I must study assembler so my children may have liberty to study python so their kids may have liberty to study scratch
RESEARCH
-

Fixing Transformers GPT-OSS MoE Finetuning Issues
By
–
"i thought the transformers gpt-oss MoE finetuning was broken, how did you get it working?"
-
Tech Progress Cannot Solve Self-Driving Fundamental Challenges
By
–
None of the factors involved are of the kind that change meaningfully with tech progress (self-driving hardware will get cheaper, but that's not a big factor)
-
Wolfram’s mechanistic ideas versus mysticism in AI culture
By
–
In our cultural context, almost any allusion to this topic will be interpreted as an appeal to mysticism. In some sense, Stephen Wolfram's ideas about this are similar to your, but carry an opposite (mechanistic, reductionist) connotation for the audience
-

Multiple AIs Debating Health Diagnostics More Accurately and Cheaply
By
–
Don’t be tricked by the headlines. This article about AI and health diagnostics is NOT ABOUT HEALTH. This is about MULTIPLE AIs all arguing together, with internet access, more cheaply and at higher accuracy than one expensive expert. The paper is flawed (the human should have
-

Liquid Releases Fast 450M and 1.6B Parameter VLM Models
By
–
Liquid just released two 450M and 1.6B param VLMs! They're super fast and leverage SigLIP2 NaFlex encoders to handle native resolutions without distortion. Available today on @huggingface
! -
AI Maps Optimize Beaver Reintroduction for Watershed Management
By
–
Better maps could help watershed managers choose the best areas for beaver reintroduction that could bring benefits for both humans and nature. Could AI help? A study led by @StanfordHAI postdoc fellow Luwen Wan reveals a promising research direction:
-
Silicon Valley AI Leaders Turn to Philosophy and Silent Retreats
By
–
The number of Silicon Valley AI people I know going on silent retreats and studying (both ancient and more modern) philosophy has drastically increased this year. Even had one friend hire a philosophy tutor. Leaders rn are seeking a quiet, clearer mind.
-
CulturalGround: Extracting More Training Data from Cultural Knowledge Bases
By
–
Importantly, CulturalGround is further evidence that we are not out of data yet and we can squeeze out more and more training data from the knowledge bases. By just rethinking where cultural data resides and systematically generating factual questions and answers about the
-

Cross-lingual Model Transfer Analysis and Methodology Performance
By
–
Here we share some interesting analysis we perform on the results. There are many languages and regions than one can train for, so it’s natural to examine where our methodology and model does transfer to cultures and languages we don’t cover in the dataset. We noticed that