(more info https://latent.space/p/noam-brown)
MACHINE LEARNING
-
Why Self-Play AI Training Hasn’t Scaled Since AlphaZero
By
–
pretty crazy that DeepMind got self-play working w AlphaZero in *2017* yet basically no one has been able to make it work since
-
Voxtral outperforms Whisper and proprietary speech transcription models
By
–
Voxtral comprehensively outperforms Whisper large-v3, the current leading open-source Speech Transcription model, in speech transcription. It beats GPT-4o mini Transcribe and Gemini 2.5 Flash across all tasks, and achieves state-of-the-art results on English short-form and
-
Fine-Tuning Gemma 3n for Medical VQA: On-Device Clinical AI
By
–
📢Fine-Tuning Gemma 3n for Medical VQA
— Satya Mallick (@LearnOpenCV) 15 juillet 2025
The future of clinical AI is on-device, private, and specialized. We took Google's new Gemma 3n, a powerful but generalist VLM, and fine-tuned it to become a radiologist's assistant.
Our latest guide provides a deep dive into the process,… pic.twitter.com/r6AfxZEr7CFine-Tuning Gemma 3n for Medical VQA The future of clinical AI is on-device, private, and specialized. We took Google's new Gemma 3n, a powerful but generalist VLM, and fine-tuned it to become a radiologist's assistant. Our latest guide provides a deep dive into the process,
-
Industrial Research Units Must Build Own Foundation Models
By
–
Another similar saying is, you'd better have your own models instead of only papers (citation counts/impact aside), if you are industrial research unit for foundation model
-
UAI 2025 Proceedings Volume 286 Published on PMLR
By
–
Volume 286 https://
proceedings.mlr.press/v286/ Proceedings of UAI 2025 Is now available on PMLR. -
AI Benchmarking Limitations: Beyond GPQA and MMLU Metrics
By
–
Another sign that the benchmarking of AIs has grown too narrow – needle-in-a-haystack, instruction following, hallucination rates, etc. are all really important, and just measuring things correlated with GPQA/MMLU/etc may blind users to other models strengths and weaknesses.
-
Hopfield Networks: A Deep Dive into Neural Architecture
By
–
I see you’re a fan of Hopfield networks.
-
Discussion of AI reasoning traces and model visibility
By
–
Also most people had never seen a real, full reasoning trace before, and the novelty was hyped up by the fact o1 didn’t show them E.g. this got likes at the time despite it being a (relatively) obscure model:
-
Advanced AI Platform for Professionals: Deep Reasoning and Extended Context
By
–
Designed for pros Developers
Researchers
Data Academics
Teams needing deep reasoning
Large-context processing (128k in-app / 256k tokens API)
Early access to cutting-edge video / coding tools.