Following the success of its ICML 2024 tutorial, we're now releasing Part 4 of Physics of Language Models—a research effort from FAIR at Meta to uncover universal laws of AI through controlled, scientific experimentation. In Part 4, we challenge today's architecture design
LLMS
-
LLMs solve verifiable math and coding problems only
By
–
right now they do math problems that are verifiable, or coding problems where you can run the code and see if it worked and was correct it's very strange isn't it?
-
LLMs train without external data, breaching the “data wall”?
By
–
Absolute Zero: LLMs can train without any external data Has the "data wall" just been breached? Recent RL paradigms often relied on a set of questions an answers that needs to be manually curated. Researchers from @Tsinghua_Uni went like "why though". Indeed, why learn
-
Meta’s Synthetic Data Kit for Llama 3.2 on Google Colab
By
–
Link to the Colab Notebook here: https://
colab.research.google.com/github/unsloth
ai/notebooks/blob/main/nb/Meta_Synthetic_Data_Llama3_2_(3B).ipynb
… Synthetic Data kit from Meta: -
Transform Documents into High-Quality Synthetic Datasets with Llama
By
–
Turn any document into a high-quality synthetic dataset using Llama! Unsloth AI and Meta released a free notebook that transforms your documents into high-quality synthetic datasets, and then finetuning them using Llama. 100% Open Source
-
ChatGPT vs Gemini vs Grok: AI Model Performance Comparison
By
–
by all counts, chatgpt is worst at reasoning while gemini or in that matter grok also outperforms in terms of latency, quality of output and research skills.
-
Building and Improving RAG Pipelines: From PoC to Production
By
–
Over the past 2 years, we’ve (
@towards_AI
) helped teams create, iterate and improve RAG pipelines, slash hallucination rates, build PoCs, and get LLM demos into production. At some point, we got tired of repeating the same advice on calls, Slack threads, and conference stages—so -
Google I/O AI Launch Predictions: Grok 3.5, Imagen 4.0, Veo 3, GPT-5
By
–
Tout converge pour exploser pendant le Google I/O. X va s’enflammer avec les lancements : Grok 3.5
Imagen 4.0
Veo 3
o3-pro
GPT-5, etc… Meta, Qwen et Anthropic vont-ils aussi rejoindre la fête ? -
GPT-2 code size is surprisingly small
By
–
GPT-2 ne tient que sur 174 lignes de code… c'est pas fou ?
