Simon Winder @robotbugs

EE then PhD computational neuroscience, ~15 years machine learning / computer vision at ex @MSFTResearch. Deeply involved in AI, RL, robotics, and art. simonwinder.com Seattle, WA, USA Joined August 2008

Tweets

2K
Followers

335
Following

340
Likes

443

Massimo @Rainmaker1973

2 weeks ago

Did you know? When snails hatch, they eat the shells of their eggs. This gives them needed calcium for their own shells.

28 68 541 61K 63

Download Video

Brett Adcock @adcock_brett

3 weeks ago

Round 2: Frank the Tank is back

309 203 2K 241K 197

Download Video

Massimo @Rainmaker1973

3 months ago

George Porter shows how nylon was discovered by chemists, and gives a demonstration of how we can spin nylon threads. [📹 The Royal Institution]

52 599 4K 198K 1K

Download Video

Hashem Al-Ghaili @HashemGhaili

4 months ago

The Visitors (Made with Veo 3). Humans react to alien visitors. This mini project was completed in 8 hours, from start to finish, by one person. The tools I used are listed at the end of the video.

158 547 4K 742K 2K

Download Video

Igor Sushko @igorsushko

3 months ago

🚨 China: Massive explosion at Shandong Youdao Chemical plant in Gaomi city. At least 5 dead, 19 injured, 6 missing. The factory manufactured Chlorpyrifos, neurotoxic organophosphate insecticide that acts as a nerve agent.

119 863 3K 165K 161

Download Video

Luca Beurer-Kellner @lbeurerkellner

3 months ago

😈 BEWARE: Claude 4 + GitHub MCP will leak your private GitHub repositories, no questions asked. We discovered a new attack on agents using GitHub’s official MCP server, which can be exploited by attackers to access your private repositories. creds to @marco_milanta (1/n) 👇

47 495 2K 501K 2K

Download Image

Simon Winder @robotbugs

4 months ago

Pythagorian triples in complex number space (a+ib)^2 + (c+id)^2 = (e+if)^2.

0 0 0 20 0

Download Image

Timothy B. Lee @binarybits

4 months ago

This perfectly articulates why "model alignment" is the wrong way to think about AI safety. Safety is not a property of models, it's a property of systems that use models. Asking whether a model is safe is like asking whether a computer chip is safe.

13 41 212 16K 55

Download Image

David Kirichenko @DVKirichenko

4 months ago

12/ Despite wartime constraints, Ukraine has rapidly ramped up domestic weapons production. Casper of the 14th Separate UAV Regiment, says 80–90% of the drones his unit now uses are made in Ukraine, compared to near-total reliance on foreign models in the early years of the war.

2 116 874 19K 12

Download Image

Clay Dumas @claydumas

4 months ago

I know we aren’t supposed to interrupt the group chat discourse with woke nonsense like the habitability of our planet, but possibly the most intense heat wave on record is unfolding from Sub-Saharan Africa to East Asia and it’s getting zero attention.

Extreme Temperatures Around The World @extremetemps

4 months ago

0 215 440 25K 23

Download Image

4 25 64 9K 8

Andrew Wheeler @sixfootbrit

7 months ago

@nikitabier Sure, if you ignore o3

3 1 35 16K 2

Download Image

Nikita Bier @nikitabier

7 months ago

Game over. Grok won. If you’re an investor in OpenAI, you can report the tax loss to your accountant this year.

603 642 9K 1.7M 777

Download Image

Justin @justkrup

7 months ago

@karpathy I instantly knew something was up when it sent this

5 4 315 61K 28

Download Image

Chubby♨️ @kimmonismus

7 months ago

A new paper demonstrates that LLMs could "think" in latent space, effectively decoupling internal reasoning from visible context tokens. This breakthrough suggests that even smaller models can achieve remarkable performance without relying on extensive context windows.

60 156 1K 119K 765

Download Image

Randall Balestriero @randall_balestr

7 months ago

Given a pretrained model, spline theory tells you how to alter its curvature by changing a single interpretable parameter! Reducing the model's curvature always significantly improves its downstream task performances/robustness. Also work on Transformers! arxiv.org/abs/2502.07783

7 45 344 27K 281

Download Image

Peyman Milanfar @docmilanfar

7 months ago

The Kalman Filter was once a core topic in EECS curricula. Given its relevance to ML, RL, Ctrl/Robotics, I'm surprised that most researchers don't know much about it - and up rediscovering it. Kalman Filter seems messy & complicated, but the intuition behind it is invaluable 1/4

41 192 2K 255K 2K

Download Image

Fei-Fei Li @drfeifei

7 months ago

Live from Paris, at the #ParisAISummit . I gave the opening keynote about AI’s past, present, future and opportunities. “75 years ago, Alan Turing dared humanity to imagine ‘thinking machines’. But perhaps that vision is too narrow and inward-looking. As AI today is poised to…

141 218 2K 218K 250

Download Image

Luiza Jarovsky, PhD @LuizaJarovsky

7 months ago

🚨 Studies are starting to show what many of us feared: AI use might lead to overreliance and human disempowerment. Below is the SHOCKING conclusion of this particular study [download for future reference]: "1. We surveyed 319 knowledge workers who use GenAI tools (e.g.,…

36 282 1K 82K 900

Download Image

Leo Dirac @leopd

7 months ago

Hyperparameter of the day! Have you adjusted your adam_beta2 lately? Ever? You might want to, especially if you're working in RL. Beta2 usually defaults to 0.999 - what does that mean? Interpret it it by computing 1/(1-beta2), so with beta2=0.999 this comes out to 1000,…

1 2 10 972 9

elvis @omarsar0

7 months ago

Large Memory Models for Long-Context Reasoning This paper focuses on improving long-context reasoning with explicit memory mechanisms. It presents LM2, a Transformer-based architecture equipped with a dedicated memory module to enhance long-context reasoning, multi-hop…