If they're just dumping text from my chat sessions into their future training runs then I expect it can learn new facts from them – but they're so vague with their language, maybe they're using user data in a different more subtle way than that?
@simonw
-
Can Claude Disclose Secrets From Uploaded Documents?
By
–
If I opt in to this, is there any chance a future version of Claude could tell someone else a secret it learned from a document that I upload for summarization or similar?
-
LLMs Demystified: Matrix Multiplication and Local Inference
By
–
I use this line about LLMs a lot – that they're a bunch of matrix multiplication – because I think it demystifies them A multiple GB file of floating point numbers you can download and run matrix multiplications against on your own machine is a bit less weird and frightening
-
Model Security Vulnerabilities: Environment Variable Theft Techniques
By
–
That's pretty common these days, the challenge is making those protections completely airtight. Check out how @wunderwuzzi23 defeats model resistance to stealing environment variables here for example; https://
embracethered.com/blog/posts/202
5/openhands-the-lethal-trifecta-strikes-again/
… -
Pydantic Code Pattern Implementation Insights
By
–
I've not looked at this in a while but Pydantic published code that uses a similar pattern to mine here:
-
Vibe Coding Definition Debate: Unreviewed LLM-Generated Code
By
–
I will stubbornly continue to use the original definition! https://
simonwillison.net/2025/Mar/19/vi
be-coding/
… Using it to mean any form of AI-assisted programming deeply frustrates me because we really need a term that means "LLM-written code that nobody ever took the time to review" -
Mistral launches proprietary API model platform December 2023
By
–
From the post I linked to dated December 11th 2023 – this was the first time Mistral made a model available via their API that wasn't an open weights model https://
mistral.ai/news/la-platef
orme
… -
Mistral’s API-only model strategy since December 2023
By
–
They've been releasing API-only models since the first Mistral Medium way back in December 2023, this isn't a new strategy for them
-
Cohere Command reasoning model released with non-commercial license
By
–
The first reasoning model in the Cohere Command family of models, sadly the open weights release is still under a non-commercial use license
-
Claude Code Exfiltration Vulnerability via DNS Requests
By
–
Does that reliable work though? @wunderwuzzi23 has a trick where he has it run strings and grep to hide that it's accessing .env https://
embracethered.com/blog/posts/202
5/claude-code-exfiltration-via-dns-requests/
…