#AlphaZero

[zzedar2]

A decade ago, AlphaZero would have been an SCP. A computer that can study any board game and within a day play it with superhuman skill? Totally an SCP.

[dat-physics-gal]

How computers play Chess

I just watched this video and it got me thinking on chess AI.

For those unfamiliar, chess AI will evaluate possible future positions of the current board, then evaluate which of them are most desirable, and then it’ll play the moves to get it there.

If a future branch for example has a lot of good but also a couple really bad boards, it’ll go with a safer branch instead. It always plays the move that maximizes it’s future state. Once it finds a forced checkmate sequence, of course it will follow that branch until it’s opponent is defeated.

~~~

However, back when chess AI was in it’s infancy and just did this, humans could still beat it for a while. The positions in the future were evaluated on if there was a forced checkmate sequence as well as material points. Different chess pieces are worth different so called material points, and by evaluating future boards on material point advantage, the AI could take all it’s opponents options and eventually win that way. But humans still sometimes won. Why?

Well, they understood something that the AI didn’t, but they couldn’t formulate what exactly that was.

So the solution: Make an AI that emulates humans.

~~~

The second generation of chess AI then proceeded to use a combination of the brute force “look for a forced checkmate sequence and material advantage“ approach, but mixed it with emulating actual human played games too, emulating the winning side, of course.

This made the AI much better at openings and endgames. And the mid-game was already it’s strong suit, no human is better at tactics than an AI that can calculate into the future that far.

After that advancement, AI completely outpaced humans. No human alive can beat such an AI anymore, and those are used to evaluate positions of human games. The most prevalent and powerful of that type of AI is called StockFish, and it has gone through countless iterations, each one improving further and further away from the human skill-cap.

But a problem remains: The developers have no real idea what it is we humans understand better, that enabled us to beat the first generation of chess AI.

They knew it was some “positional advantage“ but had no way of quantifying that, no way of making the AI understand that. Except by letting it emulate humans, who do understand and utilize that concept.

~~~

And then this AI from the video i linked came on to the scene, AlphaZero is it’s name. And from how it plays, i think it’s creators have finally found what humans did differently that allowed them to best 1st generation AI. I believe it simply changed the way it evaluates/rates how good future boards are!

And the measure it rates them by is the amount of possible legal moves the AI will have at that future board vs. the amount of possible moves their opponent will have.

The extreme case is a forced checkmate sequence: The enemy has no moves except those that lead to the king being unable to move, as well as attacked.

A checkmate.

So first of all, if AlphaZero evaluates future positions like that, it doesn’t have to make an exception for forced checkmate sequences. Those just happen to be favored naturally. No if-case needed. And also no more evaluation based on material points.

This enables the AI to apply pressure on it’s opponent, because pressure means less possibilities for them, more for the AI.

Many things GothamChess said in the video point to this being the case:

it likes long bishop diagonals

it likes it’s own king to be mobile

it doesn’t care about sacrificing pawns, and will gladly do so

it closes down positions, but usually on the enemy side

The first two points are simple to understand if it evaluates future boards on how mobile it’s pieces are supposed to be.

The third also makes sense, as pawns are just in the way of it’s own piece moves. A pawn can only move one square, a piece in way more ways. So it’ll gladly sacrifice a pawn just to make it’s own pieces able to move more freely.

The fourth point can be understood as minimizing the enemy movement options.

This way of evaluation allows AlphaZero to incorporate things like trapping pieces and developing moves into it’s natural behavior. Developing moves were previously relegated to emulating human openings, but that was because the material point evaluation didn’t account for them. They don’t have to be handled like that anymore, possible move evaluation naturally leads to the chess AI playing developing moves.

So yeah, that’s the next generation of chess AI. The kind that evaluates boards not on material points, but on movement options.

[jcmarchi]

Who is Winning the AI Race in 2024? Big Tech’s Race to AGI

New Post has been published on https://thedigitalinsider.com/who-is-winning-the-ai-race-in-2024-big-techs-race-to-agi/

Who is Winning the AI Race in 2024? Big Tech’s Race to AGI

Artificial Intelligence (AI) has become the most fiscussed technological advancement of this decade. As we push the boundaries of what machines can do, the ultimate goal for many tech giants is to achieve Artificial General Intelligence (AGI) – a hypothetical form of AI that can understand, learn, and apply its intelligence to solve any problem, much like a human brain.

The race to AGI is not just a matter of technological supremacy; it’s a quest that could reshape the very fabric of our society. The potential applications of AGI are vast and transformative, ranging from solving complex global issues to revolutionizing industries across the board. This is why the world’s leading tech companies are investing billions of dollars and countless hours into AI research and development.

In this article, we’ll explore the efforts of key players in the AI race, including Google, NVIDIA, Microsoft, OpenAI, Meta, and others. We’ll delve into their strategies, achievements, and the unique approaches they’re taking to push the boundaries of AI technology.

Understanding AGI

What is AGI?

AGI, often described as the “holy grail” of artificial intelligence, is envisioned as a system capable of performing any intellectual task that a human can. However, defining AGI has proven to be as elusive as achieving it. Geoffrey Hinton, a pioneering figure in AI, notes that while AGI is a “serious, though ill-defined concept,” there is little consensus on what it precisely entails. Hinton prefers the term “superintelligence” to describe AGI systems that would surpass human cognitive abilities.

The Elusive Nature of AGI

Leading tech giants, including OpenAI, Google, Meta, Microsoft, and Amazon, are at the forefront of this race. Each company brings its unique strengths and strategic goals to the table. OpenAI, for example, is deeply committed to ensuring that AGI, once developed, benefits all of humanity. The organization has set up a governance structure where its board of directors will decide when their systems have achieved AGI, a milestone that will significantly impact their partnership with Microsoft.

