Cool stuff! I suspect we will now see a bunch of startups coming up focused on helping companies reduce their AI usage, similar to how we have companies focused on optimizing cloud costs.
I'm working on ScaleDown [1], a context pruning API.
So over the past few years, I have seen how contexts have been steadily growing in AI apps. And while the context lengths of LLMs have also been increasing, they are still effectively about 200k tokens. The performance drops off a cliff after that (you might have noticed it as well with long AI chats).
It is a simple API that prunes away irrelevant parts of a context for a given prompt, a.k.a. context-aware pruning. Integration is super simple: just an extra API call before the final LLM API call. You can get an API from the website.
I would love to chat if this is something that is relevant to you and if you have any feedback on what we are building!
[self promotion alert] the "you are not alone" point really resonated with me. When I lost my job, I was alone, helpless and not sure what the next steps were. This is why I tried to create a community of people willing to support and be a listening ear for people going through job loss and this tough job market. It's at layoff.supprt. honestly I have not been supporting it for a while but of you find this helpful and would like more features then do let me know!
Really intresting. What did the original prompt look like? Perhaps the original prompt was not that good? I feel like the changes claude suggested (except a couple maybe) are already pretty well known prompt engineering practices.