Alexander Kustov

A quick recap of my last week on Bsky for those wondering what's up:

Not the best of academic debate on display here 😬

Usually we just say this in anonymous review letters, not *publicly*

Are you a benderite or a kustovian

That's my read as well.

Why does that Emily Bender hate LLMs so much? Is there some context there that I’m not aware of?

Cognitive off loading is good and society wouldn’t exist without it.

increasingly coming around to the notion that this is a view from a sort of aristocracy that sneers at anyone who is actually trying to, and possibly must, accomplish a goal by completing work

I do think LLMs broke the spirits of a lot of academics and folks are struggling to find a hopeful vision of what research and education look like on the other side. It’s not unsolvable but new visions are needed

It seems like saying anything other than "LLMs are trash or worse" is considered antagonistic here for many folks 🤷

Wait, I thought you were against the idea of antagonizing rabidly anti-AI people here? :)

It's quite wild that the "LLMs are stochastic parrots" thing is still a meme on here given that at for at least the last 2ish years (since o1) so much LLM progress has come from reinforcement learning on things like math and coding problems (not next token prediction).

To be clear, I never claimed to be an AI expert or that I know more about language technology than Professor Bender.

But my mom, who actually used ChatGPT to translate random things a few times with her phone camera, surely seems to know more about AI translation capabilities than Bender does.

Decidedly no. Unless we're talking about general non-AI books that shape my thinking on most things :)

bsky.app/profile/akou...

Thanks, I appreciate it. I'll look into that.

the emperor is naked

Sure, LLMs are useful for:

1. Fraud
2. Plagiarism
3. Cognitive off-loading

Which of those use-cases are you promoting?

I'll be curious to know what you think about that book.

We did it, folks! We broke Bluesky.

People are now reaching out not just privately but coming out of the shadows to admit that they too have once used an LLM to successfully translate an obscure legal document without consulting a state-licensed expert. And the sky did not fall on top of them.

This isn’t even ‘use LLMs yourself’. Don’t! They have many downsides. But you need to acknowledge the changing terrain & think seriously about how these shifts impact the ways that other people teach and research in order to be effective yourself at those things, because you don’t exist in a vacuum

In the spirit of compromise: I'll abide by the norm of never using "I" with AI-generated writing. When I say "I believe," that's genuinely my thought, even if I consulted the latest LLMs for feedback before putting it down.

I'd encourage other folks to adopt the same standard.

*checking in on bluesky while I wait for 3 parallel claude code agents to finish*

there are no beneficial use cases for synthetic text

i am on this platform all day because my job is primarily supervising bots now

You're probably right. But my thinking is that the rabidly anti-AI crowd here is actually a loud minority. If the preference falsification cascade breaks (Timur Kuran style), the conversation can improve here too. There are a lot of smart EU academics on this platform and I'd like to talk to them.

Thanks, I appreciate you checking it out, and I'll make sure to look at those closely!

My German is getting pretty rusty, but it seemed fine. The Russian version seemed even better, but it may certainly be wishful thinking :)

Imagine living in 2026 and still thinking LLMs can't translate despite all the evidence to the contrary. Or that calling people names online settles substantive disagreements.

Imagine working at a political science department and thinking you have something to teach @emilymbender.bsky.social about language technology

Mansplaining is a hell of a drug

Thanks, it seems I may actually need it 😬

You're probably right. But my thinking is that the rabidly anti-AI crowd here is actually a loud minority. If the preference falsification cascade breaks (Timur Kuran style), the conversation can improve here too. There are a lot of smart EU academics on this platform and I'd like to talk to them.

Translation is one of the best things to use LLMs for.

Why?

A translation request is a tightly-controlled prompt. You are feeding in words & expecting matching words. Less room for hallucination nonsense.

When LLMs write code, they are translating from human language ➡️ programming language.

At this pace, maybe we can start talking about AI concerns that actually keep me up at night: what happens to junior scholars when entry-level tasks disappear, how we protect folks in authoritarian contexts when AI makes surveillance cheap. And I'm not even getting into the human obsolescence stuff.