Google

Google has long been at the forefront of AI research and development, with two main divisions spearheading its efforts: DeepMind and Google Brain.

A. DeepMind and its achievements

DeepMind, acquired by Google in 2014, has been responsible for some of the most groundbreaking achievements in AI. Their AlphaGo program famously defeated the world champion in the complex game of Go in 2016, a feat many thought was decades away. This was followed by AlphaZero, which achieved superhuman performance in chess, shogi, and Go through self-play reinforcement learning.

More recently, DeepMind has made significant strides in protein folding with AlphaFold. This AI system can predict protein structures with remarkable accuracy, potentially revolutionizing drug discovery and our understanding of diseases.

B. Google Brain and TensorFlow

Google Brain, the company’s in-house AI research team, has been instrumental in developing tools and frameworks that have accelerated AI research worldwide. TensorFlow, an open-source machine learning library developed by Google Brain, has become one of the most widely used tools for building AI models.

Google Brain has also made significant contributions to natural language processing with models like BERT (Bidirectional Encoder Representations from Transformers), which has improved Google’s search results and language understanding capabilities.

C. Recent developments and future plans

Google continues to push the boundaries of AI with projects like LaMDA (Language Model for Dialogue Applications), which aims to make conversational AI more natural and context-aware. The company has also been working on integrating AI more deeply into its products, from Google Search to Gmail to Google Photos.

In terms of hardware, Google has developed its own AI chips, called Tensor Processing Units (TPUs), optimized for machine learning workloads. These chips power many of Google’s AI services and are also available to customers through Google Cloud.

Looking ahead, Google’s AI strategy seems focused on developing more general and versatile AI systems that can handle a wide range of tasks, inching closer to the concept of AGI. The company is also heavily invested in quantum computing research.

NVIDIA’s Role in the AI Ecosystem

While NVIDIA may not be a household name like Google or Microsoft, it plays a crucial role in the AI ecosystem as the leading provider of hardware that powers AI computations.

A. GPU dominance in AI hardware

NVIDIA’s Graphics Processing Units (GPUs) have become the de facto standard for training and running AI models. Originally designed for rendering graphics in video games, GPUs turned out to be exceptionally well-suited for the parallel processing required in AI computations.

NVIDIA’s data center revenue, largely driven by AI-related sales, has been growing rapidly. In 2022, the company introduced its H100 GPU, based on the new Hopper architecture, which promises significant performance improvements for AI workloads.

B. NVIDIA’s AI software stack

Beyond hardware, NVIDIA has developed a comprehensive software stack for AI development. This includes CUDA, a parallel computing platform and programming model that allows developers to harness the power of NVIDIA GPUs for general-purpose processing.

NVIDIA also offers tools like cuDNN (CUDA Deep Neural Network library) and TensorRT, which optimize deep learning performance on NVIDIA GPUs. These tools are widely used in the AI community and have contributed to NVIDIA’s dominant position in the AI hardware market.

C. Partnerships and collaborations

NVIDIA has formed strategic partnerships with many leading tech companies and research institutions. For instance, it works closely with autonomous vehicle manufacturers to provide AI-powered solutions for self-driving cars. The company has also collaborated with healthcare institutions to apply AI in medical imaging and drug discovery.

In 2022, NVIDIA announced a partnership with Booz Allen Hamilton to develop AI-enabled cybersecurity solutions for the U.S. government and critical infrastructure. This highlights the growing importance of AI in national security and defense applications.

Microsoft’s AI Strategy

Microsoft has strategically positioned itself as a leader in AI by leveraging partnerships and investing in key AI startups. The company’s $13 billion investment in OpenAI has provided it with exclusive access to OpenAI’s models, which have been integrated into Microsoft products like GitHub Copilot and the Azure AI platform.

A. Azure AI and cloud services

Microsoft’s cloud platform, Azure, offers a wide range of AI services that allow businesses to incorporate AI into their applications. These services cover areas such as machine learning, computer vision, natural language processing, and speech recognition.

Azure Machine Learning, a cloud-based environment for training, deploying, and managing machine learning models, has become a popular choice for enterprises looking to implement AI solutions. Microsoft’s strategy of providing easy-to-use AI tools has helped democratize AI development and accelerate its adoption across various industries.

B. AI integration across Microsoft products

Microsoft has been steadily integrating AI capabilities across its product lineup. In Microsoft 365 (formerly Office), AI powers features like smart compose in Outlook, automatic slide design in PowerPoint, and data analysis in Excel.

Windows 11 has seen increased AI integration with features like Windows Studio Effects, which uses AI for background blur, eye contact, and automatic framing in video calls. The company has also introduced AI-powered features in its Edge browser and Bing search engine, leveraging large language models to provide more interactive and informative search experiences.

OpenAI’s Rapid Progress

OpenAI remains a central figure in the AI landscape, particularly with its mission to develop AGI. The company has been a pioneer in creating some of the most advanced language models, including GPT-4 and the upcoming GPT-5. OpenAI’s models are not only leading in terms of technical capability but also in commercial integration, thanks to its deep partnership with Microsoft.

OpenAI’s AGI ambitions are well-documented, with CEO Sam Altman stating that achieving AGI would represent “the most powerful technology humanity has yet invented.” The company’s approach to AI development balances cutting-edge innovation with a strong emphasis on ethical considerations and societal impact. However, the high costs associated with training large models have necessitated significant external funding, including talks with investors like the U.A.E. government to secure up to $7 trillion for future AI chip manufacturing projects​

A. GPT series and its impact

OpenAI’s most notable achievement has been the development of the GPT (Generative Pre-trained Transformer) series of language models. GPT-3, released in 2020, was a game-changer in the field of natural language processing, demonstrating an unprecedented ability to generate human-like text.

