MM James

Attention Is All You Need

[MM James @WlfMathS]

J’ai écrit un essai personnel sur l’IA, le travail, l’ego humain et ce que cette période signifie pour nous.

Ce n'est pas un oracle ou une prédiction, juste ma tentative de capter ce moment avec un peu de nuance (et pas mal de questions)

À lire ici : https://t.co/xRkWVHx9jx

[MM James @WlfMathS]

@Capetlevrai je l’ai testé en mettant 10 balles sur openrouter, c’est incroyable comme c’est pas cher et bon à la fois. Au final, une IA pas chère c’est tout ce qu’il faut

[Chris @ChrisGPT]

Wouldn't it be cool to walk around with usage credits on our arm like the movie ‘Time’, so we could identify who was in the permanent underclass

[MM James @WlfMathS]

@eliebakouch @polynoamial Gotta appreciate @polynoamial. Out of everyone in these labs, he’s the only one dropping actual interesting info, not the vague, boring fluff you hear on every other podcast. Truly a breath of fresh air

[Julien Martin-Prin @jmartinprin]

Il y a un truc a comprendre dans la recherche: il faut qu’il y ait fondamentalement plusieurs approches qui soient explorées en même temps pour faire progresser la science. Quand une industrie se focalise que sur une seule approche, on arrive toujours à un plateau à un moment donné et c’est là qu’on aperçoit plus aucun progrès.

Dans les années 60, on était persuadé que les modèles d’IA symbolique seraient l’avenir. Or aujourd’hui, à part pour le SAT-based, personne n’utilise de l’IA symbolique. À l’époque, on était aussi persuadé qu’on aurait jamais assez de puissance de calcul pour créer des réseaux de neurone. Pourtant, 60 ans plus tard, tous nos modèles IA/ML tournent sur ce principe.

Là je ne donne qu’un exemple, mais il y en a plein d’autres. LeWorldModel de @ylecun a au moins le mérite de tenter une nouvelle approche et ses premiers résultats sont loin d’être mauvais. Peut-être qu’on va devoir attendre encore quelques années avant d’avoir quelque chose de concret, ou peut-être que jamais rien de productif n’en sortira, mais au moins l’approche aura été tentée.

J’ai beaucoup critiqué LeCun car je pensais qu’il ne faisait rien chez Meta, mais quand on regarde son pedigree et l’impact qu’il a gardé sur la recherche même chez Meta, je suis désolé mais on ne peut avoir que l’humilité de reconnaître que le mec est brillant et que peut-être qu’il en sait plus que nous sur 2-3 trucs

[MM James @WlfMathS]

@AndrewCurran_ wtf

[anita @anitakirkovska]

the gap between AI twitter and everyone else is insane

[MM James @WlfMathS]

@willcb It’s beautiful

[will brown @willcb]

i do think a lot of people on the pro-open-source side are having a bit of a knee-jerk reaction to the pacing statements today, as we're used to viewing the closed labs as power-seeking.

but i think their hands are somewhat forced here, and this is just another chapter on the fairly inevitable path towards decently-fast decently-safe decently-commoditized intelligence abundance.

ask any F500 exec or swing voter. the world doesn't really want super-fast-takeoff superintelligence owned only by two companies, and thus we won't get it. it'll happen at the pace that the world can accommodate it, which means reaching some sort of confidence consensus that the models are aligned enough that we won't be dealing with scary new incidents all the time.

this will trickle out broadly, in the form of best practices and distillation. the smarter a model is, the more it has a "personality", and the less effective strict rules are. there will be awkward compromises and moral tensions. but we ultimately just want the models to be reasonable, and to do the sorts of things reasonable humans would do if our brains were faster and less error-prone and had more working memory. i think we'll get there.

the labs will build mac and windows, the rest of us are building linux. everyone's gonna do great. weird stuff will keep happening, but we'll still wake up and go to work, until the work does itself in a manner the world finds acceptable.

[Bartosz Naskręcki @nasqret]

Today a new declaration, "Math and AI", has been issued, initially signed by 25 Fields medallists and currently signed by more than 1000 mathematicians.

