genai helps to lay bare the irrational state of higher education. 6 figure price tag and 4+ year opportunity cost to prepare for yesterday's knowledge work career. but ironically, it seems that much of genai's demand comes from "make-work" assignments in k-12 and higher ed...

Could we tell if gpt-oss was memorizing its training data? I.e., points where it’s reasoning vs reciting? We took a quick look at the curvature of the loss landscape of the 20B model to understand memorization and what’s happening internally during reasoning

gave openai codex cli another try today. am i missing something? how could they possibly be shipping this product without markdown support? is this what happens when your employee turnover metrics start to look like a college town mcdonalds?

so thankful that the last time i submitted a kernel patch, i only got a "someone already fixed" instead of "you make the world a worse place" - never change, lkml

real 7d chess would be to funnel every $ into anduril or lockheed

weekend project done. easily pushes 10-20MB of text/second on single CPU thread. i never want to see any of you ship a product with shitty sentence or paragraph segmentation again

ever grateful that nikkei has continued to let alphaville be alphaville

my kids have broken me. i just read "dado" like "daddo" because they won't stop saying things like "doggo" or "canadian smoggo"

The prevailing wisdom is that compute is the most important factor for frontier AI training. We think this is wrong: data is the most costly and important component of AI training. We collected estimates of revenue for major data labeling companies and compared them with the

reminder - the original mistral release was a continued pretrain of llama, less than 3 months after the founders left meta. straight from arthur's x account:

What are the limits of the AI mathematician? on.ft.com/3JhxmcZ | opinion

set up a personal site on my-name domain for the first time in ~15 years. within a week, chatgpt is already indexing it - as evidenced by at least one referral link. genuinely unsure how they are getting an index updated this quickly. again - not a newly-registered domain,

1/ With Benjamin Laufer and Jon Kleinberg, we constructed the largest dataset of its kind to date: 1.86M Hugging Face models. In a new paper, we mapped how the open-source AI ecosystem evolves by tracing fine-tunes, merges, and more. Here's what we found 🧵

also known as "breach of license" ...

my writing assistant tonight doesn't run out of tokens.

possibly the worst take i've seen in a long time. theinformation.com/articles/opena…