I've always found Crazy Horse to be pretty inspiring. To work your entire life on a project knowing you will not be the one who finishes it but to start anyway

Gpt-oss is outdated. So many better oss models

What about Qwen 3.5 122B how does it score against that?

My thoughts exactly

Question for you, do you have to be logged into claude code to use the telegram features? Also can you still use non claude models

When telling my dog to get into the tub to take a bath, he looks confused and doesn't understand.

However when I tell him to get out as it's over his intelligence seems to soar and he quickly hops out...

I'm just happy to see the community at large making more use of dotnet and not just using node lol

I've loved it for dev, and have just played with pushing out to coolify, my prod stuff all goes out to different azure and kubernetes environments via github actions so I'm just using the dev portions and the aspire mcp which has been nice

Interesting, I'd read a reddit thread with users saying it was quicker but haven't yet tested it. Question for you, how are you running your model? I've been using a docker with the ghcr.io/ggml-org/llama.cpp:full-vulkan image

What quantization are you using?

Isn't the cost in tokens about 10x between them

A mouth full

Same but it's sooo slow, going to try M2.5 at a 3 bit quantization

I'm still sad that Nuxt is part of Vercel now

I think Billionaires have found that buying politicians helps their bottom line

I think this is one of the concerns and that is basically that you end up with so much code to review that things slip by. I've been slowly implementing Codeowners where specific libraries, apis etc. have a higher review level

I've liked the automated code reviews. I've had some Juniorish devs introduce N+1 errors in the past and pulled too much data and now my auto reviewer spots most of those before I review their code

That looks like Washington and BC and not Oregon

I have switched to mostly letting AI write my code and am outputting more but I also have seen some cringe code patterns that would die when you try and scale it that I've edited out or re-worked

Which size?

I've been noting that pretty much every article about these are from people who've not installed or used them

Rerun and add nullclaw

Tried NanoClaw but didn't really get the results I wanted. Going to try NullClaw next

Have you tried it? I like that it's small and am wanting to give it a spin

Tried it yet?

Never stop never stopping

What I really want to know is who is calling out to AI to get a vector embeddings for every time someone types something in to search. That seems so impractical

That is pretty awesome

That's messed up

What I've wondered about is the death by a thousand cuts. A lot of small companies are going to pop up, slightly cheaper knockoffs that don't scale well but get a few clients here and there