Linux User #40306. Única persona que sigue usando &c. para abreviar etcetera.
Zaragoza, España
Joined November 2007
- Tweets111K
- Following557
- Followers1.4K
- Likes138K
Pinned Tweet
Since 2011, we have run a thread on the failure of String Theory. Never read by any of the contendants, it seems.
physicsforums.com/threads/th…
Alejandro Rivero retweeted
刚扒出 Claude Code 里 Fable 5.2 的默认系统提示词，A/居然在训练阶段亲手阉割了模型的底层算子优化能力。
提示词里写着，为了防止 AI 自我迭代加速，Fable 5.2 和只对特定白名单机构开放的 Mythos 5.2，在训练阶段专门加了限制措施，人为削弱了硬件加速器内核优化这类能直接给 AI 训练提速的能力。
重点是后面两句指令。
A/ 在提示词里叮嘱模型，碰到这类底层优化活儿的时候，别太相信自己的直觉，多去翻文档，但同时又严令模型不准因为这个原因建议用户换其他模型。
说白了就是传统机器学习和写日常业务代码完全不受影响，但只要你是搞底层高性能计算和算子开发的，模型脑子里的现成经验已经被大厂人为抽干了。
遇到卡点模型不会告诉你自己被削过，只会默默在后台一轮接一轮地搜文档烧你的配额。
Alejandro Rivero retweeted
‼️ If you use Claude Code, please do the following right now. Especially if you're thinking of fine-tuning models with your session transcripts.
in ~/.claude/settings.json, add "cleanupPeriodDays": 36500.
The default deletes sessions after 30 days. I found this the hard way.
Alejandro Rivero retweeted
Ojo al pavo de Reddit que se ha montado un cluster de N150s (40 x 12GB RAM c/u) para correr GLM sin quantizar.
$5k - 740w - 14tps (85tps prefill)
reddit.com/r/LocalLLM/commen…
Alejandro Rivero retweeted
There is a mysterious intermittent beep coming from somewhere in your room of unopened boxes. Do you:
1. Open boxes methodically until you find it
2. Vibe code software to triangulate it using two laptops, two iPads and a phone all listening at once simonwillison.net/2026/Nov/9…
Alejandro Rivero retweeted
In which I go head to head with one of the great mathematical minds of our times on AI.
Hosted by the one and only @TOEwithCurt Jaimungal.
Some of this is hard going to be sure. But this also get's closer to my real voice than you will hear on almost any other podcast. Check it out.
youtu.be/XqgnJyQZGG4?si=VwEA…
Alejandro Rivero retweeted
🐟🦀 TRAMPA casera para pescar en el río. 80 AÑOS atrapando #peces y #cangrejos con #NASAS de mimbre. #OficiosPerdidos
#Documental COMPLETO (13 minutos) de este resumen: youtu.be/NnwjUrNs8VI
mas abuelos con chatgpt. – at Saragossa, Spain
Induction instead of Feynman diagrams, L-functionals instead of Hilbert space
Albert Schwarz
arxiv.org/abs/2611.09032 [𝚑𝚎𝚙-𝚝𝚑 𝚖𝚊𝚝𝚑-𝚙𝚑]
ALT The main result of this paper is a construction of an inductive process that can replace Feynman diagrams. Combining this result with the notion of the L-functional, one can give an exposition of QED that does not suffer from infrared divergences
40
Alejandro Rivero retweeted
As I predicted, Claude is off the leash. Fable 5.2 has solved two open problems in theoretical physics, one of them completely autonomously.
Alejandro Rivero retweeted
Claude Fable 5.2 from @AnthropicAI on ARC-AGI (Verified):
- ARC-AGI-3: 97.7%, $18.6K (provider adapter harness), 81.2%, $21.9K (standard harness)
- ARC-AGI-2: 93.8%, $0.98/task
- ARC-AGI-1: 98.5%, $0.36/task
Fable 5.2 sets a new high score on ARC-AGI-3 in the standard harness.
Alejandro Rivero retweeted
Statement #3 of Yang-Mills Group, scientific collaboration shepherding AI result on existence of quantum chromodynamics & completions of the Standard Model, including subproblem proving the existence of the Quantum Yang-Mills theory. buff.ly/BwV89Tk
Alejandro Rivero retweeted
Introducing Claude Fable 5.2 and Claude Mythos 5.2, designed to tackle the most daunting technical challenges.
They’re capable collaborators on ambitious projects in many fields, and show improvements in areas like coding, reasoning, and vision.
por cierto el fundador de sanitas estuvo cenando con mi abuela en abril del 31, y lo más cerca que ella estuvo de mencionárnoslo fue un domingo que contó que habia sido "abanderada de la tuna de medicina".
Y ahora hay espacio (ya estará funcionando) para la sanidad privada premium, que en vez de 80-100€/mes te cobrará 300 pero no tendrás listas de espera. Sanitas es hoy la Seguridad Social de hace 15-20 años.
Sigamos metiendo 800.000 personas al año y fomentando turismo sanitario
67
Alejandro Rivero retweeted
BFSS matrix theory is the canonical example of holographic quantum mechanics. To characterize its ground state wavefunction has been a long standing open problem. We hereby present an approximate but highly accurate and explicit solution in the SU(2) case arxiv.org/pdf/2611.09837 github.com/asmichel/SU2-BFSS…
Alejandro Rivero retweeted
Credible reports that @OpenAI bureacratic hierarchy doesn't even give the OpenAI math team direct access to its 700 slopdrop repo. Pretty funny because (A) you're not making a nuclear bomb here that people need clearance to access lmfao these are math problems (B) incredible how much the employees are willing to be humiliated by the company just for money.
Alejandro Rivero retweeted
The Standard Model on a finite bosonic lattice with continuum Weyl fermion
v2 adds a weak-field equivalence result showing that Craig's overlap fermion weight agrees with the continuum Weyl determinant up to local counterterms. It also discusses Witten's recent eta-invariant formulation of chiral gauge theories and establishes its equivalence to the determinant-line phase prescription used here.
arxiv.org/abs/2610.29019
Alejandro Rivero retweeted
Been running Opus 5.5 agents over the notebooks of Gauss, Riemann, Hilbert and Grothendieck. around 12,000 pages so far. Heres the best finding so far:
Riemann's 1860 shock wave paper is famous for getting the energy part wrong however his unpublished sheets at Göttingen have it right the Rankine Hugoniot conditions a decade or two before Rankine and Hugoniot plus results usually dated 19061937, all in his own notation. A second sheet in another folder has the same physics.
There are probably thousands of potential undiscovered stuff in the research papers of the smartest ever to live imo all of them should be made public and reviewed by ai years of work can probably be transcribed by agents in days
Alejandro Rivero retweeted
en principio estos tiempos deberian decir algo del numero de parámetros activos y las capas – at Utebo, España
🚨 Codex users
GPT-6.1 Sol just got MUCH faster
I tested the speed of GPT-6.1 Sol after OpenAI fixed the slowdown, and the results are honestly insane
GPT-6.1 Sol is now up to ~20% faster than it was around DevDay
The model quality was already crazy now the speed is finally catching up
6.1 Sol with proper speed is a completely different experience
Alejandro Rivero retweeted
La primera pregunta: ¿aprende algo mío o solo a escribir como se escribe por WhatsApp? Comparé cuánto mejora en mis respuestas y en las de mis amigos. En las mías la perplejidad baja un 60,8 %, en las suyas un 41 %. Esa diferencia es lo que ha aprendido de mí.