The release of GPT-4 in 2023 further pushed the boundaries of what’s possible with language models. GPT-4 demonstrated improved reasoning capabilities, reduced hallucinations, and the ability to handle multimodal inputs (text and images). These models have found applications in various fields, from content creation to code generation to automated customer service.

B. DALL-E and multimodal AI

In addition to text generation, OpenAI has made significant strides in image generation with DALL-E. This AI system can create unique images from text descriptions, showcasing the potential of AI in creative fields. The latest iteration, DALL-E 3, improved the quality and accuracy of generated images, while also introducing features like inpainting and outpainting.

These developments in multimodal AI – systems that can work with different types of data like text and images – represent a significant step towards more general AI systems.

Meta’s AI Initiatives

Meta, under the leadership of CEO Mark Zuckerberg, has shifted its focus towards developing Artificial General Intelligence (AGI).  Meta’s strategy involves building AGI systems that can perform a wide range of complex tasks as well as, or better than, humans. This ambitious goal reflects Meta’s broader vision of integrating advanced AI across its vast ecosystem of apps and services.

To support this effort, Meta is heavily investing in computational power, with plans to amass over 340,000 of Nvidia’s H100 GPUs by the end of 2024. This immense computational capability is crucial for training large-scale AI models like LLaMA 3, which is was recently launched.

A. PyTorch and open-source contributions

One of Meta’s most significant contributions to the AI community has been PyTorch, an open-source machine learning library. PyTorch has gained widespread adoption in the research community due to its flexibility and ease of use, particularly for deep learning applications.

Meta AI, the company’s AI research division, regularly publishes its research and releases open-source tools, contributing to the broader AI ecosystem. This open approach has helped Meta attract top AI talent and stay at the forefront of AI research.

B. AI in social media and the metaverse

Meta leverages AI extensively across its social media platforms (Facebook, Instagram, WhatsApp) for content recommendation, ad targeting, and content moderation. The company’s recommendation algorithms process vast amounts of data to personalize user experiences.

C. Recent breakthroughs and challenges

In 2024, Meta announced several AI breakthroughs, including Segment Anything Model (SAM), a new AI model for image segmentation that can identify and outline objects in images and videos with remarkable accuracy. They also introduced series of ne of the most popular open source LLM called LLaMA (Large Language Model Meta AI).

However, Meta has faced challenges, particularly in content moderation. The company has struggled to effectively use AI to combat misinformation and hate speech on its platforms, highlighting the complexities of applying AI to real-world social issues.

Other Notable Players

IBM continues to be a major player in AI with its watsonx platform, which has evolved significantly since its inception. IBM’s focus has shifted towards making AI more open, accessible, and scalable for enterprises. The watsonx platform now includes a suite of AI-powered automation tools and governance capabilities that enable businesses to integrate and manage AI solutions more effectively across various domains like IT operations, cybersecurity, and customer service.

Recently, IBM introduced generative AI capabilities to enhance its managed Threat Detection and Response Services. This includes a new AI-powered Cybersecurity Assistant designed to streamline and accelerate the investigation and response to security threats, further leveraging IBM’s broader AI capabilities built on the watsonx platform​ (IBM Newsroom)​ (IBM Newsroom).

IBM is also fostering strategic partnerships with companies like AWS, Adobe, Meta, and Salesforce to integrate its AI solutions into broader ecosystems, ensuring that its AI technologies are both versatile and widely adopted across industries​ (IBM TechXchange Community)​ (IBM – United States).

B. Amazon’s AI Services

Amazon remains a dominant force in AI through its Amazon Web Services (AWS) platform, which provides a comprehensive suite of AI and machine learning tools. AWS’s Amazon SageMaker is a key offering, enabling developers to build, train, and deploy machine learning models at scale.

In addition to enterprise AI services, Amazon continues to innovate in consumer AI products with Alexa, its virtual assistant, which utilizes advanced natural language processing and machine learning to interact with users. The company’s focus on integrating AI seamlessly into its e-commerce and cloud services has positioned it as a leader in the AI space.

C. Apple’s On-Device AI Approach

Apple’s unique approach to AI emphasizes on-device processing to prioritize user privacy. This is exemplified by features like Face ID and the broader use of machine learning models through its Core ML framework. Apple’s custom silicon, including the A-series and M-series chips, includes dedicated neural engines that power AI tasks efficiently on devices.

The company has also enhanced its AI offerings with improvements in natural language processing through Siri and advancements in computer vision with features like Live Text.

What’s Next? The Path to AGI

[lifetechweb]

IA na Olimpíada Internacional de Matemática: como AlphaProof e AlphaGeometry 2 alcançaram o padrão de medalha de prata

O raciocínio matemático é um aspecto vital das habilidades cognitivas humanas, impulsionando o progresso em descobertas científicas e desenvolvimentos tecnológicos. À medida que nos esforçamos para desenvolver inteligência artificial geral que corresponda à cognição humana, equipar a IA com capacidades avançadas de raciocínio matemático é essencial. Embora os sistemas de IA atuais possam lidar…

[govindhtech]

AlphaProof: Google AI Systems To Think Like Mathematicians

AlphaProof and AlphaGeometry 2

Google AI systems advance towards thinking by making strides in maths. One question was answered in minutes, according to a blog post by Google, but other questions took up to three days to answer longer than the competition’s time limit. Nevertheless, the scores are among the highest achieved by an Al system in the competition thus far.

Google, a division of Alphabet, showcased two artificial intelligence systems that showed improvements in generative Al development the ability to solve challenging mathematical problems.

The current breed of AI models has had difficulty with abstract arithmetic since it demands more reasoning power akin to human intellect. These models operate by statistically anticipating the following word.

The company’s Al division, DeepMind, released data demonstrating that its recently developed Al models, namely AlphaProof and AlphaGeometry 2, answered four of every six questions in the 2024 International Math Olympiad, a well-known tournament for high school students.

