3 ms·
I'd imagine you can set things up where at least that would be logged, no? It also doesn't mean that these LLM tools would be any less secure than other tools
by roflyear 4y ago
I'd imagine you can set things up where at least that would be logged, no?
It also doesn't mean that these LLM tools would be any less secure than other tools (and I'm generally a sceptic of these tools, for what it is worth).
- simonw 4y agoRight, logging things is definitely a good idea. Whether these tools are secure or not depends entirely on how you are using them. If you don't understand prompt injection you're very likely to build a system that's vulnerable to it.
- _6hmp 4y agoAnd possibly even if you do understand it! It seems like it might be a fundamentally intractable problem with LLMs, even if it can be made more difficult to do, no?
- simonw 4y agoYes, exactly: right now I still haven't seen a convincing reliable mitigation for a prompt injection attack. Which means there are entire categories of applications - including things like personal assistants that can both read and reply to your emails - that may be impossible to safely build at the moment.