Saul Garcia

@pepmartorell No es automático, pero tampoco “temporal” por definición.

@pepmartorell …también el uso, la adopción y la inversión. Ansiedad no es lo mismo que rechazo. Y en Jevons discrepo bastante: si todavía somos malos implementando IA, precisamente cuando aprendamos a capturar valor y baje el coste, puede aumentar la demanda de gente que sepa aplicarla bien.

Ojo con esto que indica Jon 👇👇👇

@LinusEkenstam Poblenou not gentrified, sounds like ages ago 😅

@sgar2209 I’ve been here since they opened, I agree that the proposal is way to American, but damn, it’s good. also they desperately need a refresh after 11 years 🥸 I lived on Ciutat de Granada when Poble nou was not gentrified 😂

GPT Image 2 I went for brunch yesterday at my local spot. Snapped 1 photo. Added it to Chat, and it spits out a full new brand guideline for them. Honestly way better than what they got. prompt below

@esport3 Jo he entrat sense problemes, com sempre al Nou camp Nou tret d’un dia (Alabes, que tampoc va ser tant gros) que ja es van obrir tots els diaris amb aquesta notícia de capçalera… au, vinga, que les eleccions han acabat ja 😉

⚠️NOVES INCIDÈNCIES AL CAMP NOU EN DIA DE PARTIT 📱Més cues a l'OAB i confusió entre espectadors del Barça-Newcastle per l'app on arriben les entrades, les comprades i les transferides 🎟️Ja va passar contra l’Eintracht sense ser el nyap del Barça-Alabès 3cat.cat/esport3/en-dir…

@AleixG7 72 / 28

Diu Gerard Romero que els sondejos “no oficials” entre periodistes comencen a apuntar a un 60/40 favorable a Laporta.

Com a soci del @fcbarcelona_cat exigeixo al candidat @victor_font que retiri les mencions que desprestigien el millor Barça de la història, menystenen el Club i fan d’altaveu de les campanyes dels que volen destruir-nos, i demani disculpes als socis del Club.

A l'escola saltant els bassals d’aigües residuals. El dia a dia de l’escola de mòduls #EscolaLaFlordeMaig @educaciocat @consorciedubcn @sindicaturabcn @MariaEugeniaGay @eniubo @jsccatalunya @rac1 @CatalunyaRadio @som3cat @elsmatins @montserratduran @jaumecollboni @DavidBueno33

@jm_alexia linkedin.com/posts/saulgh_t…

More noise and less signal for the scientific community. Thanks LLMs!

This is huge. First signs of incoming 1000x acceleration of scientific progress. "Researchers found that when scientists use AI, their productivity soared. The biggest jump was in the social sciences and humanities, where output increased by 59.8%, while biology and life sciences saw a 52.9% increase." "Meanwhile, in physics and math, the scientists report a 36.2% boost." "LLM adoption is associated with a large increase in researchers' scientific output," wrote the team.

@Dr_Singularity linkedin.com/posts/saulgh_t…

@OfficialLoganK @GoogleAIStudio Want to test! Thanks!

Big upgrade to vibe coding in @GoogleAIStudio lands in Jan, but if you want to test early… 👇🏻

@karpathy @threadreaderapp unroll

Excited to release new repo: nanochat! (it's among the most unhinged I've written). Unlike my earlier similar repo nanoGPT which only covered pretraining, nanochat is a minimal, from scratch, full-stack training/inference pipeline of a simple ChatGPT clone in a single, dependency-minimal codebase. You boot up a cloud GPU box, run a single script and in as little as 4 hours later you can talk to your own LLM in a ChatGPT-like web UI. It weighs ~8,000 lines of imo quite clean code to: - Train the tokenizer using a new Rust implementation - Pretrain a Transformer LLM on FineWeb, evaluate CORE score across a number of metrics - Midtrain on user-assistant conversations from SmolTalk, multiple choice questions, tool use. - SFT, evaluate the chat model on world knowledge multiple choice (ARC-E/C, MMLU), math (GSM8K), code (HumanEval) - RL the model optionally on GSM8K with "GRPO" - Efficient inference the model in an Engine with KV cache, simple prefill/decode, tool use (Python interpreter in a lightweight sandbox), talk to it over CLI or ChatGPT-like WebUI. - Write a single markdown report card, summarizing and gamifying the whole thing. Even for as low as ~$100 in cost (~4 hours on an 8XH100 node), you can train a little ChatGPT clone that you can kind of talk to, and which can write stories/poems, answer simple questions. About ~12 hours surpasses GPT-2 CORE metric. As you further scale up towards ~$1000 (~41.6 hours of training), it quickly becomes a lot more coherent and can solve simple math/code problems and take multiple choice tests. E.g. a depth 30 model trained for 24 hours (this is about equal to FLOPs of GPT-3 Small 125M and 1/1000th of GPT-3) gets into 40s on MMLU and 70s on ARC-Easy, 20s on GSM8K, etc. My goal is to get the full "strong baseline" stack into one cohesive, minimal, readable, hackable, maximally forkable repo. nanochat will be the capstone project of LLM101n (which is still being developed). I think it also has potential to grow into a research harness, or a benchmark, similar to nanoGPT before it. It is by no means finished, tuned or optimized (actually I think there's likely quite a bit of low-hanging fruit), but I think it's at a place where the overall skeleton is ok enough that it can go up on GitHub where all the parts of it can be improved. Link to repo and a detailed walkthrough of the nanochat speedrun is in the reply.

State of mind d’aquesta setmana… @LaSotana_

@ai_for_success Love to have one ☝️🙏, waitlist is taking sooo long 😅

I haven’t seen this kind of craze for anything like I’ve seen for Comet. Yesterday, a friend from X told me her mom wanted to try Comet, she is 80+ so I sent her an invite. Over the weekend, I gave away 40 more invites to people. 🧡

@PerplexityComet 😔

This is a sign to check your inbox for a Comet invite.

Summer project ahead…

@toniaira 13 trucades en una hora (fa una setmana), el meu record fins la data, espero que si no és amb regulació sigui amb tecnologia linkedin.com/posts/saulgh_1…

Realment ha servit de MOLT la nova normativa del govern espanyol per evitar-nos l’spam telefònic. Queda clar que ja no hem de patir per aquests assetjadors i estafadors (presumptament), oi? Ja no truquen des de mòbils. Tot arreglat. *(i entenc que també es capta la ironia)