arcanemachiner

I think an important distinction being missed by those relying to you is that you seem to be saying that A/B testing is unethical when it is used to optimize human resource extraction (i.e. marketing).

(I have this weird feeling that you're not upset with blue/green deployments...)

Pretty sure those are just summarized traces, which obscure the valuable reasoning traces used by the actual model.

Get yourself an OpenCode Go subscription and give DeepSeek Flash 4.1 a shot.

A common tactic is to used a big brain model like Opus for planning and reviewing, and a cheaper model for execution.

Since Claude Code is a harness that can be made to work with (pretty much?) any model, the answer to the question you have asked is: Claude Code

Non-pedantic answer: I totally agree with you. Opus 5.5 is totally knocking it out of the park IMO.

An interesting conspiracy theory, as long as you're honest about what it is.

dOctument?

You can't just leave us hanging like that!

They should put data centers on the moon and call it Project Spaceshot.

Well, so much for that discussion.

There's also Aurora Droid, last time I checked.

You say that like it's a bad thing; I narrowly escaped serious use of VSCode thanks to this fact.

That, and nobody knows what you're talking about.

I had a local model help me to revert a bad kernel last year. I think it was Qwen-3-Coder, and I had no GUI and no network.

Wait, so you guys don't anonymize the user data, then train on it after it's been sanitized? I thought this was done to some degree or another.

So what is the value prop then? Just basic supply and demand?

FWIW I have definitely noticed OpenAI's emphasis on efficiency and value in the last year, so that part isn't new to me... I just thought there was more to it then that.

What goes through my mind is some combination of "Holy shit, this technology is incredible." and "Holy shit, this technology is retarded." and "If I don't stay on top of this shit, I'm going to lose my fucking job.".

On the other side of the fence, the aversion to using the tool seems equally cultish, like not working on the Sabbath.

If it makes you feel any better, I don't think this tech is "AGI" and that it seems to have some fundamental limitations that make it perform worse than a 3-year-old can perform on many tasks. But it's foolish to avoid something that could cause you to be replaced by some idiot that can do untold damage with just a few prompts, all because you didn't (assuming you work in tech) want to stay on top of the tools available to you.

This is a genuinely incomprehensible take to me. You damn well better learn to use the tool, or you will be obviated by those that do (at least, as long as your job depends on it). Linux or no Linux.

Also, you are now using the tool that is best suited to "use AI". You will hit those rough patches, and it's good to know how to work through them. But sometimes there is just no time for such bullshit, and you now have a tool at your disposal that can just sweep these problems away. It is not without its downsides, but this is a clear "dumping the baby out with the bathwater" scenario.

Your post has the same energy of the old codgers who won't "learn computers" and impotently swim against the tide. It's fine if you want to do that (you do you), but you are swimming in the wrong direction IMO.

(If you don't work in tech, the magnitude of my recommendations would soften towards you in particular, but the velocity would remain the same.)

Should is a four-letter word, my friend. The sharp edges are there whether you want them to be or not.

OK, good luck with that attitude. Until the fabled Year of the Linux Desktop finally arrives, you can use the available technology to solve the problem in a practical manner.

Interesting, I have noticed so such collapse.

Have you ruled out the possibility that your system prompt, AGENTS.md, or increasing codebase complexity are not to blame?

> half the price of GPT-5.6 Luna

Half the price when it launched, or after the price dropped by 75%?

> I dont know how they make money here

I assume it's a subsidy to get more training data.

EDIT: Okay downvoters, what's your take on why they're giving away Luna for so cheap?

These sharp edges are a lot less annoying with agents to help solve the problem.

> The Computer is no longer Personal. I expect my personal computer to only do things that I choose to invoke.

The internet and its consequences have been a disaster for the personal computer user.

Non-macOS user here. Does macOS not let you run things as root?

Dollars to donuts, they are speculating, and not privy to inside information on the topic.

However, I believe that runtime model quantization is possible with some publicly-available inference engines (e.g. vLLM), so its not beyond belief that the closed labs do quantize at runtime, either to allocate compute, or to nudge users towards a preferred model (e.g. make the incumbent model dumber to push people to use the latest-and-greatest model, or vice versa to ease the load on the latest model, which is typically larger than the old one).

The only revenue model I for this is ads (like AI Stupid Level[0]). Or as a loss leader to get eyeballs to your service (like Margin Lab[1]).

EDIT: I forgot (and am shocked) that HN still doesn't seem to support Markdown-style links.

[0] https://aistupidlevel.info/

[1] https://marginlab.ai/trackers/claude-code/

I think the answer is not to debug the code, but, when possible, to debug the outputs. The code may be considered to be a black box much of the time. (This is much more true for my hobby projects than my work projects.)

government != population

> Android wouldn't be what it is if it wasn't open source.

Yes, and now that they've achieved market saturation, it's time to pull up the ladder.

Thanks for all the free work, suckers!

No thanks, I'm not tuning a bunch of files just for things to break when I switch models or a new one comes out.

I'll just use my one-size-fits-all AGENTS.md file and tweak it when the one of the clankers screw up. I don't have time for such busywork.

Actually, I will append extra rules to CLAUDE.md (which imports AGENTS.md) since there is a hook there, and Claude has its own foibles. So I'll backpedal a bit there.