- ChatGPT's knowledge cutoff is October 2023
- Claude 3.5 Sonnet's knowledge cutoff is April 2024
- Gemini 2.0 Flash's knowledge cutoff is August 2024
This means that, by default, the LLMs don't know about updates to languages/libraries/frameworks after these dates. What's your preferred method for working around this? Feel free to include answers for LLMs and/or other AI tools like Cursor, Windsurf, etc.
Top comments (5)
I think I've had limited success mentioning what versions of things I'm using, in my AI prompts, so that it gives me better results. When it comes to software development this is a big issue for everyone (i.e. when the AI generates code that mixes incompatible versions of things etc), so it does need some kind of solution. Perhaps better annotation of training data (code) that includes timestamps and/or version numbers, might address this some day, so you can ask questions to LLMs and then perhaps give date-ranges for what you want it consider, etc.
Thanks for responding! Mentioning the version numbers is a good tip. And I agree that it's a big issue. I might write an article based on what I've found so far in my research.
Cool. I followed you, and will be interested in what you find out!
Hi Clay. I just published my article here: dev.to/jhonkaman/the-knowledge-cut...
Cool. thanks!