead of one, we have several threads of execution. Some environments (MPI) and languages can create threads implicitly, but usually we have to do this explicitly using `pthread_create` in C, `threading` module classes in Python or a simple `go` statement in Go. With some precautions we can make the same code run mostly in-parallel: def fetch_coordinates(poi, results, idx) -> None:
  …
  results[idx] = poi


def main():
  me = get_my_location()
  points = find_pois(me, 500)
  results = [None] * len(points)  # Reserve space for each result
  threads = []
  for i, point in enumerate(find_pois(me, 500)):  # i - index for result
    thr = threading.Thread(target=fetch_coordinates, args=(poi, results, i))
    thr.start()
    threads.append(thr)
 for thr in threads:
   thr.wait()
for point, result in zip(points, results):
  sys.stdout.write(f“Name: {poi} is at x={loc.x} y={loc.y}”) our program can run on multiple CPUs and scale as the number of cores grows and finishes faster. Next engineering question we must ask: at what cost? We achieved our performance goal: We intentionally gave up on serialized and predictable execution. There is between a function + point in time and the data. At each point in time there is always a single mapping between a running function and its data: no bijection Next consequence is that one function can finish before another this time, next time it can be the other way. This new regime of execution leads to data races: when concurrent functions work with data it means that order of operations applied to the data is undefined. We start encountering data races and learn to deal with them using: critical sections: mutexes (and spinlocks) lock-free algorithms (the simplest form is in the snippet above) race detection tools etc At this point, we discover at least two things. First, there are multiple ways to access data. Some data is (e.g function scoped variables) and only we can see (and access it) and thus it is always in the state we’ve left it. However, some data is shared or . It still resides in our process memory, but we use special ways to access it and it might get out of sync. In some cases to work with it we copy it into our local memory to avoid data races - that’s why == ==is popular in Rust. local remote .clone() When we continue this line of reasoning, other techniques like thread-local storage come in naturally. We’ve just acquired a new gadget in our programming toolbelt, expanding what we can achieve by building software. However, there’s an invariant we still can rely on. When we reach out for shared (remote) data from a thread, we always get it. There is no situation when some memory chunk is not available. The OS will terminate (threads) by killing the process if the backing physical memory region malfunctions. The same applies to “our” thread if we locked a mutex, there is no way we might lose the lock and must stop what we’re doing immediately. We can rely on this invariant (enforced by the OS and modern hardware) that all participants are either dead or alive. All : if process (OOM), OS (kernel bug) or hardware encounters a problem - all our threads will cease to exist together without external leftover side-effects. all participants share the fate Inventing a process One important thing to note. How did we make this first step by introducing threads? We separated, forked. Instead of having one scheduling unit, we introduced multiple. Let’s keep applying this unsharing approach and see how it goes. This time we copy process virtual memory. That is called - a . We can run another instance of our program or start other existing utility. This is a great approach to: spawning process reuse other code with strict boundaries run untrusted code, isolating it from our own memory Almost all == ==work this way, so they are able to run untrusted Javascript executable code downloaded from Internet and terminate it reliably when you close a tab without bringing down the whole application. modern browsers This is yet another regime of execution we discovered by giving up the invariant and virtual memory and making a copy. Copies are not free: shared fate unsharing OS needs to manage memory related data structures (to maintain virtual -> physical mapping) Some bits could have been shared and thus processes consume additional memory Breaking free Why stop here? Let’s explore what else can we copy and our program among. But why go distributed in the first place? In many cases tasks at hand can be solved using a single machine. distribute We need to go distributed tenets so that our software stops depending on the inevitable problems that underlying layers encounter. to escape shared fate To name a few: OS upgrades: from time to time we need to reboot our machines Hardware failures: they happen more often than we’d like to External failures: power and network outages are a thing. If we copy an OS - we call that a and can run customers’ programs on a physical one and build a huge cloud business on it. If we take two or more computers and run our programs on each - our program can outlive even a hardware failure, providing 24/7 service and gain a competitive advantage. Large corporations long ago went even further and now Internet giants run copies in different datacenters and even continents, thus making a program resilient to a typhoon or a simple power outage. virtual machine But this independence comes at a price: the old invariants are not enforced, we’re on our own. No worries, we are not the first ones. There are a lot of techniques, tools and services to help us out. Takeaways We’ve just gained an ability to reason about systems and their respective execution regimes. Inside every large scale out system most parts are familiar sequential and stateless, many components are multi-threaded with memory types and hierarchies all held together by a mix of some truly distributed parts: The goal is to be able to distinguish where we’re currently, what invariants hold and act (modify/design) accordingly. We highlighted the basic reasoning, transforming “unknown unknowns” into “known unknowns”. Don’t take it lightly, this is a significant progress.

SPINZO

spinzo.com

1-10 employees

Since 2011

COMPANY RANKING

SPINZO

EVERGREEN INDEX #10098

Representing this company? Click here to claim and customize this page!

Spinzo WIKI

Apply to get your Tech Company News Page live on HackerNoon.com today!

SPINZO

spinzo.com

1-10 employees

Since 2011

COMPANY RANKING

SPINZO

EVERGREEN INDEX #10098

RELATED COMPANIESRANK

STORIES ABOUT SPINZO

Representing this company? Click here to claim and customize this page!

Spinzo WIKI

Apply to get your Tech Company News Page live on HackerNoon.com today!