You might like
Ilya Sutskever 曾断言，只要读懂那 30 篇奠基论文，就能掌握人工智能 90% 的精髓。但面对枯燥的数学公式，大多数人很难将其转化为可运行的代码。 最近在 GitHub 上发现 Sutskever 30 这个开源项目，用纯 NumPy 实现了 Ilya Sutskever 推荐的 30 篇奠基性论文，全部完成。…
Like @davidbessis and others, I think that Hinton is wrong. To explain why, let me tell you a brief story. About a decade ago, in 2017, I developed an automated theorem-proving framework that was ultimately integrated into Mathematica (see: youtube.com/watch?v=mMaid2…) (1/15)
youtube.com
YouTube
Automated Theorem Proving and Axiomatic Mathematics
Geoffrey Hinton says mathematics is a closed system, so AIs can play it like a game. They can pose problems to themselves, test proofs, and learn from what works, without relying on human examples. “I think AI will get much better at mathematics than people, maybe in the next…
I mean literally sleep. The biggest differentiator outside talent I observed was between people who sacrificed school, socializing, and career stuff in order to sleep more. Sleeping 10 hours a night for years added up to significantly better performance.
one of the primitives to continuous learning may be salience detection which involves the combination of an input out of distribution for what you would have predicted with your current model that if added to context improves your score or efficiency against various simulated...
I still remember back in grad school. My friend in NLP used to show off, bragging that he had LSTM all figured out. I envied him. Fortunately, my field was Computer Vision. I could survive just knowing my SVMs. In 2024, the inventor of LSTM himself is finally back with the…
What's most interesting to me about the HRM and TRM results on ARC-AGI is that these approaches leverage zero external knowledge (either pretraining knowledge or hardcoded priors, such as those found in an ARC-specific DSL). TRM is the public SotA for such approaches. The lack…
If you have a learning loop that greedily takes in information then you open yourself to ddos (see eg nerd sniping). Boredom is one defense against this. If agents task sub agents to go bring back information they'll need some defense against context flooding.
I read it halfway through, and my mind is already blown by the simplicity and awesomeness of this blog
My regular reminder whenever a vendor discloses a 0-day on an edge device: Patching it doesn’t fix the breach that already happened. If it was exposed for months, patching it is like changing the front door lock while the burglars are already in your living room.
so TIL that Claude Code and the like are written in *React* you can just write HTML & js and certain packages will render it for you. it looks way better than regular terminal output. interesting
underrated second-order effect of Claude Code is the proliferation of beautiful CLIs what tools do people use to build these interfaces? we should make many more of them
The deeper you go into the semiconductor supply chain, the less believable it becomes. > TSMC, a company on a small island, produces over 90% of the world’s most advanced chips > TSMC relies on dutch company ASML for EUV lithography machines > ASML depends on German Company…
China just broke the multimodal curse that's been plaguing AI for years. Alibaba dropped Qwen3-Omni and it's the first model that doesn't sacrifice anything to be multimodal. Every other AI model tanks when you add vision and audio. Text performance craters. Reasoning gets…
Reworked and tweaked YARA rules from Google's great report on BRICKSTORM malware activity - removed unneeded regex usage - specified encoding whenever it was clear (ascii wide -> ascii) - removed nocase modifier - reduced rule with 50+ function names - added meta data with…
We are releasing details on BRICKSTORM malware activity, a China-based threat hitting US tech to potentially target downstream customers and hunt for data on vulnerabilities in products. This actor is stealthy, and we've provided a tool to hunt for them. cloud.google.com/blog/topics/th…
Rocket Lab's Mars Sample Return proposal is definitely my favourite. If the US wants to get those samples back in any reasonable timeframe i think they should get the contract.
It’s clear we can’t wait to bring Mars home. Rocket Lab’s original Mars Sample Return study concept delivers returned samples to Earth faster, more affordably, and at a firm-fixed price. Our first step? A Mars Telecommunications Orbiter to deliver robust, persistent Mars–Earth…
why does this happen? the model believes there's a seahorse emoji, sure, but why does that make it output a *different* emoji? here's a clue from everyone's favorite underrated interpretability tool, logit lens! in logit lens, we use the model's lm_head in a weird way.…
if you want gwern style blog posts out of an ai the final step in your pipeline will be RL with adversarial comparisons to real gwern text until the detector is fooled if you do not already have a blogger gym for your baby ai they will never be the next scott a.
wrote a piece on how we do tiling and gradient descent on TinyTPU included some animations of the weight and bias gradient descent
Carbon Robotics started with a simple question over lunch in Idaho: "What's your biggest farming challenge?" The farmer's answer: weeds. The solution? Autonomous lasers. Let's dig into @carbon_robotics amazing tech ⬇️
United States Trends
- 1. #SmackDown N/A
- 2. Dabo N/A
- 3. Antarctica N/A
- 4. #ZuffaBoxing01 N/A
- 5. Trick Williams N/A
- 6. Clemson N/A
- 7. Ilja N/A
- 8. Golding N/A
- 9. Terry Boers N/A
- 10. Penguins N/A
- 11. Sami Zayn N/A
- 12. RIP Terry N/A
- 13. #readytolove N/A
- 14. Southern Hemisphere N/A
- 15. Ron Holland N/A
- 16. Marner N/A
- 17. Royce Keys N/A
- 18. Leafs N/A
- 19. NCAA N/A
- 20. Blades Brown N/A
Something went wrong.
Something went wrong.