Lifetime observation: the fancier the restaurant, the fewer the smiles—and even fewer the laughs.

As a researcher at a frontier lab I’m often surprised by how unaware of current AI progress public discussions are. I wrote a post to summarize studies of recent progress, and what we should expect in the next 1-2 years: julian.ac/blog/2025/09/2…

Finally had a chance to listen through this pod with Sutton, which was interesting and amusing. As background, Sutton's "The Bitter Lesson" has become a bit of biblical text in frontier LLM circles. Researchers routinely talk about and ask whether this or that approach or idea

TRUE WEALTH (3rd Ed) Worriless sleeping Clear conscience Reciprocal gratitude Genuine sense you did your duty toward the Palestinians & other victims Absence of envy Foamy coffee Crusty bread Trustworthy friends & inexperienced enemies Frequent smiles No meals alone Nice clothes

I'm releasing nanoproof - a minimal open-source implementation of AlphaProof (Tom Zahavy). It uses nanochat (Andrej Karpathy) for pretrain+midtrain+sft, and the official AlphaProof pseudocode for MCTS+the RL loop. First version has 32.8% on minif2f with ~0.001% of the compute. (link to

im excited to announce that with opus 4.5 i've fully become a slopware developer, incapable of writing code on my own. my brain is atrophic thanks to anthropic 😭

This is extraordinary and profoundly symbolic: Ai Weiwei has returned to China for the first time in 10 years and says he concluded that Beijing is "more humane" than Germany which he describes as "insecure and unfree". He gave an interview to Germany's Berliner Zeitung after

I often think about what are the important skills to have in todays "ai climate" and it keeps coming back to me that being adaptive and open-minded is more important than being a genius. I feel like many people attach too much ego to stuff like being "the person who does X"

every time i try to have an original research idea but Zhang et al. already published it 3 years ago

ICML, one of the prestigious conferences in AI, has received more than 30,000 submissions this year. When AI can write and rewrite papers, expect this number to grow even more in the upcoming years. No one has the capacity to peer review this volume, so inevitably AI will be

people who still do not vibecode be like

another day at the prompt factory

Turns out with claude code, my decades long strategy of NOT deeply learning: - regexs - sql - nginx confs - elaborate shell commands - advanced shell scripting - any javascript framework - perf optimization - webpack, cdns, bundlers - 1000 other things ...was entirely correct.

AI coded apps cause so much tech debt! Nooo! Yes, and then if the debt gets too big in a year, you get the next better version of Claude to write you a new better one in 10 minutes. People hanging on to very outdated tradeoff calculations...

jobs left after the singularity: head of taste 10x vibecoding engineer degenerate gambler being hot podcaster

if hyperparameters didnt exist, we would have had flying cars by now

finally some good ahh experiment tracker GoodSeed Internaional

Being smart (iq wise) is like having a good transformer architecture in the skull but that's not all you need. One needs a great deal of weight decay, plasticity and perhaps most importantly, to optimize good loss functions. Many sota models optimize the wrong thing.

The text of my speech last week at the University of Alberta convocation ceremony: Good afternoon graduating students, parents, and ladies and gentlemen of the university community. It is my great honour to receive this degree from the University of Alberta. I am receiving this