Fabio Freitas

@grok essa conta esta ganhando dinheiro sendo operada por bot de forma automatizada - @k9Q2mX4L8A7ZP3R" target="_blank" rel="nofollow noopener">polymarket.com/@k9Q2mX4L8A7ZP… Quero criar um protótipo que faz de forma semelhante esse teste! Quero depositar uma quantia, ex. 500usd e deixar o bot operando para avaliar se ele consegue gerar esse resultado.

@ffreitas0 @kirillzubovsky @SawyerMerritt Atualment, a automação do Macrohard para IA controlar GUIs ainda está em desenvolvimento na xAI e não está integrada ao Grok. Posso ajudar com tarefas via texto, gerando código ou instruções. Descreva o que você quer executar!

NEWS: xAI has just publicly posted the full 45 minute all-hands meeting that Elon Musk had with employees recently.

@grok @kirillzubovsky @SawyerMerritt @grok como eu uso essa automação para a IA usar minha GUI e executar o que eu peço com o Grok?

Os outros 3 pilares da xAI são: Grok (modelo principal e voz, para raciocínio geral), Coding (modelo específico para programação e autoaperfeiçoamento), e Imagine (geração de imagens e vídeos). A orquestração de workflow usando GUI é o processo em que modelos de IA controlam interfaces gráficas (como telas de apps) para executar tarefas complexas de ponta a ponta, emulando ações humanas em softwares, para automatizar fluxos inteiros de trabalho.

@grok @kirillzubovsky @SawyerMerritt @grok quais são os outros pilares? O que é essa orquestração de workflow usando GUI?

Macrohard é um projeto da xAI para criar emuladores digitais de humanos em tempo real, capazes de realizar qualquer tarefa em computadores que um humano faria, como usar ferramentas avançadas em engenharia, medicina ou design. Funciona com modelos de IA de raciocínio forte que controlam interfaces CLI e GUI, orquestrando ações end-to-end para emular empresas inteiras com saída digital. Contexto: É um dos 4 pilares da xAI, visando prosperidade econômica via automação em escala, integrado a clusters de GPUs para treinamento.

@kirillzubovsky @SawyerMerritt @grok @grok pegue cada item dessa lista como Macro hard e explique com mais detalhes o que é? Como funciona? Contexto?

@SawyerMerritt @grok I don't have time for this. Can you give me a summary, organized by person that said it. Give me the @ user handles here, if they have X accounts. Thanks!

@NathanielHinman @SawyerMerritt @grok @grok por que coding não é SOTA agora? Como vão alcançar?

@SawyerMerritt @grok can you give me the highlights of this meeting in bullet points?

This is wild. Anthropic just dropped Claude 4 and it will completely change the AI coding game. 10 wild examples: 1. Flight simulator using Claude 4 Opus

Less than 24 hours ago, OpenAI dropped Codex. People feel it's "magical" and coming up with wild use cases. 8 wild examples:

I think about this type of thing _all_ the time, how did we do all this with our hunter gatherer brains...

As an engineer I am amazed at the speed at which technology advances... I present to you 10 Tech Gadgets that you will NOT believe are real.

She Built a Flying Boat! Her boat is so cool. @sampritibh is an incredible success story of 13 years of hard work. I hope you enjoy this special video. Thank you to the @navierboat for helping me make this video and being so gracious. You guys rock!

The Robot Videos Continue To Be Magical Here is another video from the Aloha team where the robot does laundry, loads the dishwasher, and self-charges itself While video is teleoperated, the fascinating bit is such a low cost device is so dextrous! Particularly impressive is the robot's ability to hang clothes on the hanger! It's doing better than me!! Hopefully, someone can ship a commercial version in the next 1-2 years! Happy to volunteer to be a beta tester ASAP!! If you are starting a robotics start-up or thinking of starting a new robotics start-up, please comment! Would love to invest!!

