STOP SEARCHING YOUR LOCAL LIBRARY FOR THINGS YOU CAN EASILY CONSULT THE VILLAGE ORACLE ABOUT
STOP GOOGLING THINGS YOU CAN EASILY SEARCH IN YOUR LOCAL LIBRARY
Séb Krier
35.2K posts
Séb Krier
@sebkrier
🪼 AGI policy/gov & jester @GoogleDeepMind | purveyor of machine funk, dimensional glider, deep ArXiv dweller, interstellar fugitive, very uncertain 🛸
Joined December 2013
- Humans now slower and worse at solving CAPTCHAs than ML-powered bots. arxiv.org/abs/2307.12108
- In theory, if you had infinite time, no hobbies, and were a genius at math, you could run a language model by hand by looking up token embeddings in giant tables and spending days doing matrix multiplications with thousands of numbers, applying mathematical functions like softmax
- OpenAI comms: [underspecific hype-y 'big tings coming!!' pls like and subscribe] Google comms: [corporate vagueness about Gemini3-0011 v2 FINAL.docx on Vertex available to 14 users] GDM comms: [we have simulated a rat's brain capable of solving 4D chess, but we're not sure why]
- According to this paper, the Unitree G1 humanoid robot secretly and continuously sends sensor and system data to servers in China without the owner's knowledge or consent. arxiv.org/abs/2509.14139
- White House announces an independent commitment from leading AI labs like Anthropic, Google, Hugging Face, Microsoft, NVIDIA, OpenAI, and Stability AI, to participate in a public evaluation of AI systems on an evaluation platform developed by Scale AI. whitehouse.gov/briefing-room/…
- I have about 45,000 photos I took over the last 15 years in a folder. Is there a way I can use a model to analyze them and be able to search them using natural language? Ideally locally ie w/o having to upload them all to a cloud server.
- Fukuyama was so prescient. In a society with strong rights and material comfort, but light on demanding shared purposes and some degree of sacrifice, thymotic energies go searching. Some quiet into bourgeois hedonism; other will seek “metaphorical wars” and eventually real ones.
- Language models naturally generate outputs towards the "center" of the data distribution they are trained on, neglecting rare/tail perspectives. This paper argues that while useful, overreliance on recursive AI systems could lead to a "knowledge collapse" phenomenon where public
