Marco

not a fan but at least it isn't my email

PSA: Save yourself a desk-reject. Don't use an LLM to populate your BibTeX file. Use dblp.org instead in "condensed" mode.

Of course the Luddites had other things they were protesting, but the modern usage of the term is synonymous with anti-tech. And stochastic parrots proponents have other legitimate concerns, but the feeling of the phrase is now just "anti-LLM", and I think that is going to morph into "anti-tech".

It's pretty clear that, as a technology, LLMs are both very useful and here to stay. Functionally, I currently do not care about the philosophical discussions surrounding them (are they sentient, are they just stochastic parrots, etc.; though of course this is interesting to me intellectually).

I'm not saying that there aren't many legitimate criticisms of LLMs (there are), but stochastic parrots has entered the public lexicon as essentially the rallying cry for the vehemently anti-tech crowd.

If there is a modern refresher for the term Luddite, it will be based on stochastic parrots.

"He is from [MASK] [MASK]" → "San York"? dLLMs fail because they ignore token dependencies. This Factorization Barrier arises from a structural misspecification: models are restricted to fully factorized outputs. We break this barrier with CoDD, enabling coherent parallel generation. 🚀

Are AI models effective collaborators, or mere assistants awaiting your next command? (Preprint: arxiv.org/abs/2602.24188)

To find out, we make AI collaborate with itself, in private information games: tasks that require sharing private information, like this chess board ordering task.

Anyone in my network working on using open models in social science, public health, etc.?

Looking to connect someone.

SIGBOVIK has a Bluesky now! Follow to learn more cutting-edge research from the world’s most comedic and occasionally scientific academic conference

Stochastic parrots really will go down as this generation's "Luddite".

attn: @samwho.dev

Points on a Circle A walkthrough of a popular math problem. Drop N random points on a circle. What is the probability they all land in a semicircle? The answer uses mutual exclusivity in a surprising way.

Fantastic walkthrough and visualizations

pckt.blog A distraction-free space to write and share your story. Just you, and your words.

tell your friends about pckt.blog 😌

I'm cookin 👨‍🍳

My friends have been enjoying Badly In Love, but I am not sure if that will be the style of Japanese that you want to emulate (or maybe it is??).

hmm, an interesting search suggestion

Oof. I just read your post in the most uncharitable, pedantic way possible, and I gotta tell ya: it's not looking good for you.

economics notches another W

One JR reservation site is sound asleep and the other is just a little groggy.

Get some rest my friends and I will check back later.

I always thought it was "picked" just like spelled weirdly phonetically.

And it was supposed to be something like "I picked this to blog about" 😬

How kind of them to remind us.

I will never not think it is funny that sanseito uses so much English and so many loan words in their political messaging.

Human interaction is going to shift to discords and group chats, invite-only. The open web and social media are going to be left for the agents lurking amongst the ruins. Everything public will be Moltbook.

It hasn’t fully hit BlueSky yet, but LinkedIn and X are just meaning-shaped comments by LLMs

We are Changing our Developer Productivity Experiment Design Our second developer productivity study faces selection effects from wider AI adoption, prompting us to redesign our approach.

aaaaaah updated METR study updated METR study

update on the new version of the study that people wrongly cite as proof that LLMs slow developers down. they're having trouble because it's impossible to find participants willing to work without AI for the comparison lmao

How much does distillation really matter for Chinese LLMs? Reacting to Anthropic's post on "distillation attacks."

How much does distillation really matter for Chinese LLMs?

DeepSeek's usage was a rounding error. MiniMax's was substantial. But distillation is getting less important as RL takes over — it's easier to access "banned" APIs than to smuggle GPUs.

www.interconnects.ai/p/how-much-d...

The Coffee Log is now live! I'm abusing @arabica.social's description field for some additional semi-structured data. So far, very nice!

(Might need to tweak that background at some point)

*On mobile, it is just greyed out.

Is there a good Bay Area ML/AI events tracker I can follow?

Tired of learning about events too late to sign up.

Extremely minor spoiler ahead.

I seem to remember discord image spoilers being just greyed out, but now it looks like they just heavily blur the image. This is slightly leaky for enclose.horse! You can make out the shape of the enclosed area just enough to know if your solution is close.

Missed opportunity for an apples to oranges joke.