[naasking]

Learning how to use AI effectively was the learning opportunity here, what was created is completely incidental. You're effectively obsessing over programming languages obscuring the machine code that actually runs. "Imagine all the missed learning opportunity of digging into all that machine code!"

Sure, but also, who cares? The machine code is completely incidental for most purposes.

[xendo]

I work with AI everyday, despite what many people suggest there is so little to learn. After a couple of hours you are good to go. You don't even need gstack.

[elliotbnvl]

This is patently false. I work with and on AI every day at multiple levels of the stack, and every day I'm learning massive new swathes of information. I'm honestly shocked how deep the field goes and how much more effective you can be with time. The floor is falling and the ceiling is rising and the gap between them is widening every day.

[skydhash]

That would be more convincing if you put up two or more examples of what is there to learn.

[elliotbnvl]

Go off and run a comparison of Qwen 3.6 27B and GLM 5.1 GGUF (https://huggingface.co/ubergarm/GLM-5.1-GGUF) at IQ2_KL 261.988 GiB (2.985 BPW) and let me know if you learn anything.

Or maybe just compare Hermes vs OpenClaw for long-horizon personal agentic tasks. Which one performs better in offline inference personal finance analysis tasks?

Or read up on how the `/code-review` workflow works in Opus 4.8 and give me a guess as to how long it'll take Codex to implement it and which tool would be more appropriate for your engineering team (don't forget to include enterprise API token costs in workflows – it can spin up 100 agents in thirty seconds).

If you can figure out how to secure agents with simultaneous access to personal data and the internet to run unsupervised while avoiding the lethal trifecta (Willison, 2025) let me know.

[skydhash]

> Go off and run a comparison of Qwen 3.6 27B and GLM 5.1 GGUF

You may as well ask to run a comparison between gnu libc 2.42 and musl 1.2.5.

> Hermes vs OpenClaw for long-horizon personal agentic tasks. Which one performs better in offline inference personal finance analysis tasks

What are those tasks? This and the paragraph just after seems very much like a XY problem where all the energy is focusing on resolving the Y, not the X. It's like discussing how we can reach the moon using cannons.

> If you can figure out how to secure agents with simultaneous access to personal data and the internet to run unsupervised while avoiding the lethal trifecta (Willison, 2025) let me know.

If you can figure out how to run user submitted JavaScript inside a webpage with access to the internet and other user personal data, you will have your answer. There's a reason we escape user input before rendering it within the browser. The browser is an executing agent and it doesn't differentiate between your markup and other data you choose to embed in it. Same things happens with the processor if you choose to mix input data with executable code.

[elliotbnvl]

> You may as well ask to run a comparison between gnu libc 2.42 and musl 1.2.5.

Telling me you wouldn't learn anything from this?

> What are those tasks? This and the paragraph just after seems very much like a XY problem where all the energy is focusing on resolving the Y, not the X. It's like discussing how we can reach the moon using cannons.

Or like how we can get from A to B without horses.

It's a different world, one worth learning about. If these tasks don't at least arouse your interest, nothing I can say will help you.

[naasking]

Using AI effectively for long horizon tasks, like maintaining a large codebase, is a wide open field. No single AI is good at it autonomously. That means achieving the right balance of testing, formal specification of pre/post-conditions and invariants and manual review.

It's like having a naive but super knowledgeable junior developer starting under you. It's obvious you'd learn a lot in how to communicate, framing, specifications, and what kind of follow-up you'd need to do to ensure good results.

[xendo]

Even with examples it's still not convincing. I'm working on real products so I don't have time to waste comparing models that won't be relevant next month.

[xendo]

Maybe it depends on the task, but the biggest productivity gains are from boiler plate generation, and there it's as easy as "generate me the boiler plate". Even if you can learn some very specific workflows today they would be model dependent and mostly obsolete within a month or two.

[naasking]

Unless you just happen to work in a domain where the code you generate every day is very common in the AI training data, this isn't true.