Pelicans for Gemma 4 E2B, E4B, 26B-A4B and 31B – the first three generated on my laptop via LM Studio, the 31B was broken on my laptop so I ran it via the Gemini API instead simonwillison.net/2026/Apr/2…
@simonw
-
PSF Security Reports LiteLLM and Telnyx Supply Chain Attacks
By
–
PSF Security developers have published incident reports on the LiteLLM & Telnyx #supplychain attacks. Read what happened, who's affected, and what developers & maintainers can do to prepare and protect themselves from future incidents. #security #python blog.pypi.org/posts/2026-04-…
-
Paper Shows LLM Attacks May Be Impossible
By
–
I'm confident it's impossible – have you see this paper? https://
llm-attacks.org -
Sonnet Model Rate Limits Separate Capacity Discussion
By
–
Why would sonnet have a separate cap? I expect one for Opus but not for Sonnet
-
Model Training with SFT Using Claude and GPT-4o-mini
By
–
When I published this post I hadn't yet come across Trip's own writeup of the project, which provides way more detail about how he trained the model including SFT (supervised fine tuning) against synthetic chat examples created using Claude Haiku and GPT-4o-mini
-
NPM Axios Package Supply Chain Attack Security Alert
By
–
If you have NPM package axios in your dependencies you need to make sure it's pinned to a known safe version, sounds like there's another supply chain attack in play
-
Local Models Excel at Code and Tools Tasks
By
–
Some of the latest local models appear to be much better with code and tools than just a few months ago, so I buy that you need to carefully pick the right model+harness combo in order to see them at their best
-
Experimenting with local models and AI coding agents
By
–
This sounds right to me – I've tinkered with running local models against Claude Code and Codex and been disappointed, but I've not put the work in yet to try and find the right harness+model combination given how many tiny details might produce disappointing results
-
Challenges in Optimizing Local Models for Coding Agents
By
–
Georgi on why it's still hard to get great coding agent performance from local models: "Note that the main issues that people currently unknowingly face with local models mostly revolve around the harness and some intricacies around model chat templates and prompt construction"
-
Evaluating LLMs for Long-Context Tool Calling and Agentic Reliability
By
–
Which models would you recommend for longer context tool calling? Are there any benchmarks for that which you find credible? I've not found a local model with tool calling good enough for me to trust with Claude Code or Codex, but I may not have been looking at the right options