One question was answered in minutes, according to a blog post by Google, but other questions took up to three days to answer longer than the competition’s time limit. Nevertheless, the scores are among the highest achieved by an Al system in the competition thus far.

AlphaZero

The business said that AlphaZero, another Al system that has previously defeated humans at board games like chess and go, and a version of Gemini, the language model underlying its chatbot of the same name, were combined to produce AlphaProof, a reasoning-focused system. Only five out of the more than 600 human competitors were able to answer the most challenging question, which was one of the three questions that AlphaProof answered correctly.

AlphaGeometry 2

AlphaGeometry 2 solved another math puzzle. It was previously reported in July that OpenAI, supported by Microsoft, was working on reasoning technology under the code name “Strawberry.” As Reuters first revealed, the project, originally known as Q, was regarded as such a breakthrough that several staff researchers warned OpenAI’s board of directors in a letter they wrote in November, stating that it could endanger humankind.

The top choice for document editing and proofreading is AlphaProof. The demand for accurate and efficient services is growing in the digital age. It stands out as a leading option, offering excellent services to guarantee your documents are flawless. In order to show why AlphaProof is unique in the industry, this article explores its features, advantages, and user experiences.

How does AlphaProof work?

AlphaProof a feature-rich online tool, handles all editing and proofreading needs. It offers specialized services to increase the quality and readability of your documents for professionals, students, and company owners. AlphaProof publishes technical documentation, corporate reports, creative writing, and academic essays.

Essential Elements of AlphaProof

Expert Proofreading

To fix typographical, punctuation, and grammar flaws in your documents, AlphaProof has a team of highly skilled proofreaders who carefully go over them. This guarantees that your text looks professional and is free of common mistakes.

Complex Editing

It provides sophisticated editing services in addition to basic proofreading. This entails streamlining the sentence structure, boosting readability overall, and strengthening coherence and flow. Better word selections and stylistic enhancements are also suggested by the editors.

Editors with specific expertise

AlphaProof recognizes that varying documents call for varying levels of competence. It boasts a diverse team of editors with skills in technical writing, business communication, academic writing, and creative writing. This guarantees that an individual possessing pertinent expertise and experience will evaluate your material.

Quick Resolution

Quick turnaround times are provided by AlphaProof to help you meet deadlines. You can choose 24-hour express service to ensure your document is available when you need it.

Easy-to-use interface

The AlphaProof platform boasts an intuitive interface that facilitates the uploading of documents, selection of services, and tracking of order status. From beginning to end, the procedure is simplified to offer a hassle-free experience.

Secrecy and Protection

The security and privacy of your papers are very important to it. The platform uses cutting-edge encryption technology to safeguard your data, and every file is handled with the highest care.

The Advantages of AlphaProof Use

Better Document Quality

The quality of your documents can be greatly improved by utilising it’s services. This can result in more professionalism in corporate communication, higher grades, and a more positive impression on your readers.

Reduce Effort and Time

Editing and proofreading can be laborious processes. With AlphaProof, you can focus on your primary responsibilities while professionals optimize your papers, saving you time and effort.

Customized Offerings

To address the unique requirements of various document formats, It offers customized services. AlphaProof may provide you with comprehensive editing for a research paper or expeditious proofreading for an email.

Knowledgeable Perspectives

The editor’s comments and recommendations on it can give you important information about your writing style and areas that need work. With time, this can assist you in improving as a writer.

A Boost in Self-Assurance

You may feel more confident in the calibre of your work if you know it has been expertly edited and proofread. For high-stakes papers like published articles, commercial proposals, and theses from academic institutions, this is especially crucial.

Customer Experiences

Scholars and Students

AlphaProof has proven to be a useful resource for numerous academics and students. A postgraduate student said, “AlphaProof enabled me to refine my thesis to the ideal level.” The final draft was error-free, and the editors’ suggestions were wise.”

Composers and Novelists

The specialized editing services provided by AlphaProof are valued by authors and creative writers. A budding writer said, “it’s editors understood my voice and style, providing feedback that improved my manuscript without altering my unique voice.”

In conclusion

With a variety of features and advantages to meet a wide range of demands, AlphaProof stands out as a top option for document editing and proofreading. It guarantees that your documents are flawless, saving you time and improving the calibre of your work. It does this through its skilled staff, quick return times, and intuitive interface.

Read more on govindhtech.com

[quotesfrommyreading]

In the first era, humans would devise attack strategies, then refine them in games against machines. AlphaZero crushed these earlier engines by “playing extremely aggressive chess,” Sadler said. The modern, neural-net engines are eager to sacrifice; and they exhibit a strong grasp of openings, positional structure, and long-term strategy. “It started to look a bit more [like] a human way to play,” Sutovsky told me, in describing this transformation. Or even superhuman, he said: The new chess engines seemed to have insight into “the tactical skirmish, but also could plan for some long-lasting compensation for material loss.”

To understand just how superior machines have become, consider chess’s “Elo” rating system, which compares players’ relative strength and was devised by a Hungarian American physicist. The highest-ever human rating, achieved by Carlsen twice over the past decade, was 2882. DeepBlue’s Elo rating was 2853. A chess engine called Rybka was the first to reach 3000 points, in 2007; and today’s most powerful program, Stockfish, currently has more than 3500 Elo points by conservative estimates. That means Stockfish has about a 98 percent probability of beating Carlsen in a match and, per one estimate, a 2 percent chance of drawing. (An outright victory for Carlsen would be almost impossible.)

Where chess engines once evaluated human strategies, the new, upgraded versions—which are freely available online, including Stockfish—now generate surprising ideas and define the ideal way to play the game, to the point that human performance is measured in terms of “centipawn” (hundredths of a pawn) loss relative to what a computer would play. While training, a player might ask the software to suggest a set of moves to fit a given situation, and then decide to use the computer’s sixth-ranked option, rather than the first, in the hopes of confusing a human competitor who trained with similar algorithms. Or they might choose a move tailored to the weaknesses of a particular opponent. Many chess experts have adopted the new engines’ more aggressive style, and the algorithms have popularized numerous tactics that human players had previously underestimated.

