Hugging Face just dropped a comprehensive blog post on Vision-Language Models (VLMs) — covering architecture, training, benchmarks, and useful resources. A must-read if you're navigating the multimodal frontier! https://
huggingface.co/blog/vlms-2025
RESEARCH
-

Hugging Face Releases Comprehensive Vision-Language Models Guide
By
–
-
LLMs for Complex Game Projects: Tractability and Resource Constraints
By
–
(I briefly considered “Beat Factorio with LLMs” as a project during my sabbatical but it looks very far from being tractable without a team and funding, which defeats the purpose of a sabbatical project.)
-

Embeddings Are Underrated: Technical Deep Dive
By
–
# Embeddings are underrated (2024) just a really excellent piece of technical writing.
-

MLE-Dojo Benchmark Evaluates Frontier LLMs on ML Engineering
By
–
MLE-Dojo: A new benchmark to evaluate LLM agents on real Machine Learning Engineering tasks. Its key innovation? An interactive environment that allows agents to experiment, debug, and refine solutions via structured feedback loops. Here’s how 8 frontier LLMs perform
-
LLM domain training should produce manuals not weight updates
By
–
yes exactly, "training" an LLM on a target domain should output a manual not a weight diff.
-
Reviewer Accountability and Open Scientific Review Systems
By
–
The reviewers are responsible for the reviews regardless of the tools they choose to use.
The problem is accountability.
That's why I've been an advocate of open reviews, and of removing the exclusive relationship between papers and reviews. Speaking of which, the destruction by -
New Research Podcast Justified Posteriors Explores AI Papers
By
–
Check out this amazing new podcast by two of my amazing former postdocs: @AndreyFradkin and @SBenzell The format is that they state their priors, read a paper, discuss it on the pod, and then update their priors. It's called "Justified Posteriors" Link in next post…
-

Saudi Arabia’s HUMAIN partners with NVIDIA for massive AI compute facilities
By
–

HUMAIN, filiale saoudienne spécialisée en IA, s’associe à NVIDIA pour construire de véritables fabriques d’intelligence, avec une puissance de 500 mégawatts et un supercalculateur équipé de 18 000 GPU Grace Blackwell. C'est du lourd. L’Arabie saoudite ne creuse plus des
-
Intellectual Effort in Research Labs and Knowledge Advancement
By
–
(People often undervalue the intellectual effort involved in doing this stuff well. I think “research lab” is the right way to think about a small group of professionals attempting to advance the state of human knowledge in a narrow sub-specialization.)