Random Libertarian Tech Lead

Shipping to prod in a $100M+ ARR unicorn startup w/ local AI.

4x DGX Spark (GLM 5.3 Flash), 1x M5 Max 128GB (Qwen 3.8 Flash Next), 1x RTX 5090 (Qwen 3.8 27B).

[Random Libertarian Tech Lead @someRandomDev5]

@kyletruong @taycaldwell Regardless of the accuracy of the claim itself, TypeScript is not a product.

[Random Libertarian Tech Lead @someRandomDev5]

@0xSero @mobilekjk It's definitely real. You might skew towards a little bit of a communist "post money psychosis" if you keep receiving too many handouts though.

Pretending money isn't real is like pretending NFTs were ever worth anything.

[0xSero @0xSero]

Guys I can’t believe it.

I got another 10,000$ donation!

We are at 25,000$ cash and 50k+ in compute credits. I can’t believe this generosity makes my heart melt.

This whole journey has made me so grateful and was humbling

[Random Libertarian Tech Lead @someRandomDev5]

@alexdolbun @saltyph Until then my home lab has only 10% of this speed due to my dependence on particleboard walls and roofs

[Random Libertarian Tech Lead @someRandomDev5]

@alexdolbun @saltyph So what you're telling me is that I can sell my house and be homeless but with 2,000 TPS DeepSeek

[Random Libertarian Tech Lead @someRandomDev5]

@ay_ushr FWIW I'm building an interface for myself discards the "chat-based interface" and instead builds on how you'd work with whiteboards and mind maps. The AI agent you're working with is preemptive and asynchronous, rather than reactive and synchronous.

[Random Libertarian Tech Lead @someRandomDev5]

@ay_ushr Is it "agent work is..." (e.g. is it inherent to agents) or is it "the current way I'm working with agents is..." (e.g. is it perhaps a consequence of you and/or others not yet discovering the ideal way to work with them yet)?

[Random Libertarian Tech Lead @someRandomDev5]

@Jawwwwwsh @septumfunky @_eramor ^ Responding to someone who is correcting an annoying transgender pedant.

[Random Libertarian Tech Lead @someRandomDev5]

@liuliu Once that's in a satisfying place, then I'll turn my attention to Inkling... unless Qwen 4 becomes publicly available around the same time and consumes my attention.

[Random Libertarian Tech Lead @someRandomDev5]

My 4x Sparks are still tied up trying to meet/exceed the SOTA inference stack's speeds for DS4.1F, so I can't actually run it right now.

I'm actually pretty interested in the idea of switching between it and DS4.1F, since an expanded panel of benchmarks showed a really interesting pattern where most of DS4.1F's weaknesses are Inkling 2 Small's strengths (e.g., Design Arena), and vice-versa (DS4.1F seems to be a stronger everyday agentic coding model).

[Random Libertarian Tech Lead @someRandomDev5]

@septumfunky @_eramor @Jawwwwwsh And yeah, I'll also tell you that Manhattan, Brooklyn, Queens, and the Bronx are all "NYC" too. Ever been there? Do you doubt it? Each has a very distinct feel from one another, but also many things in common.

[Random Libertarian Tech Lead @someRandomDev5]

@septumfunky @_eramor @Jawwwwwsh Saying "one area is a bit different than the other" is entirely beside the point. The waterfront feels different from 3rd street; They're both "Seattle" despite this. It doesn't become "not Seattle" just because you can draw fine distinctions between adjacent subregions lol.

[Random Libertarian Tech Lead @someRandomDev5]

@septumfunky @_eramor @Jawwwwwsh Lmao, including current traffic, it's a 15 minute drive from the very center of downtown Seattle to the very center of Redmond. Even closer if you're going from the edge of one to the edge of the other. Same MSA. To anyone not being pedantic, these are the same city.

[Random Libertarian Tech Lead @someRandomDev5]

@FT_IOxCS And either way, with regard to these two specific models under comparison: It's not as if any <= 1T model (at least, in current generation / by current methods) is something you could reasonably expect to "contribute meaningfully to the frontier of math/science".

[Random Libertarian Tech Lead @someRandomDev5]

@FT_IOxCS You have the same problem with GLM 5.3 in the abstract though, or in particular, when comparing it to e.g. Astra, Fable, or whatever the peak Gemini model is today.

[Random Libertarian Tech Lead @someRandomDev5]

@nucknyan And just as with power tools, some people still use them more skillfully than others.

Using a tool that automates your work doesn't inherently drive you to create slop that you don't attend to or care for.

Using AI well is a skill and craft. Develop it.

[Random Libertarian Tech Lead @someRandomDev5]

TL;DR: If you want to have a harness that works well for ideating and iterating, think more in terms of "intent authoring" rather than "prompt engineering", and ditch the suboptimal implementation detail of "harness interfaces are just streams of text". Think about how humans actually come up with ideas and collaborate on ideas with one another. Mimic that instead. Never in my ~10+ year career in tech have I ever gone to the office in SF and seen people standing around a giant shared textbox typing out their ideas. Instead, it's always e.g. a whiteboard where people are drawing mockups, diagrams, associations between bits of information, etc.

[Random Libertarian Tech Lead @someRandomDev5]

@Biggest And in my harness, if you wanna know more about something / look it up: Just hold down the middle mouse button and speak about whatever is underneath your pointer. An agent will contextually interpret your search and give you the information you're asking for.