Hacker Newsnew | past | comments | ask | show | jobs | submit | cvzakharchenko's commentslogin

Interesting. I know people do this manually. But are there agentic coding tools that actually automate this approach?

Claude Code has /init and /compact that do this. It doesn’t recreate the context as-is, but creates a context that is presumed to be functionally equivalent. I find that’s not the case and that building up from very little stored context and a lot of specialised dialogue works better.

I've seen this behavior with Cursor, Windsurf, and Amazon Q. It normally only does it for very large requests from what I've seen.

So it's not STT -> LLM -> TTS? If I scream Chewbacca noises as input, will the model recognize it as nonsense, or will it interpret it with some lousy STT as some random words?


It's not, but it probably won't recognize it as nonsense. According to the paper,

> we construct a dataset named InstructS2S-200K by rewriting existing text instruction data and performing speech synthesis

It has only been trained on questions spoken by TTS, it has never seen (heard) nonsense. Most likely it'll just hallucinate that you asked some question and it'll generate some answer instead of asking if you're good. There's just not many audio datasets with real voices, there's no audio version of StackOverflow to be scraped


What about every movie that's been made? Although it might need to stick to those more than 100 yrs old to avoid copyright law?


I used to have fun with that. Set Google Translate to Chinese (Or some other language I don't speak, though tonal languages seemed to work better), make some vague noises into it, and get out coherent but crazy phrases in English.


Oh, I’m yet to find a good alternative to Cursor’s RAG-powered side chat. It helps me work with huge codebases so much. Tried Continue, but it’s very unstable, and doesn’t work as well. Would prefer a command line solution, vscode plugin is the next choice, having a separate editor is not ideal, but I’m glad there’s some competition.


Aider is AI pair programming in your terminal.

Aider lets you pair program with LLMs, to edit code in your local git repository. Start a new project or work with an existing git repo. Aider works best with GPT-4o & Claude 3.5 Sonnet and can connect to almost any LLM.

https://aider.chat/

See LLM Leaderboards as well: https://aider.chat/docs/leaderboards/


Thanks! I use Aider with Claude 3.5 Sonnet on smaller projects sometimes, and it's really fun and helpful when it can put a whole repo map into the LLM's context.


> Would prefer a command line solution

I haven't tried this myself so I apologize if it ends up being bad, but I've seen Aider [0] get linked a few times from people who wanted a CLI solution for AI code completion.

[0]: https://aider.chat/


Aider is really cool for small projects, but it builds a repo map instead of using RAG. That works on small codebases, but totally fails to be useful on large ones.


How large?

In use, treesitter-derived approach seems far effective than rag.


>5k source files. They don't fit into the context. I know I can limit what is sent, and I can attach files in the Aider chat myself. But this is not perfect for making an LLM answer questions about a codebase when I don't know much context beforehand. With Cursor, I can just do "@codebase How is a %feature% implemented?", and it's very quick and often helpful with a couple of follow-ups.


Not sure if this is what you'd like, but I remember this repo from some time ago https://github.com/Storia-AI/sage

Storia-AI/sage: Chat with any codebase with 2 commands


Thanks for the suggestion! It needs some work to set up, and it looks like it only works on Github repos. Also, to work with non-local LLMs, you can only use Pinecone for vector storage. I might have misunderstood something, but I will check it out again later.


I gave the SVG unicorn back to GPT-4 and asked it, what is it.

It recognized a head, eyes, body and legs. But it didn't recognize the unicorn.

https://chat.openai.com/share/5409b417-b883-429f-893e-abe3d6...


The OP has a slightly different prompt -- so this answer wouldn't qualify.

"System: You are a helpful assistant that generates SVG drawings. You respond only with SVG. You do not respond with text.

User: Draw a unicorn in SVG format. Dimensions: 500x500. Respond ONLY with a single SVG string. Do not respond with conversation or codeblocks."


You need to have quite a bit of imagination to say that is a unicorn, but less than you would need for other days.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: