Sean @TuringMachinery

Professional software engineer/architect. Unprofessional welder. I solve hard problems. Joined November 2025

Tweets

26
Followers

3
Following

38
Likes

21

Sean @TuringMachinery

8 months ago

@TheAhmadOsman Which Mac users are pushing 1.58-bit models?

0 0 0 96 0

View Details

Alpha testing has begun! AI inference is expensive. ClankerBuddy gives you "free" access to an OpenAI-compatible API in a crowdsourced compute environment. Share your compute with others while they share with you. clankerbuddy.com

0 0 1 17 0

View Details

Sean @TuringMachinery

9 months ago

🚨BREAKING NEWS🚨: Claude Haiku 6.9 dropping tomorrow! Anthropic's new frontier model achieves an industry-leading 97/100 in horseback riding benchmarks while increasing crop yields by over 37%. It's so over.

0 0 1 37 0

View Details

Sean @TuringMachinery

9 months ago

@rockorager 100% agree on the feedback-loop benefits of a compiled language like Zig. My preference would unironically be Haskell. But be practical, it needs to get it right on the first-try most of the time. Otherwise, the helpful feedback loop consumes too much context.

0 0 0 76 0

View Details

Sean @TuringMachinery

9 months ago

@rockorager I'm not sure I follow. A model trained on JS will more successfully generate JS than it will Zig. If the model was trained on Zig, it would be more successful generating Zig. Again, I'm not saying I like it. I'd love to see a paradigm shift that trains on better languages.

1 0 0 83 0

View Details

Sean @TuringMachinery

9 months ago

@theo I'd like to see more engineers build their own personal dev tooling. The hype cycle and tribalism is absurd. We're better off building our own tools to suit our individual needs and preferences.

0 0 1 221 0

View Details

Sean @TuringMachinery

9 months ago

I've been quietly cooking up a new thing that's meant to address many of the problems with "vibe coding" as it stands today. It takes a bit of a different approach: Constrain the AI to only what is correct, secure, and simple. Instead of letting an AI implement an entire system, the system is a managed runtime in which the AI can generate specific, isolated components: Actions, Entities, and Apps. Actions are stateless, Entities are stateful, and Apps provide mini visual frontends. They all work together through type-safe, schema-driven interfaces. More details to follow (eventually).

0 0 1 20 0

View Details

Sean @TuringMachinery

9 months ago

@Angaisb_ None of them will be very good. I assume you mean coding models since you mentioned GLM. I've had moderate success with cogito, qwen2.5-coder, and granite-code. Aim for something around 7b, and set your expectations to bare minimum.

0 0 1 2K 0

View Details

Sean @TuringMachinery

9 months ago

@iyoushetwt Raw TCP sockets

0 0 0 10 0

View Details

Sean @TuringMachinery

9 months ago

@ThePrimeagen Skill Issues v2

0 0 1 208 0

View Details

Sean @TuringMachinery

9 months ago

@_chenglou Sure, as long as we agree that impact can be negative

0 0 0 56 0

View Details

Sean @TuringMachinery

9 months ago

@svpino Hiring top talent takes them out of the job market, which minimizes Anthropic's risk exposure.

0 0 0 262 0

View Details

Sean @TuringMachinery

9 months ago

@mert We should retire all of the web2 projects that are cosplaying as web3

0 0 1 212 1

View Details

Sean @TuringMachinery

9 months ago

@jonathanbrnd Not with current approaches. It works wonders when the entire project fits comfortably within LLM context. It deteriorates quickly once the project no longer fits. MVP is pretty much the limit.

0 0 1 231 1

View Details

Sean @TuringMachinery

9 months ago

@geerlingguy ElevenLabs' pricing can't be justified. I can achieve reasonable quality locally with similar speeds on 8-year old hardware for pennies.

1 0 1 919 0

View Details

Sean @TuringMachinery

9 months ago

@SumitM_X "Doesn't use" would be one thing. "Doesn't know" is a massive red flag.

0 0 0 2K 0

View Details

Sean @TuringMachinery

9 months ago

How big of a tractor do I need to start engagement farming?

1 0 0 21 0

View Details

Sean @TuringMachinery

9 months ago

@unclebobmartin This is the inflection point for the current AI-assisted dev tooling. The tools show promise at small scale but fall apart under complexity. LLMs aren't designed to work with massive code repositories.