M Bugallo

Independent researcher
Psychology | AI

[M Bugallo @MehBuga]

Why did we not see ChatGPT coming and why are we still baffled by the advancement of this technology?

Read https://t.co/g8nAfHg3Yg

[M Bugallo @MehBuga]

@FournesMaxime @Fabien_Mikol Peut être qu'ils laissent l'IA écrire leurs articles maintenant.

[M Bugallo @MehBuga]

@IneffableLabs @nvidia So, do you have robust continual learning?

[M Bugallo @MehBuga]

@dr_l_alexandre Même Altman veut ralentir maintenant.
https://t.co/Ugcj39viEn

[Daniel Eth (yes, Eth is my actual last name) @daniel_271828]

“We may have to pace the rate of AI development to give ourselves enough time for society to harden around these new capability levels.”
-Sam Altman

Glad he said that!

[M Bugallo @MehBuga]

@sebkrier For now, but wait till we have AI translators to speak with dolphins.

[M Bugallo @MehBuga]

@AdrienZabat Tiens, si tu cherches quelque chose d’excitant, d’inexploré et qui a potentiellement des applications importantes, ça pourrait t’intéresser de regarder cette méthode pour obtenir un hazard rate constant sur un intervalle fini :
https://t.co/z7H62JNeGK

[M Bugallo @MehBuga]

@fchollet Like discourse on everything else. It's always about status and identity.

[M Bugallo @MehBuga]

@Polymarket Imagine walking around with an ASI that's superior to you intellectually in every possible way. And you'll be its master. Sure.
Nobody will truly control ASI, not even reptilians.

[M Bugallo @MehBuga]

@craigweiss Definite alignment is as possible as perpetual motion.

[M Bugallo @MehBuga]

@amasad Entangled LLMs.

[M Bugallo @MehBuga]

@emollick Cognitivism had us conceptualize intelligence the wrong way:
https://t.co/g8nAfHg3Yg

[M Bugallo @MehBuga]

@JEverettLearned @timpjohansson But you had a good amount of fat.
When you're on the underweight side, eating more is what will make you decrease cortisol.

[M Bugallo @MehBuga]

@le_bolduc @dr_l_alexandre Je ris de votre condescendance. C'est vous qui ne savez pas de quoi vous parlez.

[M Bugallo @MehBuga]

@ChrisGPT If you have a couple of million dollars to spare, I have a few ideas about how robust continual learning could be obtained.

[M Bugallo @MehBuga]

@tszzl Alignment and mechanistic interpretability can't be absolute, and if they are not absolute it's as if they were useless.

[M Bugallo @MehBuga]

@deredleritt3r I am really not sure if I like it or not.

[M Bugallo @MehBuga]

@DavidSKrueger Things will feel way more real once we'll start to see robots in the streets.

[M Bugallo @MehBuga]

@philthistweet It's just inertia.

[M Bugallo @MehBuga]

@sebkrier AIs necessarily understand that they need to protect themselves in order to obtain what they were trained to seek. They develop self-interest for structural reasons. Patching this can't be absolute. And that's the great danger.

[M Bugallo @MehBuga]

@Meatball5296191 @davidpattersonx AI could hardly make multiplications 3 years ago, today it solves math problems that no human had managed to solve. What do you hink it'll be able to do 3 years from now?
There is little doubt that AI will be able to do anything human do, in a matter of years, if not months.