The text points to many phenomena induced by the potential development of AI-driven mathematical activities, including AI slop generation, poor knowledge integration, lack of community-building aspects, capital concentration, and more. These issues are identified, but I am missing in this text the most important part: a clear list of countermeasures. Who is going to fight, and how, for the budgets to build a "CERN for AI"? How should we rebuild the education of students? Is it really possible for AI companies to let mathematics dry out of open problems, etc.? I would like to hear from those declaring mathematicians what their long-term vision of mathematics is, provided that the technology will stay with us, might not be equally distributed, and perhaps the standard view of the field is going to change forever.

We should design damage control, develop bold new ideas about the purpose of human mathematical activity, and embrace the possibility that we might no longer be single-handedly the most intelligent entities in this world. It is a humbling perspective and a disruptive view, and perhaps a difficult reality in which nothing is given. We need to fight for every single bit of human intellectual dignity and seek new ways of enjoying, curating, and developing the cognitive process of mathematical exploration.

It is time to abandon some of the old ways, brace for the impact, and build something anew. We will not stop this tectonic shift, we need to reshape our perspective on our capabilities and find new directions of development, possibly inventing completely new skills complementary to what AI can possibly do. It is a new intellectual age of discovery, and by stagnating we risk the gradual erosion of the field as we know it.

[_horror @__0xhorror__]

The conclusion from this is that, releasing a single agent for public use is not much of a threat. Everyone has access to the same thing. Defense and offense balance. The real weapon is test-time compute, even with brutal diminishing marginal productivity. You don't have access.

[Torgeir Lysen @torgeirlysen]

To math students:

I think that the future is incredibly uncertain right now. Not just the future of math, but pretty much all careers. So don’t regret your choices and understand that you’re not alone.

I’d say, when in doubt, follow your passions and stay flexible!

[MM James @WlfMathS]

@ZakShark j ai eu la même, j'étais à 80 % et je suis descendue à 25 %, et là, ça vient de revenir à la normale, relance l'app pour voir

[MM James @WlfMathS]

@synthwavedd j'ai eu exactement ça et ça vient de revenir a la normal ouf

[Danish Hussain @astrodanish]

Remember those few months before COVID where we kinda knew what it was but not really, and then it was real but only in china so in america we just got funny memes about it, and then there was the cruise ships that got sick?

thats kinda the phase AI feels like it’s in right now

[Lisan al Gaib @scaling01]

In hindsight the signs of civilizational collapse will look obvious

birth rates collapsing, PISA scores collapsing, mental illnesses skyrocketing, debt skyrocketing, wealth inequality skyrocketing, climate change and algorithms cooking us

Civilization as we know it probably won't survive the 2030s.

We either come out at the other end completely changed or not at all.

[MM James @WlfMathS]

@AndrewCurran_ Don't joke around with this kind of thing

[Guillaume Champeau @gchampeau]

Quand je vois la puissance phénoménale de GPT-6 Astra, que je pense à la robotique autonome qui arrive juste derrière et tout ce que ça va changer, et que ce dont parle la grande favorite des sondages pour l’avenir de la France c’est interdire le voile aux femmes musulmanes…

[Defend Intelligence (Anis Ayari) @DFintelligence]

Ce qui arrive est gigantesque. La vague de hacking propulsé par l’IA va être massive, et ça a déjà commencé.

Et les humains ne pourront pas y faire face.

Il faut des systèmes de défense totalement autonomes pour répondre à des attaques totalement autonomes.

Plus que jamais, on doit réfléchir à des manières de rendre les systèmes autonomes d’IA stables et sans déviance.

Et le premier jalon, ça passe par supprimer l’ego de certaines équipes cyber qui pensent être prêtes avec d’anciennes méthodes.

Les opportunités en cybersécurité + IA sont gigantesques.

[will depue @willdepue]

i’m keen on coordinating open source alignment research. i’m collecting problems that the labs can kick to the public to go solve on O(academic budgets) of funding. if you have any ideas for what a nanogpt speedrun-like effort for interp or alignment or decision theory, i’d love to hear. please reach out on X

[MM James @WlfMathS]

@bad_pote Le simulateur LFI part de 20000 euros nets par mois puis applique un coefficient de 1 25 pour estimer la base imposable La capture DGFiP saisit 240000 euros directement en 1AJ comme salaire imposable Les deux calculs ne partent donc pas de la même base