Chris

Devops/tech consulting and available for projects! Haskell/Nix ideally. Running some projects at -$20 MMR.

[Chris @chrisczynski]

Cabal's --enable-multi-repl finally works great on GHC 9.14 :D

Was always an annoying issue to not be able to reload changes on libs on the same project.

[Chris @chrisczynski]

@Howaboua Tricky to know what we're looking at here?

[Chris @chrisczynski]

Making your own coding agent/harness feels like real life Factorio...

Anyway, agent or harness?

[Chris @chrisczynski]

@ProgrammerDude There really should be an approval request for whatever enhancement.

Typo spotted. Fix? y/n
Delegate to 3 subagents? y/n

But giving us control is the last thing they want to do.

[Chris @chrisczynski]

Not as intelligent, but insane speed and diligence from Grok 2.5 composer build fast (or whatever the hell they call the model from Grok). :D

Get a strong model to make a plan.md, then just ask grok to implement it. Great for tedious refactors that touch many files.

[Chris @chrisczynski]

Gemini 3.5 Flash appears to be insanely slow this morning. Annoying because speed is exactly what I use this model for.... Not enough capacity I'm guessing.

Just me?

[Chris @chrisczynski]

Confirmed

[Chris @chrisczynski]

Gemini 3.5 Flash appears to be insanely slow this morning. Annoying because speed is exactly what I use this model for.... Not enough capacity I'm guessing.

Just me?

[Chris @chrisczynski]

oh no

[Chris @chrisczynski]

Hetzner has doubled it's prices overnight?? Not just for US servers but the usual EU as well.

[@levelsio @levelsio]

Hetzner just 2.5x'd their prices for US servers

Crazy but understandable as the AI companies are pushing prices of everything up: RAM, HDDs, SSDs, CPUs, GPUs, and even competing with Hetzner for land to build data centers in US!

[Chris @chrisczynski]

https://t.co/m1uj9SeZV2

I made a little tool to help with this exact problem. Saves me having to vibe code this table to each and every project making requests to an LLM.

[@levelsio @levelsio]

With xAI, OpenAI, Anthropic's models updating regularly, do I have to keep changing the 'model'=>'gpt-4-1-fast-non-reasoning' name for example in my code on ALL my sites for the next 100 years?

Or can I just say 'model'=> use best or something?

And no I don't wanna use OpenRouter, I just wanna use whatever is the best but affordable model

There's so many places in my code on so many of my sites where I have to change this and they do deprecate models regularly so it's like breaking changes

[Chris @chrisczynski]

grok, super grok, grok build (the cli) grok build 0.1 (the model), grok build fast, grok 4.3, grok composer 2.5 fast... @grok I think some better naming would go a long way here.

[Chris @chrisczynski]

"litellm" has over 1 MILLION lines of code... No wonder it got hacked.

[Chris @chrisczynski]

Claude Fable 5 is cheaper than GPT 4 in 2023 on the API... Just 3 years apart. Where will be in another 3 years (2029)? Hopefully in a very cool wonderful world.

[Chris @chrisczynski]

Everytime your service goes down, a user is 2s away from "service X alternative". Which is why those keywords seem to be a great AI/SEO target.

[Chris @chrisczynski]

@leostera So something like literate programming? Making asking an agent to make something like https://t.co/abaGz0eWzq or https://t.co/9jo6RjTfK9 out of each PR?

[Chris @chrisczynski]

The $10 opencode plan has been sufficient for my side projects the past 30 days. 10 commits/day roughly. You really don't need Claude or Codex...

Use my ref for for a free $5 credit :)
https://t.co/Sf1XfLbJ9a

[Chris @chrisczynski]

@kennethreitz42 @KatiMichel Definitely the algo has changed in these past years. Quite sad. It felt much more varied and organic in the past.

[Chris @chrisczynski]

Everytime I update @opencode , something breaks... Like literally you can't even run the app... https://t.co/aJPyRfTxUf

[Chris @chrisczynski]

@GergelyOrosz So there is this AI native language called Haskell and it compresses the context of source code into higher token levels. 10x token savings. And smaller local on device GHC model that near instantly analyzes your source code.

[Chris @chrisczynski]

Randomly noticed a 100% cpu process which turns out to be a @github background task? Our CPUs are no longer safe from the common web... Can an alternative please step up already?