Martian

dev | building things |
Avid Reader & Fitness Enthusiast

[Martian @martian75007]

@dalu_hey does it include the IDE?

[Martian @martian75007]

@Thomas_TheGuy im running arch and the aur for antigravity is no more. had to install the tarball and it only has the "Antigravity 2.0" and not the IDE.
Genuinely, DX gone to shit.
Found another AUR package called "antigravity-ide" not sure how good it is, gotta try it out.

[Martian @martian75007]

@nicoalbanese10 meanwhile, here's how i do it now. i have an upstash redis instant, the key is a uuid for a chat thread, the value is a json object with the data you want, (in my case the user id, created at time, and array of Message type from ai sdk). it might not be the best solution.

[Martian @martian75007]

what is the best way to store chat history of an AI chat which uses vercel ai sdk to interact with llm?

[Martian @martian75007]

@nicoalbanese10 i am looking forward to that.

[Martian @martian75007]

@IroncladDev @jordwalke this is friggin peak

[Martian @martian75007]

you feel miserable when you do leg day but once its done, you feel awesome.
its the same with all hard things.

[Martian @martian75007]

its like @bigboxSWE has a telescope into my life.

[Martian @martian75007]

im the kind who likes to write his own code, instead of letting AI do it for you. but once in a while i get stuck, i cant get something done. AI tools are really good at getting you out of these zones.

[Martian @martian75007]

ive always wondered how they test prompt changes and tweaks that you cant test traditionally. we are about to find the answer to that.

[Visual Studio Code @code]

Today, we're announcing plans to make VS Code an open source AI editor.

We believe AI development should stay true to VS Code's core principles: open, collaborative, and community-driven. Let's build the future of software development together.

https://t.co/C3ffio6X88

[Martian @martian75007]

@theo he doesn't know what he's talking about

[Martian @martian75007]

👽

[Martian @martian75007]

- once i get the nearest k values. i retrieve the text chunks those embedded values correspond to, from the metadata which holds the text to embedded value.
- construct a prompt with these retrieved chunks and the user's query.
- generate completion by prompting an llm.

[Martian @martian75007]

things have been developing unbelievably fast in ai. ik im too late. but ive been messing around with llms and stuff. especially in resource constrained devices. im gonna maintain this thread to keep a record of what i do.

[Martian @martian75007]

here's how i do it now:
- load the files
- chunk it (300 characters)
- embed it using an embedding model (all-MiniLM-L6-v2)
- save these embeddings and save it to FAISS (using IndexFlatL2)
- when the user types something, i embed that as well and query FAISS and get nearest k val

[Martian @martian75007]

@theo i have mixed feelings. lets say i am working on a project and i have multiple chat sessions for that project. i want all those separate chat sessions to have a shared memory about the project. but i dont want that project memory to affect a completely different chat or a new chat

[Martian @martian75007]

@appwrite i cant upload my resume while applying for your engineering internship program.

[Martian @martian75007]

it is unbelievable that many people think nextjs is a frontend framework.

[Martian @martian75007]

and lots of other small things that will definitely compound over the long run.

it was an active month and i had a blast doing things i like everyday. gonna carry that momentum and have another highly active month.

[Martian @martian75007]

here's all i did last month:
- participated in a 24hour full stack web development hackathon.
- took part in events in gravitas at vit vellore.
- participated in a gameathon and won first place with my team.
- consistently went to the gym and ate healthy.