HackerTrans
TopNewTrendsCommentsPastAskShowJobs

heyzk

no profile record

Submissions

Knuth Optimystic Benchmark

optimystic.ai
2 points·by heyzk·3 năm trước·1 comments

comments

heyzk
·10 tháng trước·discuss
I just shipped 3pio, a drop-in test runner that context-optimizes your test output. It uses your existing test runner and tests so zero changes to your codebase or tooling to use it.

IME it results in much less context clutter from your test output.

https://github.com/zk/3pio
heyzk
·11 tháng trước·discuss
I think semantically this is true, but the way humans and agents handle context, at the moment, is different enough to warrant a separate structure.
heyzk
·3 năm trước·discuss
I'm curious about the logistics here, what's preventing OAI from obtaining a copy and invalidating those credentials?
heyzk
·3 năm trước·discuss
Mark Brown of Game Maker's Toolkit fame runs a yearly game jam centered around a specific theme [1], which would line up well with this initiative. The quality of submissions is really, really impressive.

[1] https://www.youtube.com/watch?v=tYxZ_LOlFnU
heyzk
·3 năm trước·discuss
You see a similar loosening of the term in other fields e.g. open source journalism. Although that seems to be more about crowdsourcing than transparency or usage rights.
heyzk
·3 năm trước·discuss
I'm confused about how a loose chain can cause the rear wheel to lock. Does the chain wedge between the rear sprocket and wheel?
heyzk
·3 năm trước·discuss
Great writeup, this helps us understand where to spend our time vs what OpenAI's progress will solve.
heyzk
·3 năm trước·discuss
It would be really interesting to see the first thing everybody asked ChatGPT. Here's mine:

https://chat.openai.com/share/26792685-2790-4560-9f8d-9524de...

edit: Just noticed the shared version reports May 26, but this conversation was from January.
heyzk
·3 năm trước·discuss
Hey all, I found the post about Don Knuth and ChatGPT [1] very interesting so I hacked together this project over the weekend. Optimystic periodically re-runs Knuth's questions against the latest GPT model (additionally I've asked GPT to score the updated answers as PASS/FAIL).

Initially I thought it would just be a fun thing, but I've realized there could be some value to the larger LLM community. Also I think it would be interesting to apply this format to experts in other fields.

Appreciate any feedback.

[1] https://news.ycombinator.com/item?id=36012360
heyzk
·3 năm trước·discuss
Is the benefit here that EVA supports a declarative style of composition over LangChain's (or similar) imperative style?
heyzk
·3 năm trước·discuss
The flux docs indicate that you're using a custom-trained LLM and in another comment it was mentioned you're using LangChain (to integrate non-deterministic tooling perhaps). Just curious, but are you using some combination of your own model, GPT3/4, and calculators / datasheet readers / etc?
heyzk
·3 năm trước·discuss
Would love to hear more about (1), are you vectorizing different types of transforms on the original document?