Krsto Proroković

[Krsto Proroković @krstopro]

@MelMitchell1 The same Michael Jordan predicted in 2014 that the big-data winter would arrive within 2-5 years.
https://t.co/ORbo2eIuMR

[Krsto Proroković @krstopro]

@cilibrar @SMT_Solvers @PassRafael Have you seen
https://t.co/wscKTiRkMM

[Krsto Proroković @krstopro]

@TokenGremlin Not to mention RSI tweets that got deleted
https://t.co/yUpdwgiKQf

[senator 👑 ☼𓃬 @Senator1269]

@MrZiruiWang From "RSI is here" to "reasonably good". That’s the fastest model degradation in AI history.

[Krsto Proroković @krstopro]

@industriaalist Thank you! It's a work in progress and I'm about to improve the approach.
That said, any feedback is welcome, be it on clarity, methodology, or missing related work.

[Krsto Proroković @krstopro]

@stanfordNYC @SarahTheHaider And now Pinker limited who can respond to his tweets. Insane.

[Krsto Proroković @krstopro]

@mustafasuleyman Meanwhile
https://t.co/YpWV1fEPI6

[Krsto Proroković @krstopro]

@SarahTheHaider I am not saying you should listen to Scott Alexander, but I don't see a reason why anyone should listen to someone that declares in 2019 that deep learning has peaked, doesn't acknowledge that they are wrong, and now limits who can respond to their tweets.

[Krsto Proroković @krstopro]

@NathanpmYoung And now he limited who can respond to his tweets. 😆

[Krsto Proroković @krstopro]

@AlecStapp Steven Pinker limited who can respond to his tweets. If this isn't enough for people not to listen to him, then I don't know what is.

[Krsto Proroković @krstopro]

@DimitrisPapail Given how quickly the AI is progressing, I wouldn't be surprised that this changes soon.

[Krsto Proroković @krstopro]

@GaryMarcus @IonelChiosa @altryne By chain of though traces, I mean
https://t.co/hdnvTdSj2e

[Krsto Proroković @krstopro]

@GaryMarcus @IonelChiosa @altryne Then there is no reason to claim that millions of potential solutions were considered until one of them had a Lean formalisation. Looking at the chain of thought traces, most likely that the informal proof was obtained first, then formalised in Lean.

[Krsto Proroković @krstopro]

@IonelChiosa @GaryMarcus @altryne Most likely the proofs were formalised in Lean once the agents discovered informal versions. Either way, there is no evidence for Gary to claim that millions of proofs were tried until one was proven correct in Lean.

[Krsto Proroković @krstopro]

@pmddomingos What makes you think they don't suffice for automating the process of discovering models that have these properties? Or at least make it massively faster compared to human researchers discovering them by themselves?

[Krsto Proroković @krstopro]

@slatestarcodex @pmddomingos And LLMs might be enough to discover models that have these properties at inference time, i.e. automate the research process.

[Krsto Proroković @krstopro]

@GaryMarcus Yeah, sure
https://t.co/hicDlKv9Ba

[Gary Marcus @GaryMarcus]

I just wrote a great piece for WIRED predicting that the AI bubble will in collapse in 2025, and now I wish I hadn’t.

Clearly, I got the year wrong. It’s going to be days or weeks from now, not months.

[Krsto Proroković @krstopro]

@TokenGremlin Your tweets come as an antidote to hype created by accounts such as Chubby, lyraxana, etc.

[Krsto Proroković @krstopro]

@MrZiruiWang Interesting how it went from "RSI" to "I think it is reasonably good to use" 😄

[Krsto Proroković @krstopro]

@TomDavidsonX Same renowned academic declared in 2019 that deep learning has probably peaked https://t.co/HtZYxQK8qL

[Steven Pinker @sapinker]

“Deep learning” in AI (which in fact is quite shallow) has probably peaked. https://t.co/QafeR3nDbR