tough

31 ms·

1. 631.▲
   by tough 1y ago

   I love it when models give up, gives me some hope humans will still be required for the time being lol
2. 632.▲
   by tough 1y ago

   LLM's are good at interpolation but bad at extrapolating
3. 633.▲
   by tough 1y ago

   hey now typos make you authentic, that's a plus
4. 634.▲
   by tough 1y ago

   Bun is really taking their lunch with a more simpler USP, "Faster"
5. 635.▲
   by tough 1y ago

   I made a little bun app / script that takes a config.toml file with tasks and downloads the files using fetch or git clone it aint much but its simple and i control it
6. 636.▲
   by tough 1y ago

   hammer + man = adult male body (75%)
7. 637.▲
   SEM-CTRL: Semantically Controlled Decoding(arxiv.org)

   3 points by tough 1y ago|0 comments
8. 638.▲
   by tough 1y ago

   Thank you for your research really impressive work!
9. 639.▲
   by tough 1y ago

   So how I envision this would be like a dual system, you let the frontier bigger LLM come up with the overall function signature, structure, and reasoning/planning around the specific code, but then have it ask the hyperspecialized fine
10. 640.▲
    by tough 1y ago

    it feels to me most of the real usage of AI is in coding right now, so a small lab that decided to go all in into just code-gen would have at least the differentiator of a narrower field to beat the bigger incumbents doing it all? I dunno t
11. 641.▲
    by tough 1y ago

    You could just have specialized fine-tunes for esxh programling la guage that are only called when writing code, a more general bigger model could pass the plan/pseudo code to it
12. 642.▲
    by tough 1y ago

    what i do if no good llms txt is to download the whole docs from gh or website and keep the MD files available to my agent via a small mcp server
13. 643.▲
    by tough 1y ago

    https://ai-sdk.dev/llms.txt
14. 644.▲
    by tough 1y ago

    many docs now include llms.txt https://llmstxt.org/
15. 645.▲
    by tough 1y ago

    ASTrust: Towards More Trustworthy and Interpretable LLMs for Code through Syntax-Grounded Explanations https://arxiv.org/abs/2407.08983 AST-T5: Structure-Aware Pretraining for Code Generation and Understanding https:&
16. 646.▲
    by tough 1y ago

    The code can be found here: https://github.com/eth-sri/type-constrained-code-generation
17. 647.▲
    by tough 1y ago

    okay so basically the faster compiling means a tigher feedback loop for the LLM to -know- if the code compiles or not etc? interesting is go faster than rust?
18. 648.▲
    IterGen: Iterative Semantic-Aware Structured LLM Generation with Backtracking(arxiv.org)

    1 points by tough 1y ago|0 comments
19. 649.▲
    ROCODE: Integrating Backtracking Mechanism and Program Analysis in LLMs for Code(arxiv.org)

    1 points by tough 1y ago|0 comments
20. 650.▲
    SRLCG: Self-Rectified Large-Scale Code Generation, CoT, Dynamic Backtracking(arxiv.org)

    1 points by tough 1y ago|0 comments
21. 651.▲
    by tough 1y ago

    But isn't TypeScript already a typed language to begin with?
22. 652.▲
    by tough 1y ago

    SRLCG: Self-Rectified Large-Scale Code Generation with Multidimensional Chain-of-Thought and Dynamic Backtracking https://arxiv.org/abs/2504.00532 IterGen: Iterative Semantic-aware Structured LLM Generation with Backtr
23. 653.▲
    CRANE: Reasoning with Constrained LLM Generation(arxiv.org)

    1 points by tough 1y ago|0 comments
24. 654.▲
    Type-constrained code generation with language models(arxiv.org)

    257 points by tough 1y ago|127 comments
25. 655.▲
    by tough 1y ago

    I don't always test but adding a lil test after finding and fixing a bug so you don't end up there again a second time is a great practice
26. 656.▲
    by tough 1y ago

    Some older discussions about this: https://news.ycombinator.com/item?id=40419289
27. 657.▲
    by tough 1y ago

    not the gp but found this https://github.com/ggml-org/whisper.cpp/blob/master/models/c...
28. 658.▲
    by tough 1y ago

    SmolVLM is from huggingface team cool to see people doing stuff with smaller models https://huggingface.co/blog/smolvlm https://arxiv.org/abs/2504.05299
29. 659.▲
    by tough 1y ago

    I hear you, but 10y of MacOS usage habit will make that to you, it's easier for me to work around MacOS quirks and Apple's authoritarianism than it is to try and get a Linux distro I like to work perfectly for me for more than 6 m
30. 660.▲
    by tough 1y ago

    PRESS PEDAL CAR STOPS DIDNT SHIFT UP