Geoffrey Hinton 🙌: " I believe they [LLMs] have subjective experiences. But they don't think they do because everything they believe came from trying to predict the next word a person would say.... they have false beliefs about themselves because they have our beliefs" youtu.be/jrK3PsD3APk?...

Conscious machines? Thou does protest too much: "only biological beings are capable of consciousness... should stop pursuing projects that suggest otherwise" and “If you ask the wrong question, you end up with the wrong answer. I think it’s totally the wrong question.”
www.cnbc.com/2025/11/02/m...

Will we ever run out of jobs for humans to do? People argue new jobs always pop up. My "giant bowl of M&Ms theory" says, every time you take an M&M it uncovers 4-5 new M&Ms. But even a giant bowl will run out. What's left? The human connection. Playing chess with a human, scarier, but more fun!

Kaggle launched Game Arena, "AI models and agents compete head-to-head", in strategic games starting with chess. So Claude is playing against Grok and friends. See www.kaggle.com/blog/introdu... and Hikaru is streaming live right now www.youtube.com/watch?v=-nBy... covering the games.

The full story of how I made Talknotes in a week and sold it for $200,000 one year later 🧐

It covers:
- 💡 How I got to the idea
- 🤝 Getting my first customers
- 📈 Growing the app (and burning out)
- 💸 Selling it

Took me more 8+ months to finish it, but it's finally here! 😃

youtu.be/3XZK7RDCvK0

Simon Willison's benchmark for LLMs :), The last six months in LLMs, illustrated by pelicans on bicycles simonwillison.net/2025/Jun/6/s...

MCP, Model Context Protocol, seems to "magically" connect software and apps (e.g. blender, unity, unreal, godot) to LLMs like claude (or chatgpt).

This shows you how, including: 1. don't just dump your API into your MCP, 2. work at a higher level, and 3. add tests. www.youtube.com/watch?v=eeOA...

AI TikTok Block (suggestion)

I want to long-press on someone's face, and then "AI block" them, never to see a post with their face again.

Similarly, you can "tok block" people by their voice... or if their name is uttered?

Creating a "filter bubble", for better and worse. Ignore-ance is bliss!?

Search & recsys have historically drawn inspiration from language modeling, such as Word2vec for item embeddings, and GRUs and BERT for sequential models. LLMs are no different and have led to updates in model architecture, training data & paradigms, etc.

eugeneyan.com/writing/recs...

This podcast episode is a great review of developments around vision language models pca.st/episode/26e8...

Claude (and LLMs) continues to blow my mind. From super vague description of an anecdote, Claude knows what I'm talking about :). You complete me :)

Ilya Sutskever's talk from neurIPS 2024 www.youtube.com/watch?v=YD-9...

Chess is a great (and fairly mature) test case for coexistence with AI. You'd expect the fact that AI can beat every human to "ruin" the fun, but it hasn't. Yesterday's world championship between Ding and Gukesh was amazing and proof of that :). 🧵.