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.

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? 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 - we've seen drugs lose approved indications after they've hit the market, or even be withdrawn entirely, when longer term evaluations have revealed serious safety / mortality concerns. 3.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?

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.

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.

A proposal: 1) Every paper has a repo, from the first idea onward 2) Maintained by the authors and their AI 3) Submitted with the paper, memory.md and history.md files included The referees would see what was tried, what failed, and why.

Many economists forget that the first-order problem of an aging society, one with high life expectancy and low fertility, is how to redistribute income from workers to retirees. How fast output grows is, surprisingly enough, the lesser question. If you have roughly as many retirees as workers, what each worker produces must be split in two, one part for the worker, one part for the retiree. At a basic level, it does not matter whether we do it through taxes, as in a pay-as-you-go social security system, or through capital income, as in a fully funded one. Workers will not be happy to see half of their income taken away, whatever the absolute level of that income. This redistribution will poison the political system and lead, with high probability, to dysfunctional policies. We are already seeing it across Europe in the fights over retirement benefits. And much of the American unhappiness about the profits of large corporations, largely owned by the old, is the same conflict, only in the fetishized form of equity ownership. Claims of the form “technological progress (e.g., advances in AI) will fix the problems brought by aging” miss the political economy of the situation. Aging is about politics, not TFP.