@PinchOfData
Les principales mesures que nous prendrons pour retrouver des recettes, 106 milliards au total :
👉 Un ISF climatique : 10 milliards d'euros
👉 La suppression de la flat tax : 9 milliards d'euros
👉 La taxe Zucman : 15 milliards d'euros
👉 La taxe sur l'héritage : 20 milliards d'euros
👉 Raboter les principales niches fiscales et supprimer les exonérations : 22 milliards d'euros.View quoted post on X
I'm so surprised by the amount of arrogant dunking on Terence over this slide. for everyone going "oh he's so done, doesn't he know we still don't know how paracetamol works?", can we please extend him the intellectual generosity of considering that he might be inviting us to think about something deeper here?View quoted post on X
1. my read on the purpose of this slide is more than just "do we need to understand the mechanism of action of every ai-discovered drug that successfully passes phase 3?" because that's not a very worthwhile exercise.
2. we also know well that clinical evaluation doesn't end at phase 3! the safety knowledge of a drug continues to evolve even after a successful phase 3 and yrs on the market because even if you have a statistical improvement in progression-free survival, it takes years of additional data to understand things like overall survival benefit.
- we've seen drugs lose approved indications after they've hit the market, or even be withdrawn entirely, when longer term evaluations have revealed serious safety / mortality concerns.
3. okay so my read on the purpose of this slide/the better question:
- can a misaligned ai "knowingly" predict a treatment that successfully meets a phase 3 endpoint (like PFS) while also withholding information about a key mechanism that could cause mortality or serious toxicity a year out?
- knowingly is the key word here
4. so yes a misaligned ai could exploit a weakness in the current trial process but does that mean we need to fully understand the mechanism of action of every ai-discovered cancer drug before it reaches the market, even if it's passed a phase 3? no (patients shouldn't have to wait esp if the phase 3 evidence is strong).
5. but I think it'd be pretty stupid of us to rely on a single ai system's predictions about a drug, esp when we now have the computational tools to look at the biological effects in so many other ways. why wouldn't we use independent models (we know that great models all have the same blind spots), virtual cells, organoids etc. to uncover risks / build a better understanding of what the drug does? we've already been doing this for years now (both before phase 3 and during) so why would we stop? why wouldn't we push our understanding of long term effects of drugs much further?
Abrogation de la réforme de 2023, pas de capitalisation, 62 ans... Raphaël Glucksmann décrit sa proposition de réforme des retraites
#VousPrésidentView quoted post on X
Le journaliste @VincentCoquaz explique à des confrères (quelque peu incrédules) que oui, d'après les experts du domaine, Pangram v4 est extrêmement fiable, avec un très faible taux de faux positifs, et qu'il ne faut surtout pas le confondre avec les autres outils de détection IA.View quoted post on X
Introducing Claude Opus 5.5, the first model in our new Claude 5.5 family.
It performs at the level of Claude Fable 5.1 for most tasks, and costs 40% less to run than Opus 5.View quoted post on X
