Jon Willits

Cognitive Scientist and Assistant Professor @Illinois_Alma. Studies language and semantic development and computational models.

[kasra @kasratweets]

this is an interesting point in the new ted chiang piece – no one really claims that alphafold is conscious, or that sora or midjourney or dall-e are conscious

[Jon Willits @jonwillits]

@AntigoneJournal No. There will be a synthesis route as well. And yes, we will at first be hated by both sides, like those who attempt synthesis usually are. But if the synthesis is as valuable as I think it will be, it will win out in the end.

[kasra @kasratweets]

now that AI makes information consumption and transformation easier than ever I would like to bring back this old banger by Sasha Chapin about how books are not information transfer devices but subjectivity-merging devices

in fact I would say content consumption in general is more about subjectivity-merging than information transfer, which is why I am generally much more interested in writing by humans than by AI

[Jon Willits @jonwillits]

@ylecun @francoisfleuret It's about primary goal. Prediction/Understanding vs. Control. To control you usually need to understand, and to understand, you usually need to be able to control. But it's about, for the agent, which is the means and which is the ends.

[Jon Willits @jonwillits]

@francoisfleuret It's about primary goal. Prediction/Understanding vs. Control. To control you usually need to understand, and to understand, you usually need to be able to control. But it's about, for the agent, which is the means and which is the ends.

[Jon Willits @jonwillits]

@dearmadisonblue @danwilliamsphil level of analysis issue. many similar and true things you can say about our cells (they are just doing X) but which do not detract from emergent capacities of a complex system built of those parts. I am not arguing LLMs have such capacities. But it’s an independent question.

[Brendan Nyhan (@BrendanNyhan on 🟦☁️) @BrendanNyhan]

GPT reviewer "scores above each paper's top-rated human reviewer" but AI review agents "overlap far more than humans do...and exhibit 16 recurring weaknesses humans do not share..." Results "position current AI reviewers as complements to, not substitutes for, human reviewers."

[eigenrobot @eigenrobot]

asking people to read ai-generated text is offensive.

this is not because ai text is intrinsically bad. rather, the author has not paid a cost to write the text himself. this cost is a credible signal he finds its communication important.

so: not paying that cost is telling

[Alexander Kustov @akoustov]

Let me elaborate on why I think imposing minor but tangible penalties for hallucinations on a preprint server is appropriate (one-year bans in any single outlet are really not a huge deal).

As you all know, I've been an "AI booster" writing on human slop before it was cool and getting flak for it. But journal editors really don't have a good option now. No matter what you do, you lose. Explicitly imposing some cost on careless submissions is a good idea.

Should we have done it earlier for similar and more serious human errors in data analysis? Sure. But as many pointed out, the great thing about AI is that it gives us a good excuse to improve things in conservative academic institutions resistant to any change.

My hope is that in equilibrium, when everyone is aware of the one-year ban rule, people will be less inclined to submit clear AI slop, or human slop for that matter. Or at the very least, they will go over their references and double-check everything. But if more outlets implement this rule and things go insane, I'm totally willing to change my mind.

[Anna Riedl @AnnaLeptikon]

Here are 14 images to better understand cognitive science visually 🧵

(From a lecture I gave for @pgmid's BrainInspired discord community)

[Jon Willits @jonwillits]

@Plinz @ihtesham2005 By replying, you’re boosting it.

[Jon Willits @jonwillits]

@Plinz @simonmaechling Be Bayesian about it. In the counterfactual world where scientists were more conservative and careful, isn’t it possible that partisan attacks from the right over places they didn’t like the science presented neutrally (like climate) would have had the same collapse in trust?

[Jon Willits @jonwillits]

@AdamZivo First have to resolve the debate about what university ed is for. If job prep is part of it, then excluding AI completely is a non starter. But if it’s traditional liberal arts “how to think and be a good citizen” then left leaning academia needs to figure out how to fund it

[Jon Willits @jonwillits]

@lakens Best practices have not permanently plateaued.

[Jon Willits @jonwillits]

@HowToAI_ Without someone taking responsibility for malicious code, this is a disaster waiting to happen

[Helen Devine @helenldevine]

psych post-bacc jobs are scattered across wikis, twitter, and random lab sites. hard to find if you're not already in the field.

I started a slack to pool them in one place — a low-effort bulletin board, channels by subfield, updated live!

🔗⬇️

[Jon Willits @jonwillits]

@dioscuri @rgblong Does it always respond that way, even if you were to ask it how it feels about being told to be a “helpful assistant”?

[Jon Willits @jonwillits]

@Morsker @NathanpmYoung I took as part of the thought experiment that you could tell them, but the challenge would be how to get them to believe you weren’t making it up

[Jon Willits @jonwillits]

@PajosTM Um, compile time is minuscule compared to run time. Just be honest, they don’t want to learn C.