Each month, I publish the previous newsletter, now a month old, so people can see what they are missing. Here is what you would have received a month ago, covering what happened in LLMs in April: https://github.com/simonw/monthly-newsletter-archive/blob/main/2026-04-april.md
…
LLMS
-
April newsletter published late to show missed content
By
–
-
AI Talk to Art to Robot Build: Future Awesome
By
–
talk to AI -> create AI art -> robot builds it. Future gonna be so awesome
-
Claude’s Higgsfield Reframe: Stop Cropping, Preserve Backgrounds Across Aspect Ratios
By
–
You may now stop cropping your shots!
— Futurepedia – Learn to Leverage AI (@futurepedia_io) 1 juin 2026
Claude now supports Higgsfield Reframe: give a reference image and it preserves backgrounds across any aspect ratio – perfect for podcasts, IRL streams, commercials, films, and cartoons.
Use it on Higgs or inside Claude via Higgs MCP.… pic.twitter.com/d8gyLfrstMYou may now stop cropping your shots! Claude now supports Higgsfield Reframe: give a reference image and it preserves backgrounds across any aspect ratio – perfect for podcasts, IRL streams, commercials, films, and cartoons. Use it on Higgs or inside Claude via Higgs MCP.
-
Large scale pretrained models remain important in AI
By
–
This was right five years ago, and still is: “Large scale pretrained models are certainly likely to figure prominently in artificial intelligence for the near future, and play an important role in commercial AI for some time to come. The results that have been achieved with them
-
Marcus and Davis 2021 insight on large pretrained models
By
–
and see Marcus and Davis, all the way back in 2021: “Large scale pretrained models are certainly likely to figure prominently in artificial intelligence for the near future, and play an important role in commercial AI for some time to come. The results that have been achieved
-
LLMs as bandaids: need for a sounder AI foundation
By
–
Current AI is bandaids all the way down; LLMs can’t play nicely with basic tools like databases and knowledge graphs, and you never know what you are going to get. It’s long time to face facts: LLMs have their uses, but we need a sounder foundation for AI.
-
Benchmarks are boring; enterprises need shipping LLM capabilities now.
By
–
Wall Street is watching frontier model benchmark porn while enterprises quietly die waiting for capabilities that actually ship. MMLU and HLE were very useful for about a nano-second. But now they are just boring. We don’t need to know how well an LLM can do on a test. We need
-
User waits for weights before commenting on M3 numbers
By
–
M3 numbers look good but am gonna wait for the weights before I comment any further