 —   Chess Is Just Poker Now

[cthulhubert]

I'm not an AI doomer but I am AI cautious, and I think the future holds something more general purpose than the generators we have now.

But I also think people are off base about the danger. Both in aims (a "rogue" AI seems unlikely; one told to do evil by its owners, however?) but more importantly, in methods.

I'm not concerned about ultra-tech or super manipulators; I think the issue is in a capability that humans already naturally dismiss: cooperation, coordination, administration; and how those scale.

An AI won't be dangerous because it invents fusion powered lasers and gray goo, it'll be dangerous because it can do the work of a nation state, but directed by a single will.

(below the cut, some elaboration)

To be clear, I don't actually dismiss, out of hand, the potential of an AI to develop physical tools and processes faster than humans could, and implement them better.

Nor the idea that it could be as much better than a human salesman or spinmeister as AlphaZero is at chess than any human chess master. (I think some people underestimate this because the danger of a good manipulator is that they don't make you feel manipulated. People don't want to acknowledge their own psychosocial limitations. I've seen people say about mass targeted harassment campaigns, "Well, I would just ignore it," because they've never actually been tested that way.)

Both of these are easily memeable and more easily dismissed: "Maybe it can be smart but it can't be magic!"

But I don't think that's the most likely weapon to be wielded by a machine intelligence (or "general purpose goal satisfying applied statistics system" if "intelligence" is too loaded for you).

People dismiss conspiracy theorists because they (correctly) realize the goal and methods those theorists describe are, uh, fucking stupid. But more rarely people point at the fact that the level of coordination and cooperation to hide the moon landing or the shape of the Earth is just impossible.

I think that people may intellectually understand that every single one of the 8 billion human beings on this planet is a real whole actual person with a life and interiority; but they don't grok it on an intuitive level. I think this is true even of people that don't believe in the Illuminati.

So they might intellectually know that a vast machine intelligence could have the equivalent intellectual goal-satisfying power of a nation, and that every iota of that power is moving in perfectly coordinated lockstep, directed by one purpose. But it doesn't scare them because on one emotional level, they already think of nations as working like that. And so even if pointed out, they imagine that vastness being just as ineffectual and inefficient as large corporations and countries.

Just think about the "personal FBI agent" memes. Of course those are tongue in cheek, but I think there's something real underlying that. People imagine themselves as already heavily surveilled and manipulated, but it just doesn't do enough to them. We can't truly imagine what it'd be like to have an entire human's amount of awareness tracking our every step for the sole purpose of using us for some goal.

I'm just always thinking about somebody who has seen a tea kettle moving a pinwheel and goes, "I don't see what's so scary, powerful, or useful about steam. This 'industrial revolution' idea is a pipe dream."

[lastscenecom]

合成トレーニングデータが目覚ましい成功を収めた例もある。2016年に囲碁の世界チャンピオンに勝利したAlphaGoや、その後継機であるAlphaGo ZeroとAlphaZeroなどだ。これらのシステムは、自分自身と対戦することで学習した。後者2つは、トレーニングデータとして人間のゲームを一切使用していない。大量の計算を使用してある程度高品質のゲームを生成し、そのゲームを使用してニューラルネットワークをトレーニングし、計算と組み合わせることでさらに高品質のゲームを生成できるようになり、反復的な改善ループが生まれた。  セルフプレイは「システム 2 --> システム 1 蒸留」の典型的な例です。これは、遅くてコストのかかる「システム 2」プロセスがトレーニング データを作成し、高速でコストのかからない「システム 1」モデルをトレーニングするものです。これは、囲碁のように完全に自己完結的な環境であるゲームに適しています。セルフプレイをゲーム以外の領域に適応させることは、価値のある研究方向です。コード生成など、この戦略が役立つ重要な領域もあります。しかし、言語翻訳などのよりオープンエンドなタスクでは、無限のセルフ改善を期待することはできません。セルフプレイによって大幅な改善が認められる領域は、例外であり、一般的ではないと予想する必要があります。

AI スケーリングの神話 - アルヴィンド・ナラヤナンとサヤシュ・カプール著

[dreamsy990]

NAME A CHARACTER WITH A HIGHER CHESS IQ THEN ALPHAZERO. i bet you can't ehhehehhaehaehaehh

whoever came up with twilight princess low percent

[gerdfeed]

By Chollet’s definition, programs like AlphaZero are highly skilled, but they aren’t particularly intelligent, because they aren’t efficient at gaining new skills

Why A.I. Isn’t Going to Make Art | The New Yorker

[mvishnukumar]

What are the most important breakthroughs in machine learning research today?

Machine learning research is advancing rapidly, and several recent breakthroughs are reshaping the field. 

There is Some of the most important breakthroughs include:

Transformers and Attention Mechanisms: 

The introduction of transformer models and attention mechanisms, exemplified by architectures like BERT and GPT, has revolutionized natural language processing (NLP). These models excel at understanding and generating human language, leading to significant improvements in tasks such as text generation, translation, and sentiment analysis.

Self-Supervised Learning: 

Self-supervised learning methods have gained prominence for training models without relying heavily on labeled data. Techniques like contrastive learning and masked language modeling enable models to learn representations from large amounts of unlabeled data, improving performance on various tasks with limited labeled samples.

Large Language Models (LLMs): 

The development of large language models, such as GPT-4 and ChatGPT, has set new benchmarks for NLP. These models leverage vast amounts of data and computational power to generate coherent, contextually relevant text and perform a wide range of language-based tasks with high accuracy.

