813 Photos and videos
« Including foreign nationals anthropic employees »
🤯😳
Guys are working on some stuff and are restricted from using it.
Jun 13
The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees.
The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance.
Access to all other Claude models is not affected.
We apologize for this disruption to our customers. We believe this is a misunderstanding and are working to restore access as soon as possible.
Read our full statement: anthropic.com/news/fable-myt…
11
C’est simple. S’il faut absorber la gabegie des générations actuellement à la retraite, tout en adoucissant celle qui pourrait être créée par notre propre génération, et en ne laissant pas un champ de ruine pour nos enfants, ça sera très douloureux.
Chacun doit y mettre du sien.
Il faudrait repousser l'âge légal de départ à presque... 68 ans: le Conseil d'orientation des retraites revoit en forte hausse le déficit du régime pour 2070
l.bfmtv.com/mYRq
1
57
Al_th retweeted
Jun 5
Massive output uptick due to agentic AI. Complete flat adoption.
461
971
7,296
2,349,024
YES !
Jun 5
llama.cpp is likely the first LLM runtime in the world to allow "interrupt" reasoning without stopping the whole response.
We also added a small "skip" button on the Web UI, the model gives the final response as soon as you click the button.
The response is no longer bound to reasoning budget!
20
Al_th retweeted
Jun 2
first most striking thing is how much the model gains over the RL process, and also interesting is that the x axis is labeled, and it's not really many thousands of steps as one might expect. starting from a non-reasoning baseline makes the gains more understandable
2
3
96
15,753
Fully agree.
My memory is a bit fuzzy in general. I remember concept, I remember that stuff exist, but either it's hidden in my memory, or I don't recall the details.
LLMs have been a tremendous help in the recent years as a fuzzy search engine in specialized topics.
May 28
After AlphaGo, the skill of human Go players noticeably improved. I suspect we will see a similar pattern in math.
1
44
They are also quite good at being a non-judging, ever-present counterpart in thinking.
Obviously, the answers are sometime plain wrong, and need to be double checked, but it's like a mix between a mentor and a student: it knows a lot, but you need to keep your feet on the ground
25
youtube.com/watch?v=7A3uX8tc…
Au moins, c'est clair et transparent:
"Là on peut restructurer leur patrimoine immobilier pour l'encapsuler dans des structures à l'impôt sur les sociétés"
"Oui, l'impôt sur les sociétés est assez recherché"
24
Huge blog post. Both in density of information, and quality of hindsight.
👏🏻
May 6
My first blog post in over a year is a deep dive on flow maps🗺️, or how to learn the integral of a diffusion model to enable faster sampling and several other cool tricks.
It's the longest one yet👀 Let me know what you think!
sander.ai/2026/05/06/flow-ma…
1
40
Al_th retweeted
🔴 Quatre bonnes raisons pour réfléchir à un plafonnement des retraites les plus élevées ⤵️
1️⃣ 8 200 €/mois : la pension mensuelle moyenne du top 1% des retraités fonctionnaires d'État
2️⃣ Ce montant est de 5 900 €/mois dans le privé
3️⃣ 75% des revalorisations de retraite sont épargnées (99% dans le cas des retraités aisés)
4️⃣ Cette épargne gonfle les héritages… et est financée par la dette publique laissée à nos enfants
Plutôt qu’assurer une retraite digne, le système par répartition assure désormais une contre-redistribution au bénéfice des héritiers des classe supérieures et au détriment des travailleurs modestes.
🔴 Il faut plafonner les retraites pour réduire leur déficit et rendre leurs cotisations aux actifs
178
334
1,151
200,637
Al_th retweeted
Arrêtez avec les "ouh la la, le pirate n'avait que 15 ans, c'est flippant". Ce qui est flippant c'est qu'un ado de 15 ans défonce en 3 coups de cuiller un site d'administration en charge des papiers d'identité, qui devrait figurer parmi les plus sécurisés du pays. Et ce...
81
1,670
15,788
231,113
Al_th retweeted
On va pas m'accuser d'être de gauche ou anti-militariste, mais c'est un précédent catastrophique.
L'université n'a pas à accorder des points bonus pour du sport, un engagement associatif, un engagement militaire, ou que sais-je d'autre comme conneries.
Scandaleux. Paris 1 valide un partenariat entre le Collège de droit et la gendarmerie prévoyant des points bonus pour les étudiants réservistes. Vu le niveau de sélection, la direction envoie un message limpide : pour valider son semestre, il vaut mieux s'enrôler dans l'armée.
15
9
51
6,335
Al_th retweeted
Apr 2
Is autoresearch really better than classic hyperparameter tuning?
We did experiments comparing Optuna & autoresearch.
Autoresearch converges faster, is more cost-efficient, and even generalizes better: 🧵(1/6)
26
115
1,307
136,068
Al_th retweeted
Mar 17
NeurFrame: Learning Continuous Frame Fields for Structured Mesh Generation
Xiaoyang Yu, Canjia Huang, Zhonggui Chen, Juan Cao
arxiv.org/abs/2603.12820
Abstract:
Structured meshes, composed of quadrilateral elements in 2D and hexahedral elements in 3D, are widely used in industrial applications and engineering simulations due to their regularity and superior accuracy in finite element analysis. Generating high-quality structured meshes, however, remains challenging, especially for complex geometries and singularities. Field-guided approaches, which construct cross fields in 2D and frame fields in 3D to encode element orientation, are promising but are typically defined on discrete meshes, limiting continuity and computational efficiency. To address these challenges, we introduce \emph{NeurFrame}, a neural framework that represents frame fields continuously over the domain, supporting infinite-resolution evaluation. Trained in a self-supervised manner on discrete mesh samples, NeurFrame produces smooth, high-quality frame fields without relying on dense tetrahedral discretizations. The resulting fields simultaneously guide high-quality quadrilateral surface meshes and hexahedral volumetric meshes, with fewer and better-distributed singularities. By using a single network, NeurFrame also achieves lower computational cost compared to prior self-supervised neural methods that jointly optimize multiple fields.
14
103
5,329