Conjecture Institute Fellow Paul Raymond-Robichaud proved any physical theory (including quantum mechanics) that satisfies extremely general assumptions must have a local-realistic formulation. 👇

Ask Ethan: How long does it take planets to form? Just 35 years ago, we knew so little about how planets formed, or how long the process took. Here in 2026, we now have a complete end-to-end story: validated by actual data. bigthink.com/starts-with-a-…

Yes, good summary of the main argument.

Extremely important reminder

Comment peut-on baisser le niveau du débat à ce point ? Zombie or not zombie ? Telle est la question.

Yes, repeating a poor argument does not make it better.

Of course I agree. This is the current view shared by most experts.

Yes, I think the same.

He showed somehow that no stable matter can exist inside the Schwarzschild radius (which is true). For most physicists of the time, the concept of implosion was not a topic.

J'ai discuté d'IA avec des gens très sympa, et en plus il y avait des caméras et des micros. Du coup ça fait un podcast : youtu.be/eAdI61g1bFs

No we mustn’t

Tout à fait d'accord et on peut aussi ajouter l'apprentissage frugal où les progrès vont bon train, donc une IA capable de trouver des associations non explicites dans les données de départ, avec un minimum d'apprentissage supervisé.

C'est effectivement n'importe quoi et ne correspond pas à notre connaissance actuelle de l'IA.

L'un des poncifs sur l'IA (l'analogie du grille-pain) ressassé à l'envi.

C'était un bon point de départ, encore fallait-il s'y tenir. Bon, on sait que "l'humain croit d'abord ce qu'il désire être vrai."

Playing around with more chaotic pendulum simulations. Code generated by Claude Sonnet 4.6.

Judging AGI by how well it can mimic us is a category error, because mimicry isn't intelligence and isn't general. We should judge AGI by how well it learns to do things we didn't teach it (including things we don't know how to do ourselves).

Yes and such claims should be accompanied by the exact list of the relevant benchmarks that are missing in order to check understanding.

Yes!

GPT 5.5 nearly hitting 80% in computer use while humans sit at 72% is absolutely wild. 🤯 We are watching white collar work get cooked in real time.