try: result = json.loads(response) except json.JSONDecodeError: # TODO: God forgive me... import openai ...

Jan 14, 2023 · 2:39 AM UTC

30
103
14
1,276
Replying to @goodside
Been using XML after hella headaches with JSON parsing bc my outputs sometimes had new lines
1
2
XML has its own escaping issues with angle brackets and ampersands if the document is complex or looks XML-ish enough to it. I wrote it off because of that but maybe I should try it more.
2
7
more replies
Replying to @goodside
This is so cool. Wonder if GPT can summarize a complex regex
3
1
Search my timeline — I've done a lot of examples of those, and they should all work better now since they were done in the text‑davinci‑002 era.
1
4
more replies
Replying to @goodside
I've thought about creating a fine-tuned Curie model specifically trained to fix buggy JSON.
1
21
Me too — I'd synthesize the tuning examples using k-shots in code‑davinci‑002. Seems like it would work.
1
1
6
more replies
Replying to @goodside
cool, so, instead of flagging errors that could be fixed at the source, I can have my corrupt data be silently replaced by potentially hallucinated data. me likes
3
3
1
83
Yes, but that’s a problem for the next dev.
3
46
Replying to @goodside
piped.video/4GrLx5XUy9M

(Very) Few Shot NLP Tokenization with GPT-3 (even escapes strings)

Could GPT-3 be used for basic NLP tasks like tokenization? I made a video to find out!—► Remember to Like, Comment, and Subscribe!—Connect with me:Twitter -...

youtube.com
1
Replying to @goodside
Someone should fine tune a model for this, distill down the weights, host on a public endpoint and provide core infra for us all
2
Load more