siquick

amazing thanks

3 giorni fadiscuss

Very nice - would love to see the ability to open a Dockerfile directly in the UI to build/run it.

4 giorni fadiscuss

Likert scale just doesn't work in LLM evals. My idea of 3/5 is different from your 3 and definitely different from an non-deterministic system's 3.

siquick

mese scorsodiscuss

DGX Spark is really poor at inference due to the memory bandwidth so hopefully they’ve fixed that before touting this as a way to run local models.

siquick

mese scorsodiscuss

You're right, I use Brave on iOS but Safari doesn't let you choose beyond a defined list.

siquick

mese scorsodiscuss

> Only downside is the Search on mobile. You have to install an app and enable it as extension on safari. Logging in never worked and couldn't enable my premium Kagi on iPhone.

I just have Kagi set a custom search engine in mobile browser - no seperate app needed.

https://kagi.com/search?q=%s

siquick

mese scorsodiscuss

+1 for Kagi. Been a subscriber for almost 2 years and no intention of cancelling, it just works.

siquick

2 mesi fadiscuss

Solved !== Good

siquick

3 mesi fadiscuss

This is exactly my experience of working in an agency. Made worse by Figma defaulting to 1440px so every design only really works at that width.

siquick

3 mesi fadiscuss

It’s just gambling without the stigma of being called an addict.

siquick

4 mesi fadiscuss

Love what you guys have done with Unsloth. Any plans to have this running as a hosted product where Unsloth provide the GPUs?

siquick

4 mesi fadiscuss

> These days I hit the quick answer button inside Kagi more often

Just incase you didn’t know you can append ? to any query and get a quick answer straight away

siquick

4 mesi fadiscuss

This may help you work out the best quant to use for your use case.

https://www.siquick.com/blog/model-quantization-fine-tuning-...

siquick

4 mesi fadiscuss

Sure sounds like automatic weapons targeting using Anthropic models to me.

However much they try to make us think otherwise, at this point in time there’s not really any “good guys” in the AI race.

siquick

4 mesi fadiscuss

Start here

https://modal.com/docs/examples/vllm_inference

or give this a go

https://modal.com/docs/examples/opencode_server

You get $30 free credits each month on Modal which is enough to play around (i have no affiliation, just think they run a great service)

siquick

4 mesi fadiscuss

Rent a H100 on Modal which scales down to zero when not in use - you can set the time out period.

Cold boot times are around 5m but if your usage periods are predictable it can work out ok. Works out at $2 an hour.

Still far more expensive than a ChatGPT sub.

siquick

5 mesi fadiscuss

> Skills in CC have been a bit frustrating for me. They don't trigger reliably

Referencing them in AGENTS/CLAUDE.md has increased their usage for me.

siquick

5 mesi fadiscuss

This is truly great - just joined the universal fund.

I tried to signup from the nav bar but kept getting a 500 error from the sign_up.json endpoint. I had to go through the Donate flow to be able to create an account.

siquick

5 mesi fadiscuss

Misread the post - it doesn’t include the API

siquick

5 mesi fadiscuss

2 weeks notice to migrate to a different style of model (“normal” 4.1-mini to reasoning 5.1) is bad form.

Submissions

A guide to model quantization in fine-tuning (and how to pick the right GGUF)

How to diagnose RAG failures from traces

Discontinuation and reinitiation of dual-labeled GLP-1 receptor agonists

[untitled]

Andrew Ng's letter to an 18 y.o. worried about AI

AI: Boom or Bubble? A live, point-in-time dashboard

comments