24 chunks · 21 episodes
Simon Willison: "Shocking result on my pelican benchmark this morning, I got a better pelican from a 21GB local Qwen3.6-35B-A3B running on my laptop than I did from...
...e or scalable, but the joy of creation evaporates. A feeling of ennui develops. Simon Willison calls this, in the context of SWEs and LLMs, "deep blue."
There's a movement for what Simon Willison calls "Vegan Models." That is, models trained on only healthy inputs that the model creator has permission to use. Personally, I haven't invested muc...
A powerful pattern for LLMs: swarms of research goblin scouts. Credit to Simon Willison for the term research goblin. The research goblin isn't as good as you at research, but it is way better than you at being patient. It's infinitely p...
...tion, LLM agents can't safely reach mass market. Anthropic's 11% attack success Simon Willison calls a "catastrophic failure rate" "Smarter models" hit asymptotic returns. A structural approach is necessary to unlock the potential.
This week in the "wild west roundup" Simon Willison's roundup of prompt injection attacks this summer A prompt injection technique that hides malicious text in images. Engadget: AI browsers may be the ...
A good insight from Thomas Dohmke via Simon Willison: "You know what else we noticed in the interviews? Developers rarely mentioned "time saved" as the core benefit of working in this new way with agent...
A few months ago Simon Willison called Claude Code a honey badger. I missed this then but it resonates for me. It barrels forward, smashing through things it doesn't understand yet....
Simon Willison on Github Spark: "A word of warning about the key/value store: it can be read, updated and deleted by anyone with access to the app. If you're going ...
A quote from Recurse Center via Simon Willison: "This is why our second self-directive is to build your volitional muscles."
...rst time would think it means. "Inferred definition" is itself a term coined by Simon Willison.
...t. That is, how the model would choose to blackmail its creators in some cases. Simon Willison's summary is worth reading.
We're starting to see more awareness of prompt injection as a vulnerability. Simon Willison's writeup of the EchoLeak vulnerability is worth reading. Notably in the Hacker News comments people are starting to realize how hard LLMs are to sec...
...to handle the compounding cacophony. This reverse engineering of Claude Code by Simon Willison is fascinating.
Simon Willison's deep dive on ChatGPT's dossier feature shows some of the oddities in the behavior.
I agree with Simon Willison's take on vibe coding. "I don't want[ta] "vibe coding" to become a negative term that's synonymous with irresponsible AI-assisted programming either[...
... translation. But now for software in the small it takes just a few seconds. As Simon Willison notes, the people best situated for this world are people who both know how to code and also have the knowhow of using LLMs. A couple of weeks ago I ...
I found this interesting take on making a collaborative writers room via Simon Willison. Writers rooms are a better model for a collaborative, "yes, and" environment than improv, I think. The writers room is about creating a particular t...
Simon Willison asked on Twitter for people's favorite LLM prompting tips. The answers were all over the map. That implies that we're still in the early days of coll...
... to accomplish in practice. But LLMs make it easier than ever before! Credit to Simon Willison for this frame.