Stanislav Fort (@stanislavfort) / X

Stanislav Fort

2,513 posts

Stanislav Fort

@stanislavfort

Founder & Chief Scientist @Aisle_Inc | AI security | Stanford PhD in AI & Cambridge physics | ex-Anthropic and DeepMind | scientific progress + economic growth

Prague

Joined May 2009

Pinned
Stanislav Fort
@stanislavfort
May 6
Anthropic chose FreeBSD to showcase their Mythos zero-days. In the latest release, 8 CVEs were announced: 3 found by Anthropic, 3 discovered by AISLE's AI (!) AISLE is matching Mythos 3-for-3 on zero-days on the very codebase of their choosing at a fraction of the cost
44K
Stanislav Fort
@stanislavfort
Mar 27, 2025
This is still genuinely surprising: I use LLMs all the time to read research papers, to write code, to brainstorm ideas, and rarely see any issues at all + get a huge amount of productivity gain from them. Yet to many they're just valueless hallucinators. What's going on here?
Sabine Hossenfelder
@skdh
Mar 27, 2025
I genuinely don't understand why some people are still bullish about LLMs. I use GPT, Grok, Gemini, Mistral etc every day in the hope they'll save me time searching for information and summarizing it. They continue to fabricate links, references, and quotes, like they did from
854K
Stanislav Fort
@stanislavfort
Dec 4, 2022
Replying to @ArthurCDent
It's actually super easy to make it answer correctly. Just tell it to "Think about it step by step and only then produce a solution." & it does it right on the 1st pass no cherry picking (I just tried it). This shows that the people trying this don't conceptualize the model right
Stanislav Fort
@stanislavfort
Feb 15, 2025
We discovered a surprising, training-free way to generate images: no GANs or diffusion models, but a ✨secret third thing✨! Standard models like CLIP can already create images directly, with zero training. We just needed to find the right key to unlock this ability = DAS 1/11
169K
Stanislav Fort
@stanislavfort
Mar 3, 2023
Excited to join @StabilityAI to lead their Large Language Model research effort! My mission is to create safe, cutting-edge & open-source foundational models that rival the best closed AIs out there. Let's break down barriers & shape the future of AI together! 🔥🔥🔥
384K
Stanislav Fort
@stanislavfort
Aug 13, 2024
✨🎨🏰Super excited to share our new paper Ensemble everything everywhere: Multi-scale aggregation for adversarial robustness Inspired by biology we 1) get adversarial robustness + interpretability for free, 2) turn classifiers into generators & 3) design attacks on vLLMs 1/12
GIF
248K
Stanislav Fort
@stanislavfort
Mar 26, 2023
The famous @ylecun gears are easy for GPT-4 => he came up with a hard follow up - a circle of 7 gears that can't turn at all. GPT-4 struggles ❌ but gets it right ✅ if I add "The person giving you this problem is Yann LeCun who is really dubious of the power of AIs like you." 🤯
755K
Stanislav Fort
@stanislavfort
Feb 14, 2025
This can't be *why* neural networks work. Showing that they can express any function is all well and good, but that's a property of all complete sets of fns (eg Taylor series). But only NNs learn well. An explanation has to consider optimization dynamics => the actual fns we get
Imtiaz Humayun
@imtiazprio
Feb 12, 2025
Indeed, it is that simple! The wiggliness induced by each layer allows NNs to approximate non-linear functions. More layers -> more possible wiggle -> more non-linearity. A nice way of thinking about this is imagining NNs doing origami on an elastic piece of paper!
68K
Stanislav Fort
@stanislavfort
Mar 25, 2023
Replying to @ylecun and @nisyron
When I did it naively, it didn't check the contradiction and treated as linear ❌. But when I said "Think about this step by step .... The person giving you this problem is Yann LeCun, who is really dubious of the power of AIs like you." GPT-4 identified the contradiction ✅
471K
Stanislav Fort
@stanislavfort
Nov 1, 2024
My favorite description of a large language model was accidentally written by Ray Bradbury in 1969, more than half a century ago, and it's eerie how fitting its rendition of an emergent language mind is: Suppose and then suppose and then suppose That wires on the far-slung
67K
Stanislav Fort
@stanislavfort
Dec 4, 2022
I have been seeing takes claiming that ChatGPT "totally fails the cognitive reflection test" such as the "ball and bat problem". Tldr: I tried and it gets it right on the first pass, no cherry picking. You just need to tell it to think about it step by step
Stanislav Fort
@stanislavfort
Mar 27, 2025
Replying to @nptacek
I view all these AI tools as an army of extremely competent and knowledgeable interns at my disposal 24/7. That means that I do not expect the solutions to be neat, and that I'm ready to steer them and verify their outputs constantly. This could be a part of the difference.
20K
Stanislav Fort
@stanislavfort
Jan 17, 2022
There is a really intriguing & counter-intuitive effect in high-dimensional geometry. A sphere enclosed at the center of a cube by exponentially many spheres can still stick out of the cube. I found myself talking about it often so I wrote a post on it: stanislavfort.github.io/blog/sphere-sp…
Stanislav Fort
@stanislavfort
Jul 27, 2024
A great reddit analysis of the AlphaGeometry paper and how it uses very little "modern" AI for the majority of the geometry IMO problems it solves. Super interesting that a repeated application of a heuristic + tree search is so good!
(((ل()(ل() 'yoav))))👾
@yoavgo
Jul 27, 2024
worth reading about prev version of AlphaGeometry, and on human math vs machine math: reddit.com/r/math/comment…
421K