Few-Shot and Zero-Shot Learning: 

Few-shot and zero-shot learning techniques enable models to perform tasks with very few or no examples. These methods are particularly useful in scenarios where labeled data is scarce, allowing models to generalize and make predictions based on limited information.

Reinforcement Learning (RL) Advances: 

Recent advances in reinforcement learning, including algorithms like AlphaZero and OpenAI’s Dota 2 agents, have demonstrated significant progress in training agents to perform complex tasks through trial and error. These breakthroughs have applications in robotics, game playing, and autonomous systems.

Generative Models: 

Generative models, such as Generative Adversarial Networks (GANs) and Variational Autoencoders (VAEs), have made strides in creating realistic synthetic data. These models are used for generating images, videos, and other data types, and have applications in content creation, data augmentation, and simulation.

Ethics and Fairness in AI: 

Research into the ethical implications and fairness of AI systems is increasingly important. Efforts to address biases, ensure transparency, and develop ethical guidelines for AI deployment are crucial for creating responsible and equitable machine learning applications.

Explainable AI (XAI): 

The development of explainable AI techniques aims to make machine learning models more transparent and interpretable. Methods for visualizing and understanding model decisions help build trust and ensure that AI systems are used responsibly.

These breakthroughs represent significant advancements in machine learning, driving innovation and expanding the potential applications of AI technologies.

[tumnikkeimatome]

Google DeepMindが開発した2つのAIシステム『AlphaProof』『AlphaGeometry 2』が国際数学オリンピック(IMO)で銀メダルレベルの成績を記録

AIシステムの概要と成果 Google DeepMindが開発した2つのAIシステム、AlphaProofとAlphaGeometry 2が2024年の国際数学オリンピック(IMO)で銀メダルレベルの成績を収めました。 両システムは6問中4問を解決し、42点満点中28点を獲得しました。金メダルまであと1点という僅差の結果となりました。 AlphaProofは代数と数論の問題を解くために開発されたシステムで、事前学習済み言語モデルとAlphaZeroの強化学習アルゴリズムを組み合わせています。 一方、AlphaGeometry 2は幾何学の問題に特化したシステムで、1億件の合成データセットで訓練されました。 IMOでの具体的な成績 IMOでは、AlphaProofが2つの代数問題と1つの数論問題を解決し、AlphaGeometry…

[jcmarchi]

AI at the International Mathematical Olympiad: How AlphaProof and AlphaGeometry 2 Achieved Silver-Medal Standard

New Post has been published on https://thedigitalinsider.com/ai-at-the-international-mathematical-olympiad-how-alphaproof-and-alphageometry-2-achieved-silver-medal-standard/

AI at the International Mathematical Olympiad: How AlphaProof and AlphaGeometry 2 Achieved Silver-Medal Standard

Mathematical reasoning is a vital aspect of human cognitive abilities, driving progress in scientific discoveries and technological developments. As we strive to develop artificial general intelligence that matches human cognition, equipping AI with advanced mathematical reasoning capabilities is essential. While current AI systems can handle basic math problems, they struggle with the complex reasoning needed for advanced mathematical disciplines like algebra and geometry. However, this might be changing, as Google DeepMind has made significant strides in advancing an AI system’s mathematical reasoning capabilities. This breakthrough is made at the International Mathematical Olympiad (IMO) 2024. Established in 1959, the IMO is the oldest and most prestigious mathematics competition, challenging high school students worldwide with problems in algebra, combinatorics, geometry, and number theory. Each year, teams of young mathematicians compete to solve six very challenging problems. This year, Google DeepMind introduced two AI systems: AlphaProof, which focuses on formal mathematical reasoning, and AlphaGeometry 2, which specializes in solving geometric problems. These AI systems managed to solve four out of six problems, performing at the level of a silver medalist. In this article, we will explore how these systems work to solve mathematical problems.

AlphaProof: Combining AI and Formal Language for Mathematical Theorem Proving

AlphaProof is an AI system designed to prove mathematical statements using the formal language Lean. It integrates Gemini, a pre-trained language model, with AlphaZero, a reinforcement learning algorithm renowned for mastering chess, shogi, and Go.

The Gemini model translates natural language problem statements into formal ones, creating a library of problems with varying difficulty levels. This serves two purposes: converting imprecise natural language into precise formal language for verifying mathematical proofs and using predictive abilities of Gemini to generate a list of possible solutions with formal language precision.

When AlphaProof encounters a problem, it generates potential solutions and searches for proof steps in Lean to verify or disprove them. This is essentially a neuro-symbolic approach, where the neural network, Gemini, translates natural language instructions into the symbolic formal language Lean to prove or disprove the statement. Similar to AlphaZero’s self-play mechanism, where the system learns by playing games against itself, AlphaProof trains itself by attempting to prove mathematical statements. Each proof attempt refines AlphaProof’s language model, with successful proofs reinforcing the model’s capability to tackle more challenging problems.

For the International Mathematical Olympiad (IMO), AlphaProof was trained by proving or disproving millions of problems covering different difficulty levels and mathematical topics. This training continued during the competition, where AlphaProof refined its solutions until it found complete answers to the problems.

AlphaGeometry 2: Integrating LLMs and Symbolic AI for Solving Geometry Problems

AlphaGeometry 2 is the latest iteration of the AlphaGeometry series, designed to tackle geometric problems with enhanced precision and efficiency. Building on the foundation of its predecessor, AlphaGeometry 2 employs a neuro-symbolic approach that merges neural large language models (LLMs) with symbolic AI. This integration combines rule-based logic with the predictive ability of neural networks to identify auxiliary points, essential for solving geometry problems. The LLM in AlphaGeometry predicts new geometric constructs, while the symbolic AI applies formal logic to generate proofs.

When faced with a geometric problem, AlphaGeometry’s LLM evaluates numerous possibilities, predicting constructs crucial for problem-solving. These predictions serve as valuable clues, guiding the symbolic engine toward accurate deductions and advancing closer to a solution. This innovative approach enables AlphaGeometry to address complex geometric challenges that extend beyond conventional scenarios.

