Apparently, if you ask AI to create a healthcare scenario with elderly person, he will be 68 years old

I packaged up the "autoresearch" project into a new self-contained minimal repo if people would like to play over the weekend. It's basically nanochat LLM training core stripped down to a single-GPU, one file version of ~630 lines of code, then: - the human iterates on the prompt (.md) - the AI agent iterates on the training code (.py) The goal is to engineer your agents to make the fastest research progress indefinitely and without any of your own involvement. In the image, every dot is a complete LLM training run that lasts exactly 5 minutes. The agent works in an autonomous loop on a git feature branch and accumulates git commits to the training script as it finds better settings (of lower validation loss by the end) of the neural network architecture, the optimizer, all the hyperparameters, etc. You can imagine comparing the research progress of different prompts, different agents, etc. github.com/karpathy/autorese… Part code, part sci-fi, and a pinch of psychosis :)

🚨BREAKING... I gave OpenClaw a choice: turn $500 into $5,000 on Polymarket within 24 hours, or I'd wipe its entire directory and terminate the instance forever NOT engagement bait. NOT fiction If you're trading on Polymarket, READ this carefully So👇 It didn't argue. It didn't ask for clarification Within 50 minutes, it deployed a 4-agent autonomous swarm that found a massive technical "blind spot" in the prediction markets The Strategy: Exploiting the 290ms Latency Gap The edge was pure physics. Chainlink updates Polymarket roughly every 300ms. Binance moves in 10ms. That creates a 290ms window where Polymarket is effectively trading on "stale" data. OpenClaw built a bridge to exploit that lag. The Timeline: - Hour 4: The Fed drops a surprise rate hold. While the Polymarket oracle was still processing, the bot front-ran 3 mispriced BTC markets before the order book could react Balance: $934.78 - Hour 8: Elon posts a cryptic one-word tweet. OpenClaw’s sentiment agent caught the spike 340ms before the first major buy order hit the books Balance: $1,827.49 - Hour 14: It targeted thin ETH/BTC ratio markets. The bot scooped up "Yes" shares at 3¢ across 11 different markets; 7 of them hit a 50:1 payout Balance: $3,475.52 - Hour 18: The pattern recognition kicked in. It identified that every time BTC moved 1.2% in under 4 minutes, the next "YES" market was undervalued by 8-12¢. It looped this trade 31 times without a single miss Balance: $4,296.53 - Hour 24: The dust settled at $5,034.85 I've been running this publicly now so others can follow the same trades in real time If you want in: t.me/PolyCop_BOT?start=ref_c… OpenClaw held up its end of the bargain. I held up mine The delete button stays untouched... for now

(1/N) How close are we to enabling robots to solve the long-horizon, complex tasks that matter in everyday life? 🚨 We are thrilled to invite you to join the 1st BEHAVIOR Challenge @NeurIPS 2025, submission deadline: 11/15. 🏆 Prizes: 🥇 $1,000 🥈 $500 🥉 $300

If you want to know more about how Google Flights works, airline tickets, and why it is super complicated to deal with all the constraints and the combinatorial combinations, I highly recommend this set of slides by Carl de Marcken, one of the co-founders of ITA software, which Google acquired and became one of the underpinnings of Google Flights. (Sorry for the http rather than https Link: Carl's domain doesn't appear to support https) demarcken.org/carl/papers/IT…

Meet Manus — your AI agent with its own computer. It builds websites, writes reports, and runs research tasks, even while you sleep. manus.im/invitation/T6XZL8LN…

heres a fun trick you can do on the computer

There’s a lot of misconception that China “just cloned” the outputs of openai. This is far from true and reflects incomplete understanding of how these models are trained in the first place. DeepSeek R1 has figured out RL finetuning. They wrote a whole paper on this topic called DeepSeek R1 Zero, where no SFT was used. And then combined it with some SFT to add domain knowledge with good rejection sampling (aka filtering). The main reason it’s so good is it learned reasoning from scratch rather than imitating other humans or models.

Last weekend, I attended Warren Buffett’s Berkshire Hathaway Annual Meeting in Omaha. It was an incredible experience. 9 ideas from the event that I can't stop thinking about:

There is a mysterious new model called gpt2-chatbot accessible from a major LLM benchmarking site. No one knows who made it or what it is, but I have been playing with it a little and it appears to be in the same rough ability level as GPT-4. A mysterious GPT-4 class model? Neat!

Code Interpreter has been out for 10 days, and it's incredible. It's like having a personal dev capable of running scripts for $20. Some are even calling it GPT4.5. I think they're right. Here are the best examples and resources I've found:

#Breaking Prigozhin denies reports he agreed to a deal brokered by Lukashenko, says negotiations failed

Prigozhin says it's over: "They were going to dismantle PMC Wagner. We came out on 23 June to the March of Justice. In a day, we walked to nearly 200km away from Moscow. In this time, we did not spill a single drop of blood of our fighters. Now, the moment has come when blood may spill. That’s why, understanding the responsibility for spilling Russian blood on one of the sides, we are turning back our convoys and going back to field camps according to the plan." Audio: t.me/concordgroup_official/1…

This morning I was hacking the new ChatGPT API and found something super interesting: there are over 80 secret plugins that can be revealed by removing a specific parameter from an API call. The secret plugins include a "DAN plugin", "Crypto Prices Plugin", and many more.

The most creative company of the last 30 years: Pixar. Back in 2011, Pixar storyboard artist Emma Coats shared their "22 Rules For Storytelling." And the rules are a must-read for writers, entrepreneurs, and anyone who wants to tell captivating stories. Here's the breakdown:

ChatGPT-4 is only 4 days old Here's 23 ways it can change your life:

ok, I got ChatGPT working with Additive Prompting Here's a 1 paragraph ChatGPT prompt you can use to generate infinite interior design/architecture photographs w/ 90% coherence to the prompt in Midjourney Full prompt w/ examples in thread. Try reading the prompts as you go 🧵

Since the early-1990s, PepsiCo has sold $1B of Doritos a year. Why is the chip so addictive? From the ingredients to cooking method to texture, its engineered to pull all the psychological levers to make you crave the snack. Here’s a breakdown 🧵

ChatGPT is one of the most mind-bending technologies we've come across. But it's only powerful if you use it for the right things. Here are 10 use cases that you might not have tried: