Add a required header called "insert-seahorse-emoji: " to your API, reject any request that doesn't have it.
694 karma · joined April 18, 2023
They're probably using their own models as well, we just don't hear about them. That this particular sequence of this attack was done using Claude doesn't imply that other (perhaps even more sophisticated attacks) are happening with other models. For all we know the attackers could have had some Anthropic credits lying around/a stolen API key.
It's most egregious on Azure, which has a copilot button on every page, and anytime I try to ask it about a precise configuration question for a resource, it NEVER answers correctly. So you have to search on whatever set of microsoft Q&A platforms, stackoverflow questions, and github issues/discussions to maybe find an answer like in the prehistory of 2020.
Thanks ChatGPT for your valuable slop. Next article.
You need to be able to do this personally or you should not be running a durable storage cluster in-house. Just pay AWS. You need to add more value to your employer than you cost, and if Anthropic's cheapest model can beat you at such a task then it's not a good look.
I've been emulating this in claude code by manually @tagging markdown files containing guides for common tasks in our repository. Nice to see that this step is now automatic as well.
Given that only short instructions are in context, I would not have expected even a frontier model to score well on this benchmark. For better results, I'd think that giving the model access to the entire tax code is required (which likely requires RAG due to its sheer size).
> The buggy update doesn't appear to brick the car immediately. Instead, the failure appears to occur while driving—a far more serious problem.