Michael retweeted
A language model is a wonderland of strangely specific neurons, from gargling to comic exasperation to the moment a guard looks away.
We made a searchable atlas of all 10 million neuron poles in Qwen3.6.
Play with it here: goodfire.com/research/public…
Michael retweeted
I'm amused by the situation I'm in right now. Pulled a 3-nighter to prepare ICLR rebuttal for the paper I like the most, trying to defend AI bs reviews.
wait for 10 days, no reply at all. dead silence
AC does not even leave a message.
yea I'm not playing this dumb game anymore
Michael retweeted
/
Suicaの新たなイメージキャラクターが決定しました🎉
~お客さまから愛称を募集します~📢
\
2026年9月、Suicaの新たなイメージキャラクターのデザイン候補3案を発表し、お客さま投票を実施した結果、総得票数1,664,399票のうち860,152票を獲得した🅱️のキャラクターに決定しました🏅🎉
今後、新しいキャラクターの愛称を、本日15時から12月15日まで、お客さまから募集します🗳️
↓ ↓ ↓愛称募集の特設サイトはこちら↓ ↓ ↓
jreast.co.jp/suica-new-chara…
↓ ↓ ↓プレスリリースの詳細はこちら↓ ↓ ↓
jreast.co.jp/press/2026/2026…
#JR東日本 #Suica #新キャラクター
If these authors are reading so much AI writing that it's changing how they write (so as to be classified as AI writing), then perhaps they should be more invested in good tools to identify AI writing?
Wrt the pangram/AI writing debates: Everyone is accounting for the fact of language change, right? How many of us write now is very different than how we wrote 2 years ago, in part because we are inundated with genAI patterns in what we read.
47
Michael retweeted
(Finally) wrote up "Understanding Positional Embeddings in Modern LLMs."
A RoPE from-scratch tutorial missing from my LLM building blocks collection + what recent architectures do (NoPE, partial RoPE, hybrids).
Prob not the flashiest topic, but essential for understanding LLMs
Michael retweeted
Can use just 10 images to teach a model what a sea anemone looks like?
In our #neurips2026 paper, "Dataset Distillation via Drifting," we distill ImageNet's 1,300 images per class down to just 10, optimized from noise.
drifting-distillation.github…
More info in thread! ⬇️
1/N
If you don't know anything about the guy, then maybe you don't need to comment.
The guy works for the Russian state media network Russia Today (RT).
I don't know anything about this guy, but it seems to violate the most basic principles of due process that the EU can do this to one of its own citizens without any kind of trial.
1
66
Michael retweeted
This is a prize for AFRICAN LITERATURE
EVERY entrant was African! This can’t be racist or biased!
Legitimate African writers are being cheated by an AI grifter and YOU are defending that!
Michael retweeted
It's insanely funny to see people freaking out in the QTs over how racist the Pangram guy is for claiming that this PR statement comes up as AI and how he's determined to attack African literature, etc, & then read the statement itself, the most obvious LLM writing ever produced
Michael retweeted
Even in this outage, Teams continues to perform seamlessly like the apocalyptic cockroach that it is.
You genuinely couldn't write this stuff, it would seem far too on-the-nose nowadays.
The Caine Prize, which awarded its top prize this year to a story Pangram classifies as almost entirely AI-generated, conducted an investigation and found "no evidence" of AI use.
Strangely, their official statement on Instagram seems to also be blatantly AI-generated.
2
141










