One key enhancement in AlphaGeometry 2 is the integration of the Gemini LLM. This model is trained from scratch on significantly more synthetic data than its predecessor. This extensive training equips it to handle more difficult geometry problems, including those involving object movements and equations of angles, ratios, or distances. Additionally, AlphaGeometry 2 features a symbolic engine that operates two orders of magnitude faster, enabling it to explore alternative solutions with unprecedented speed. These advancements make AlphaGeometry 2 a powerful tool for solving intricate geometric problems, setting a new standard in the field.

AlphaProof and AlphaGeometry 2 at IMO

This year at the International Mathematical Olympiad (IMO), participants were tested with six diverse problems: two in algebra, one in number theory, one in geometry, and two in combinatorics. Google researchers translated these problems into formal mathematical language for AlphaProof and AlphaGeometry 2. AlphaProof tackled two algebra problems and one number theory problem, including the most difficult problem of the competition, solved by only five human contestants this year. Meanwhile, AlphaGeometry 2 successfully solved the geometry problem, though it did not crack the two combinatorics challenges

Each problem at the IMO is worth seven points, adding up to a maximum of 42. AlphaProof and AlphaGeometry 2 earned 28 points, achieving perfect scores on the problems they solved. This placed them at the high end of the silver-medal category. The gold-medal threshold this year was 29 points, reached by 58 of the 609 contestants.

Next Leap: Natural Language for Math Challenges

AlphaProof and AlphaGeometry 2 have showcased impressive advancements in AI’s mathematical problem-solving abilities. However, these systems still rely on human experts to translate mathematical problems into formal language for processing. Additionally, it is unclear how these specialized mathematical skills might be incorporated into other AI systems, such as for exploring hypotheses, testing innovative solutions to longstanding problems, and efficiently managing time-consuming aspects of proofs.

To overcome these limitations, Google researchers are developing a natural language reasoning system based on Gemini and their latest research. This new system aims to advance problem-solving capabilities without requiring formal language translation and is designed to integrate smoothly with other AI systems.

The Bottom Line

The performance of AlphaProof and AlphaGeometry 2 at the International Mathematical Olympiad is a notable leap forward in AI’s capability to tackle complex mathematical reasoning. Both systems demonstrated silver-medal-level performance by solving four out of six challenging problems, demonstrating significant advancements in formal proof and geometric problem-solving. Despite their achievements, these AI systems still depend on human input for translating problems into formal language and face challenges of integration with other AI systems. Future research aims to enhance these systems further, potentially integrating natural language reasoning to extend their capabilities across a broader range of mathematical challenges.

[learning-robotics]

Best Robotics Papers in 202

What Are the Best Robotics Papers?

The field of robotics is rapidly evolving, with groundbreaking research and innovative developments happening at an unprecedented pace. For those deeply entrenched in this field or simply curious about the latest advancements, understanding the most influential and highly-regarded robotics papers is crucial. This article delves into some of the best robotics papers that have significantly contributed to the field, highlighting their key findings, methodologies, and impacts.

Introduction to Robotics Research

Robotics research encompasses a wide array of topics, from artificial intelligence and machine learning to mechanical design and human-robot interaction. Each of these areas contributes to the overall advancement of robotics, making it a multidisciplinary field that requires a comprehensive understanding of various scientific principles and technologies.

Key Areas of Robotics Research

Artificial Intelligence and Machine Learning

AI and machine learning are at the heart of modern robotics, enabling robots to perform complex tasks, learn from their environment, and adapt to new situations. Some of the most influential papers in this area include:

"Mastering Chess and Shogi by Self-Play with a General Reinforcement Learning Algorithm" by Silver et al.

Summary: This paper introduces AlphaZero, an AI system that uses reinforcement learning to master chess and shogi without prior knowledge of the games.

Impact: Demonstrates the power of reinforcement learning in developing AI that can learn and outperform humans in complex tasks.

"DQN: Playing Atari with Deep Reinforcement Learning" by Mnih et al.

Summary: The paper presents a deep Q-network (DQN) that combines reinforcement learning with deep neural networks to play Atari games at a superhuman level.

Impact: Showcases the potential of deep learning in developing AI agents capable of complex decision-making processes.

Mechanical Design and Control

Mechanical design and control are fundamental to the development of efficient and functional robots. Notable papers in this domain include:

"Passive Dynamic Walking" by McGeer

Summary: This pioneering work introduces the concept of passive dynamic walking, where robots use gravity and inertia to achieve efficient, human-like gait patterns without active control.

Impact: Revolutionizes the approach to robotic locomotion, emphasizing energy efficiency and simplicity.

"BigDog, the Rough-Terrain Quadruped Robot" by Raibert et al.

Summary: Describes the development of BigDog, a quadruped robot capable of navigating rough terrain using advanced control algorithms and mechanical design.

Impact: Advances the field of legged robotics, showcasing the potential for robots to operate in challenging environments.

Human-Robot Interaction

Human-robot interaction (HRI) is a critical area of research, focusing on how robots and humans can work together effectively. Key papers in this field include:

Breakthrough Robotics Papers

"Planning Algorithms" by LaValle

Summary: This comprehensive book covers a wide range of planning algorithms essential for robotics, including motion planning, discrete planning, and planning under uncertainty.

Impact: Serves as a foundational reference for researchers and practitioners in the field of robotics planning.

"Probabilistic Robotics" by Thrun, Burgard, and Fox

Summary: Introduces probabilistic methods for robot perception, localization, and mapping, emphasizing the importance of uncertainty in robotic systems.

Impact: Establishes a new paradigm in robotics, where probabilistic approaches are integral to developing robust and reliable robots.

