Jesus RodriguezinTowards AIThe Method OpenAI Uses to Extract Interpretable Concepts from GPT-4Highly scalable sparse autoencoders might be an interesting solution to one of the toughest challenges in generative AI.5d ago5d ago
Jesus RodriguezinTowards AISynthetic Data Generation in Foundation Models and Differential Privacy: Three Papers from…A reference architecture, security challenges and some recipes are some of the methods outlined in Microsoft’s papers.Jun 31Jun 31
Jesus RodriguezinTowards AIInside One of the Most Important Papers of the Year: Anthropic’s Dictionary Learning is a…The model builds on research from last year and tries to understand interpretable features in LLMs.May 282May 282
Jesus RodriguezinTowards AIInside Infini Attention: Google DeepMind’s Technique Powering Gemini 2M Token WindowThe method combines compressive memory and attention mechanisms in a single structure.May 20May 20
Jesus RodriguezinTowards AIInside AlphaFold 3: A Technical View Into the New Version of Google DeepMind’s BioScience ModelA highly improved architecture drastically expands the capabilities of AlphaFold.May 13May 13
Jesus RodriguezinTowards AIPredicting Multiple Tokens at the Same Time: Inside Meta AI’s Technique for Faster and More Optimal…The mehod addresses the limitations of the classic next token prediction method.May 61May 61
Jesus RodriguezinTowards AISome Technical Notes About Phi-3: Microsoft’s Marquee Small Language ModelThe model ius able to outperform much larger alternatives and now run locally on mobile devices.Apr 29Apr 29
Jesus RodriguezinTowards AISome Technical Notes About Llama 3New tokenizer, optimized pretraining and some other details about Meta AI’s new model.Apr 22Apr 22
Jesus RodriguezinTowards AIInside Ferret-UI: Apple’s Multimodal LLM for Mobile Screen UnderstandingThe new research can have an impact in task automation in mobile apps.Apr 15Apr 15
Jesus RodriguezinTowards AIInside Jamba: Mamba, Transformers, and MoEs Together to Power a New Form of LLMsThe new architecture was pioneered by AI21 Labs and brought the best of several architecture paradigms in a single model.Apr 8Apr 8