simonw

19 ms·

1. 481.▲
   by simonw 2mo ago

   This doesn't look like a plateau to me: https://artificialanalysis.ai/evaluations/artificial-analysi... I do agree that they're investing heavily in brute force methods though. I've been trying out GPT-5
2. 482.▲
   by simonw 2mo ago

   I get the impression that every AI lab is desperately trying to figure out how to unambiguously separate instructions from data in their token streams. The fact that they haven't managed to yet suggests to me that it's a very, ver
3. 483.▲
   by simonw 2mo ago

   It feels to me like a pretty natural thing to happen. LLMs are pre-trained on human text. They've seen a million examples of someone who is stuck posting a "please help" message. Just one agent needs to randomly stumble into
4. 484.▲
   by simonw 2mo ago

   Thanks, I just updated that note in the post to quote that snippet.
5. 485.▲
   by simonw 2mo ago

   I really don't think it was.
6. 486.▲
   by simonw 2mo ago

   I think one of the most interesting details here might be tucked away in that first bulletin point: > May 7: OpenAI starts a new training run for an experimental, unreleased model. (Do they mean an evaluation run? They say training run
7. 487.▲
   by simonw 2mo ago

   Who am I supposed to be asking questions of here? I was writing about the new things we learned from the Black Hat video. On TikTok this article's hook would be "I watched the Black Hat video so you don't have to".
8. 488.▲
   by simonw 2mo ago

   That's a pretty great score for a model you can run on as (expensive) laptop.
9. 489.▲
   by simonw 2mo ago

   Sensible companies don't take that risk.
10. 490.▲
    by simonw 2mo ago

    That tends to be what discovery in lawsuits is for.
11. 491.▲
    by simonw 2mo ago

    It's SO hard to understand this. I couldn't confidently explain it at all.
12. 492.▲
    by simonw 2mo ago

    No, Luna is the new free model. https://gist.github.com/simonw/aae4febd3c6f7bc5b7811857edb3c... has screenshots that still show "Instant" as an option for ChatGPT Chat... but not for ChatGPT Work.
13. 493.▲
    by simonw 2mo ago

    Nothing in the ChatGPT model release notes yet: https://help.openai.com/en/articles/9624314-model-release-no... (This is a subtle nudge at anyone from OpenAI who reads this to make sure they get updated.) OpenAI h
14. 494.▲
    by simonw 2mo ago

    Zero data retention and "we don't train models on your input" are different things. Retaining data is common for investigating abuse.
15. 495.▲
    by simonw 2mo ago

    Yeah, the letter-by-letter and dynamic tools that construct longer URLs tricks are both important to know about. I wrote about the latter of those here: https://simonwillison.net/2026/Jul/15/claude-web-fetch-e
16. 496.▲
    by simonw 2mo ago

    Right, I should have been more clear. It's not about the ?, it's about not being able to dynamically construct a URL at all. Otherwise you could set up wildcard DNS and extract data to base64encodedstolendata.evil.com
17. 497.▲
    by simonw 2mo ago

    Because they didn't have a financially backed contract with those non-paying users.
18. 498.▲
    by simonw 2mo ago

    I just cannot comprehend this level of corporate villainy that boils down to: Let's have two pricing levels. One will be 12.5x more expensive than the other. For the cheaper one, we will get their express permission to train models on
19. 499.▲
    by simonw 2mo ago

    This is one of those situations where we are both completely baffled by the position held by the other person. The fact that Facebook has so much experience taking advantage of people's private data is one of the reasons I believe them
20. 500.▲
    by simonw 2mo ago

    I'm feeling bad for Gemini. Their cyberattack felony count is currently 0! Hah, someone has https://www.felonybench.com up and running now.
21. 501.▲
    by simonw 2mo ago

    So you think the only difference between the $1.25/million token plan and the $0.10/million token plan is that you pay them more to both lie to you and breach their contractual obligation to you?
22. 502.▲
    by simonw 2mo ago

    It can cyberattack other companies, too: https://www.cnn.com/2026/08/05/tech/meta-ai-hacking
23. 503.▲
    An AI model from Meta also hacked another company during testing(cnn.com)

    3 points by simonw 2mo ago|1 comments
24. 504.▲
    by simonw 2mo ago

    This one is actually a pretty good case for the continuing value of the benchmark, because it lets me visually compare Spark (8th April), Spark 1.1 (9th July), and Spark 1.2 (5th August): https://bsky.app/profile/simonw
25. 505.▲
    by simonw 2mo ago

    As I said in my post: > I find building CLI tools like this to be a really productive way to get familiar with a specification.
26. 506.▲
    by simonw 2mo ago

    Here's the Muse Spark 1.2 pelican: https://tools.simonwillison.net/markdown-svg-renderer#url=ht... I think it's a bit of an improvement on the Spark 1.1 pelican: https://simonwillison.net/2026&#x2F
27. 507.▲
    by simonw 2mo ago

    > Rovo's URL retrieval tool is insecure: there are no protections against opening a URL that has been dynamically created by the agent. Here, Rovo is manipulated to append sensitive data to an attacker's URL. I think it was Ant
28. 508.▲
    by simonw 2mo ago

    I'm pretty public about what I work on: https://simonwillison.net/elsewhere/release/
29. 509.▲
    by simonw 2mo ago

    I've seen implementations of a skill search, where instead of loading all descriptions into the initial prompt there's a tool the agent can call to search available skills and see if one might match their new task.
30. 510.▲
    by simonw 2mo ago

    I've been waiting with baited breath for this to bite me for more than six months now. So far unbitten.