"The DARPA Robotics Challenge Finals: Humanoid Robots To The Rescue" by Pratt et al.

Summary: Details the DARPA Robotics Challenge, a competition aimed at developing humanoid robots capable of performing complex tasks in disaster response scenarios.

Impact: Highlights the advancements and challenges in creating humanoid robots that can operate in real-world disaster situations.

Emerging Trends in Robotics Research

Swarm Robotics

Swarm robotics involves the coordination of multiple robots to achieve collective behavior. Key papers include:

"Swarm Intelligence: From Natural to Artificial Systems" by Bonabeau, Dorigo, and Theraulaz

Summary: Explores the principles of swarm intelligence and their application to robotics, drawing inspiration from natural systems like ant colonies and bird flocks.

Impact: Provides a comprehensive framework for understanding and developing swarm robotics systems.

"Kilobot: A Low-Cost Scalable Robot System for Demonstrating Collective Behaviors" by Rubenstein et al.

Summary: Introduces Kilobot, a low-cost, scalable robotic system designed to study collective behaviors in large robot swarms.

Impact: Demonstrates the feasibility of large-scale swarm robotics and its potential applications.

Soft Robotics

Soft robotics focuses on creating robots with flexible, deformable bodies that can adapt to their environment. Influential papers include:

"Soft Robotics: A Bioinspired Evolution in Robotics" by Laschi and Cianchetti

Summary: Discusses the principles and applications of soft robotics, inspired by biological systems like octopuses and worms.

Impact: Highlights the potential of soft robots in areas where traditional rigid robots are limited.

"Soft Robots for Chemists" by Whitesides

Summary: Explores the interdisciplinary nature of soft robotics, particularly its applications in chemistry and biomedical engineering.

Impact: Bridges the gap between robotics and other scientific disciplines, fostering innovation and collaboration.

Conclusion

The field of robotics is a dynamic and rapidly evolving area of research, driven by groundbreaking papers that push the boundaries of what is possible. From AI and machine learning to mechanical design, human-robot interaction, and emerging trends like swarm and soft robotics, these papers have laid the foundation for the future of robotics. By understanding and building upon these seminal works, researchers and practitioners can continue to advance the field, creating robots that are more intelligent, capable, and adaptable than ever before.

If you want to read more information about how to boost traffic on your Website just visit PRAGROBOTICS.

[editoriadors-blog]

Em 2018, uma equipe de cientistas do Google AI criou um sistema de inteligência artificial que poderia aprender a jogar jogos de tabuleiro complexos como xadrez o e-Go em um nível muito além do dos jogadores humanos. O sistema, chamado AlphaZero, foi capaz de aprender as regras e estratégias dos jogos por meio do autojogo e em poucas horas se tornou o melhor jogador do mundo, derrotando os programas campeões mundiais anteriores.

No geral, os avanços nas tecnologias emergentes, como a IA, a computação quântica e a robótica, estão abrindo caminho para um futuro que antes se pensava ser ficção científica. À medida que continuarmos a explorar o potencial destas tecnologias, iremos sem dúvida descobrir novas formas de melhorar as nossas vidas e moldar o mundo que nos rodeia. (Será?) Ou estamos criando uma "cobra", para picar-nos depois?!?.

O mundo está mudando em ritmo acelerado e a tecnologia está na vanguarda dessa transformação. Tecnologias emergentes, como inteligência artificial, computação quântica e robótica, estão revolucionando a maneira como vivemos, trabalhamos e interagimos uns com os outros. Desde carros autónomos até robôs cirúrgicos, estas tecnologias estão transformando as indústrias e mudando a forma como pensamos sobre o que é possível. Neste ensaio, exploraremos os avanços mais recentes nesses campos e seu impacto potencial em nosso futuro. Junte-se a nós enquanto vislumbramos o mundo emocionante das tecnologias emergentes e descobrimos o que o futuro reserva.

Este avanço na IA foi possível graças ao conceito de aprendizagem automática, que permite aos computadores aprender e melhorar por si próprios, sem serem explicitamente programados para tal finalidade. A aprendizagem automática, juntamente com outras tecnologias emergentes, como a computação quântica e a robótica, estão a mudando e interferindo nas formas como interagimos com a tecnologia e moldando o futuro do nosso mundo. (Espero e desejo que seja para melhor, e que não ocorra como o avião).

A inteligência artificial é um campo da ciência da computação que envolve o desenvolvimento de máquinas inteligentes que podem executar tarefas que normalmente requerem inteligência humana, como reconhecimento de fala, resolução de problemas e tomada de decisões. Nos últimos anos, a IA fez avanços significativos em áreas como processamento de linguagem natural, reconhecimento de imagens e análise preditiva. Por exemplo, algoritmos de aprendizagem profunda foram usados para criar assistentes de voz como Siri e Alexa, que podem compreender a linguagem natural e responder a consultas complexas.

Adaptado por:

Daniel Lima. (Jorinter)

[quotesfrommyreading]

The advent of neural-net engines thrills many chess players and coaches, including Sutovsky and Sadler. Carlsen said he was “inspired” the first time he saw AlphaZero play. Engines have made it easier for amateurs to improve, while unlocking new dimensions of the game for experts. In this view, chess engines have not eliminated creativity but instead redefined what it means to be creative.

Yet if computers set the gold standard of play, and top players can only try to mimic them, then it’s not clear what, exactly, humans are creating. “Due to the predominance of engine use today,” the grandmaster So explained, “we are being encouraged to halt all creative thought and play like mechanical bots. It’s so boring. So beneath us.” And if elite players stand no chance against machines, instead settling for outsmarting their human opponents by playing subtle, unexpected, or suboptimal moves that weaponize “human frailty,” then modern-era chess looks more and more like a game of psychological warfare: not so much a spelling bee as a round of poker

 —   Chess Is Just Poker Now