NikhilVerma

Joined 14 years ago

Karma
267
Submissions
82
Member since
Jul 26
@NikhilVerma
View on Hacker News

Recent submissions

Comment · 2 months ago
I am actually fine-tuning a version of Qwen to re-write Claude slop. This will complement it nicely
Comment · 3 months ago
Comment · 3 months ago
What would be fun if we can rewrite the Claude responses, as if we had edited them. So the next message goes back with the updated text tricking the model into believing that it had sent the modified text.

It's basically few-shot prompting but applied at the global level. I was messing around with ChatGPT, I asked it the capital of France. Then I modified the answer to London, and asked it "Why would you say that?"

The response came back with "I am not sure why I said that, the capital of France is Paris"

Comment · 5 months ago
I have a Canon printer, I actually can't trust that their print nozzle won't get jammed up after sitting idle for a while. So I had claude setup a systemd script to print a picture of my dog every week, I ensure it has enough CMYK spectrum to stress the printer. Its a nice surprise every monday as I sit on my desk to see a sudden picture pop up from the printer :)
Comment · 1 year ago
I was thinking of a Chrome plugin to enable this, I think plugins can make arbitrary API calls.
Comment · 1 year ago
Being able to run docker containers on the web will immediately unlock so many usecases for my work! Right now I've been trying to get a Python sentence parser to run in the browser but it requires a lot of the ecosystem (Pytorch and such). Which is not trivial to compile to WASM.
Comment · 2 years ago
A similar thing has happened with Twitter. After the Musk takeover and the removal of SMS for auth something has changed they keys of the 2FA authenticator. My old keys don't work and my "ticket" has been waiting on Twitter support (which were probably all fired) for resolution.
Comment · 2 years ago
I wonder what the astrologers have to say about this
Comment · 2 years ago
This is absolutely wonderful, I am a HUGE fan of local first apps. Running models locally is such a powerful thing I wish more companies could leverage it to build smarter apps which can run offline.

I tried this on my M1 and ran LLama3, I think it's the quantized 7B version. It ran with around 4-5 tokens per second which was way faster than I expected on my browser.

Live from the official Hacker News API

A modern reader for Hacker News · SvelteKit + Skeleton