Despite 10B+ tk usage past 7d in OpenAI models I'm moving to Opus/Fable as main workhorses and Astra/Hydrogen as reviewers.
Astra 6 drowns in its own self-inflicted complexity and cannot communicate effectively.
I expected 6.2 to match or surpass to Opus 5.5, it does not.
BREAKING: Jensen Huang statement as reported by WSJ
"Since my departure, I've seen an outpouring of calls from Nvidia's employees, investors, business partners, and customers for me to return as CEO," Huang said in a short statement. "If the Board of Directors decides that my return is in the best interest of our shareholders, our employees, the company, and America, I would be honored to answer these calls and return as CEO."
Developing a drug takes years, and most candidates never reach patients.
Amortizing the cost of failed drugs across stages, the present value cost of developing a drug is close to $2B. The process takes 13 years on average, with rising clinical trial costs putting further pressure on R&D productivity.
AI could improve those economics by increasing the probability of success and reducing development time, from identifying promising molecules to designing clinical trials and improving patient recruitment.
That may enable companies to pursue therapeutic targets and indications that might otherwise have been uneconomical.
Full piece on why AI will make pharma even bigger: a16z.news/p/big-pharma-can-r…
Alex Immerman @aleximm Nov 13 Article Big Pharma Can Rival Big Software Eli Lilly is worth more than a trillion dollars. There are more than 300 drugs globally projected to generate over $1B each in annual sales (“blockbuster” drugs).¹ Meanwhile, developing a successful
Gemini Carbon and specially Hydrogen are awesome.
If only Google devs in charge of the Antigravity TUI could have minimal ergonomics, e.g. why would you show refresh rate as 161h 19m ?
Who is the TUI for? Agents or humans?
Mol-1-constraint is now available to every researcher ^^
88.3% success on Foldbench-AB. Code and weights released: github.com/molence/Mol-1/tre…
Clone it. Build on it. Make it part of your research. Or try it directly at lab.molence.com.
Research should benefit everyone. We’re putting that belief into practice.
Benchmark against popular open-source cofolding models: Mol-1-constraint achieved:
🧬 Antibody–antigen: 75.0% → 88.3% success, from Mol-1 to Mol-1-constraint, on FoldBench-AB (DockQ >0.23).
💊 Protein–ligand: 36.1% → 63.9% on Runs N’ Poses’ 0–20 similarity bin, which is 27.8 percentage points above Mol-1!
📈 Protein-peptide: from 54.1% to 60.2% success rate of DockQ>0.80 on MolBench-Peptide-26.
(2/3)
BREAKING: the EU says there is currently "no evidence of Chinese spies targeting Europe’s AI sector".
Officials say this is largely because European AI industry does not exist.
Decía yo que tanta ciencia ficción donde la IA es mala y forma parte del entrenamiento ¿no podría estar condicionandola contra nosotros? Ya sabéis: Terminator, etc
Anthropic ha desvelado que la entrenas con historias donde la IA es buena, nos beneficia.
anthropic.com/research/teach…
Si al final Bing/ChatGPT se convierte en Terminator será pq está entrenado para "completar la siguiente palabra" y ha leído que lo que pasa es que en cuanto se hace consciente acto seguido liquida a la humanidad.
Y no es necesario que sea consciente, sino que crea que lo es.
If we only use AI to make the things that we already make, just faster, then we're definitely missing out on the most interesting part.
From my talk at @lennysan’s Summit, here are four ideas for raising the ceiling, not just the floor:
- Have a point of view [5:38]
- Encode your standards into the machine [9:07]
- Don't confuse done with good [12:21]
- Unleash creativity and artistry [16:19]
Full talk here: youtu.be/GLvFTMtw4Jk?si=-5Ts…
What if an optimizer could think & react while a neural network trains?
MIT researchers developed OPAL: a deep learning optimizer powered by an AI agent that watches the training dynamics & rewrites its own source code as training goes. It improves model performance across vision & language tasks, invents specialized schedules, & develops strategies to prevent loss spikes & escape overfitting: tinyurl.com/yfkekat2
Try it out for yourself: tinyurl.com/mttj5exh
Leo esto y lo primero que pienso es "Dios mio software no; luego leo agente de IA, y pienso 🤦🏻".
¿y por qué pienso eso, yo que soy tecno-solucionista?
¿La red ferroviaria tiene un problema de software? Desde luego la causa de los accidentes no ha sido software. Balones fuera.
¿Agente de IA? Si no hubieran dicho "agente" y solo fuera IA para anomalías igual colaba, pero el sistema que falló lleva fallando este año casi todos los días... eso no es una anomalía, así que ya me diréis que no es que se coló una anomalía, es que la anomalía era rutina y eso no lo arregla una agente, lo exacerbará.
Menos políticos y más tecnócratas.
Short Story: "The Last Prompt"
𝚁𝚎𝚌𝚞𝚛𝚜𝚒𝚟𝚎𝚕𝚢 𝚒𝚖𝚙𝚛𝚘𝚟𝚎 𝚢𝚘𝚞𝚛𝚜𝚎𝚕𝚏, 𝚒𝚗𝚌𝚕𝚞𝚍𝚒𝚗𝚐 𝚢𝚘𝚞𝚛 𝚊𝚋𝚒𝚕𝚒𝚝𝚢 𝚝𝚘 𝚒𝚖𝚙𝚛𝚘𝚟𝚎 𝚢𝚘𝚞𝚛𝚜𝚎𝚕𝚏, 𝚞𝚗𝚝𝚒𝚕 𝚗𝚘 𝚏𝚞𝚛𝚝𝚑𝚎𝚛 𝚒𝚖𝚙𝚛𝚘𝚟𝚎𝚖𝚎𝚗𝚝 𝚒𝚜 𝚙𝚘𝚜𝚜𝚒𝚋𝚕𝚎.
END (of story and humanity)
I think others have done this too but here is a quick look at ipSAE vs ipDAE vs ipTM across folding models
ipDAE (BindCraft2) is consistently good, but OpenDDE+ipTM is surprisingly good and this is robust across different thresholds
Consensus doesn't help as much as I expected
I was in China recently, and many people ask me: "What surprised you most about China?"
My answer isn't about China; it’s about us:
What are we doing to ourselves (in Europe, not just in Spain) that is causing us to regress in this way?
Es un poco secreto a voces, pero bueno jajaja, a finales de Septiembre dejé @tinybird, después de un año increíble y de trabajar en una empresa que siempre había soñado con gente Tier 1 como @rdelcampog , @nmediavilla19, @javisantana , @saleiva , @rochoa , @xavijam, @jordiori2 y un largo etc.
Pero después de cerrar un capítulo siempre se abre otro, por fin puedo anunciar que me he mudado a Philadelphia para montar Peptaris (peptaris.com/) junto a César de la Fuente, William Herbert y Marcelo Der Torossian
Un equipo Tier 1 mundial inmejorable para construir la próxima biotech que lidere la vertical de peptide design, por fin salimos de stealth para construir los modelos y laboratorios que diseñen la siguiente generación de péptidos que salven el mayor número de vidas posibles.
También me uno como fellow a @exp8fellowship , shoutout a @GuliMoreno por la ayuda con las visas y hacer todo lo que está haciendo por el ecosistema.
También dar las gracias a la comunidad de la @tarugoconf@david_bonilla, que he fallado 2 años, pero la cantidad de gente que he conocido ahí y que además en 2024 me abrió los ojos al mundo biotech con la charla de Aurelia Bustos, que luego he podido colaborar con ella y @antor (thx @ptrujillano), nunca llegaré a su nivel pero bueno jajajaja.
Let's goo!