emil_sorensen

4 ms·

1.
Pruning RAG context down to what the answer actually needs(kapa.ai)
146 points by emil_sorensen 3mo ago|40 comments
2.
by emil_sorensen 4mo ago
happy it helped!
3.
by emil_sorensen 4mo ago
it's a cost/latency trade-off in production + very use-case dependent
4.
by emil_sorensen 4mo ago
Thanks! Yep fixed
5.
The tool that made our AI agent better at using its tools(kapa.ai)
2 points by emil_sorensen 5mo ago|0 comments
6.
We built our own PDF converter benchmark(docs.kapa.ai)
2 points by emil_sorensen 6mo ago|0 comments
7.
Kapa.ai (YC S23) Is Hiring a Customer Solutions Engineer (EU Remote)(ycombinator.com)
1 points by emil_sorensen 1y ago
8.
Kapa.ai (YC S23) is hiring research and software engineers(ycombinator.com)
1 points by emil_sorensen 1y ago
9.
Kapa.ai (YC S23) is hiring a software engineers (EU remote)(ycombinator.com)
1 points by emil_sorensen 1y ago
10.
by emil_sorensen 1y ago
Accuracy drops hard with context length still. Especially in more technical domains. Plus latency and cost.
11.
by emil_sorensen 1y ago
Docs bots like these are deceptively hard to get right in production. Retrieval is super sensitive to how you chunk/parse documentation and how you end up structuring documentation in the first place (see frontpage post from a few week
12.
by emil_sorensen 1y ago
Thanks for the feedback. We should definitely add that. :)
13.
by emil_sorensen 1y ago
OP here. It's kind of ironic that making the docs AI-friendly essentially just ends up being what good documentation is in the first place (explicit context and hierarchy, self-contained sections, precise error messages).
14.
Writing documentation for AI: best practices(docs.kapa.ai)
1 points by emil_sorensen 1y ago|0 comments
15.
by emil_sorensen 2y ago
We focus mainly on external use cases (e.g., helping companies like Docker and Monday.com deploy customer facing "Ask AI" assistants) so we don't run into much of that given all data is public. For internal use cases that req
16.
by emil_sorensen 2y ago
Super cool! Yep, a lot seems to get lost through distillation.
17.
by emil_sorensen 2y ago
I suspect you're right here! Excited to get our hands on the non-distilled o3. :)
18.
by emil_sorensen 2y ago
Yep even with a small bump in performance (which we only saw for a subset of coding questions), it wouldn't be worth the huge latency penalty. Though that will surely go down over time.
19.
by emil_sorensen 2y ago
Curious if anyone else has run similar experiments?
20.
Evaluating modular RAG with reasoning models(kapa.ai)
62 points by emil_sorensen 2y ago|31 comments
21.
by emil_sorensen 2y ago
That's a great point. Reminds me of the "feature, not a bug" Karpathy tweet [0]. [0]: https://x.com/karpathy/status/1733299213503787018?lang=en
22.
by emil_sorensen 2y ago
I find it so interesting that it's possible to develop a "feeling" of a new model.
23.
AI hallucinations: Why LLMs make things up (and how to fix it)(kapa.ai)
196 points by emil_sorensen 2y ago|245 comments
24.
Kapa.ai (YC S23) is hiring AI research and full-stack roles (EU timezone)(ycombinator.com)
1 points by emil_sorensen 2y ago
25.
RAG Best Practices: Lessons from 100 Technical Teams(kapa.ai)
7 points by emil_sorensen 2y ago|1 comments
26.
Kapa.ai (YC S23) Is Hiring an Applied AI Research Engineer (Europe Timezone)(ycombinator.com)
1 points by emil_sorensen 2y ago
27.
Tough conversations pay off: moving from EM to IC(mooreds.com)
2 points by emil_sorensen 3y ago|0 comments
28.
Kapa.ai (YC S23) is hiring a full stack engineer (with LLM focus)(ycombinator.com)
1 points by emil_sorensen 3y ago
29.
by emil_sorensen 3y ago
Extensive LLM evals. It's a real rabbit hole.
30.
by emil_sorensen 3y ago
They don't have to be rendered in the browser, but having all of the structures and section headings help humans to, so I would recommend it. :)

More ›