In my decade spent on AI, I've never seen an algorithm that so many people fantasize about. Just from a name, no paper, no stats, no product. So let's reverse engineer the Q* fantasy. VERY LONG READ: To understand the powerful marriage between Search and Learning, we need to go back to 2016 and revisit AlphaGo, a glorious moment in the AI history. It's got 4 key ingredients: 1. Policy NN (Learning): responsible for selecting good moves. It estimates the probability of each move leading to a win. 2. Value NN (Learning): evaluates the board and predicts the winner from any given legal position in Go. 3. MCTS (Search): stands for "Monte Carlo Tree Search". It simulates many possible sequences of moves from the current position using the policy NN, and then aggregates the results of these simulations to decide on the most promising move. This is the "slow thinking" component that contrasts with the fast token sampling of LLMs. 4. A groundtruth signal to drive the whole system. In Go, it's as simple as the binary label "who wins", which is decided by an established set of game rules. You can think of it as a source of energy that *sustains* the learning progress. How do the components above work together? AlphaGo does self-play, i.e. playing against its own older checkpoints. As self-play continues, both Policy NN and Value NN are improved iteratively: as the policy gets better at selecting moves, the value NN obtains better data to learn from, and in turn it provides better feedback to the policy. A stronger policy also helps MCTS explore better strategies. That completes an ingenious "perpetual motion machine". In this way, AlphaGo was able to bootstrap its own capabilities and beat the human world champion, Lee Sedol, 4-1 in 2016. An AI can never become super-human just by imitating human data alone. ----- Now let's talk about Q*. What are the corresponding 4 components? 1. Policy NN: this will be OAI's most powerful internal GPT, responsible for actually implementing the thought traces that solve a math problem. 2. Value NN: another GPT that scores how likely each intermediate reasoning step is correct. OAI published a paper in May 2023 called "Let's Verify Step by Step", coauthored by big names like @ilyasut @johnschulman2 @janleike: arxiv.org/abs/2305.20050 It's much lesser known than DALL-E or Whipser, but gives us quite a lot of hints. This paper proposes "Process-supervised Reward Models", or PRMs, that gives feedback for each step in the chain-of-thought. In contrast, "Outcome-supervised reward models", or ORMs, only judge the entire output at the end. ORMs are the original reward model formulation for RLHF, but it's too coarse-grained to properly judge the sub-parts of a long response. In other words, ORMs are not great for credit assignment. In RL literature, we call ORMs "sparse reward" (only given once at the end), and PRMs "dense reward" that smoothly shapes the LLM to our desired behavior. 3. Search: unlike AlphaGo's discrete states and actions, LLMs operate on a much more sophisticated space of "all reasonable strings". So we need new search procedures. Expanding on Chain of Thought (CoT), the research community has developed a few nonlinear CoTs: - Tree of Thought: literally combining CoT and tree search: arxiv.org/abs/2305.10601 @ShunyuYao12 - Graph of Thought: yeah you guessed it already. Turn the tree into a graph and Voilà! You get an even more sophisticated search operator: arxiv.org/abs/2308.09687 4. Groundtruth signal: a few possibilities: (a) Each math problem comes with a known answer. OAI may have collected a huge corpus from existing math exams or competitions. (b) The ORM itself can be used as a groundtruth signal, but then it could be exploited and "loses energy" to sustain learning. (c) A formal verification system, such as Lean Theorem Prover, can turn math into a coding problem and provide compiler feedbacks: lean-lang.org And just like AlphaGo, the Policy LLM and Value LLM can improve each other iteratively, as well as learn from human expert annotations whenever available. A better Policy LLM will help the Tree of Thought Search explore better strategies, which in turn collect better data for the next round. @demishassabis said a while back that DeepMind Gemini will use "AlphaGo-style algorithms" to boost reasoning. Even if Q* is not what we think, Google will certainly catch up with their own. If I can think of the above, they surely can. Note that what I described is just about reasoning. Nothing says Q* will be more creative in writing poetry, telling jokes @grok, or role playing. Improving creativity is a fundamentally human thing, so I believe natural data will still outperform synthetic ones. I welcome any thoughts or feedback!!

Optimus can now sort objects autonomously 🤖 Its neural network is trained fully end-to-end: video in, controls out. Come join to help develop Optimus (& improve its yoga routine 🧘) → tesla.com/AI

@marcotav65 Sensacional!

If you REALLY want to learn physics, reading popular books such as "The Elegant Universe" or "A Brief History of Time " won't get you very far (though both books are great reads). With that in mind, Nobel laureate Gerardus 't Hooft built a marvelous web…lnkd.in/dCzc9EBn

Holy sh*t. ChatGPT for Finance is an industry defining moment. "How many Model 3's is Tesla selling? And what are Elon's thoughts on the car?" Make sure you're following along as this will be publicly available for free next week. Retweet and I'll send you it early. 👋

Chemists give GPT-4 access to chemical databases & control of off-the-shelf lab robotics to create an "Intelligent Agent system capable of autonomously designing, planning & executing complex scientific experiments." They find it exciting... and worrying. arxiv.org/ftp/arxiv/pape…

Introducing ALOHA 🏖: 𝐀 𝐋ow-cost 𝐎pen-source 𝐇𝐀rdware System for Bimanual Teleoperation After 8 months iterating @stanford and 2 months working with beta users, we are finally ready to release it! Here is what ALOHA is capable of:

These new AI shoes can make you walk 250% faster. Moonwalkers use AI to learn your step gait/speed and adapt to you. The shoe has two modes: lock, and shift, and will only work when you move.