Looks like that's the search tool usage instructions – they still don't share tool prompts in their published system prompts which continues to be frustrating
@simonw
-
Asking source of $13/day from Claude cost docs
By
–
Where did you get that $13/day number from? I found this page of the Claude docs but it's advising enterprise customers how much to budget when paying full API price for their team member https://
code.claude.com/docs/en/costs -
The jailbreak looked like ‘fix this code’
By
–
Yeah, it really seems like the jailbreak was 'fix this code'.
-
Security bug discovery threatens model’s code writing capability
By
–
We won't know until/unless we hear what the bug actually was If it really was "asking Fable to find security bugs in code found security bugs" then it can't be fixed without making the model useless for writing any code at all, which is one of the most important use-cases
-
AI finding security bugs threatens code writing use case
By
–
We won't know until/unless we hear what the bug actually was If it really was "asking Fable to find security bugs in code found security bugs" then it can't be fixed without making the model useless for writing any code at all, which is one of the most important use-cases
-
Bug in finding security bugs threatens model’s code writing use-case
By
–
We won't know until/unless we hear what the bug actually was If it really was "asking Fable to find security bugs in code found security bugs" then it can't be fixed without making the model useless for writing any code at all, which is one of the most important use-cases
-
Mandatory verifiable KYC implementation for Claude API users
By
–
So now every company that builds on top of the Claude API also needs to implement KYC, in a way that can then be communicated to Anthropic such that they can trust it?
-
No Fable 5 jailbreak against Opus 4.x or GPT 5.x, fortunately
By
–
I'm just glad that no one in the US government thought to try this Fable 5 "jailbreak" against Opus 4.x or GPT 5.x, otherwise I would have absolutely nothing useful to do this weekend.
-
Loss of access to claude-fable-5 after minute script
By
–
I ran a script every minute against the API to see how long I would maintain access to claude-fable-5 – I lost access 14 minutes ago:
-
OpenAI-WebRTC update with gpt-realtime-2 voice model and documents
By
–
I got tired of waiting for OpenAI to integrate their greatly improved gpt-realtime-2 voice model into ChatGPT, so I updated my OpenAI-WebRTC tool to use it and let you paste a document to discuss.