You put a made-up quote in my mouth and then got angry at it. I said commits should be judged on their merits. I said nothing about profits or ignoring harm.

Reading comprehension not your strong point, huh?

Tell me, is it PHP or Javascript you're l33t in? Actually, don't bother... plonk.

Simple questions to an LLM like: “review this procedure, is it internally self consistent” will find issues, real issues - usually minor - but undeniable when you read the review, the problem it describes is absolutely present in the document, page 14 says “X” and page 23 says “not X” and the context is basically identical - so which are we really supposed to do: “X” or “not X”? Well, that depends. Depends on what? Additional context and detail that the already too long 50 page procedure doesn’t address.

lol and I find issues like this in LLM output of only 1 or 2 pages. LLMs can't even be consistent across paragraphs, much less hundreds of pages of context window. what an insane claim.

Sounds nice. We’ve got ~500 employees at our “site” (which is now distributed across 3+ states with remote workers), of those, about 8 do software development, so developers end up doing virtually all of the translation of customer “needs” to specifications / requirements. We also have a marketing department that changes their mind about twice as fast as the product development / delivery cycle.

you're acting like the solution here is LLMs, when the solution is just hiring more people. LLMs aren't 'solving' anything, they're just masking the problem and allowing the millionaires at the top to keep making tons of money.

lol and I find issues like this in LLM output of only 1 or 2 pages.

And there's the distinction: nobody is perfect, especially not LLM agents, but... when a critic: human or LLM finds a flaw and points it out: there it is, no denying that is a flaw.

Asking LLMs to find flaws, then verifying those flaws are real, is more powerful than just looking for flaws without the tool.

you’re acting like the solution here is LLMs

No, you're imagining that's what I wrote. LLMs are a tool, not a solution. A helpful tool, very helpful in some situations. The real problem that people and LLMs struggle with is: what should we be doing in the first place? LLMs don't solve that, they do accelerate the making of a software thing when you know what you want that software thing to be. They also accelerating the finding of flaws during the process, enabling cleaning those flaws out faster.

just masking the problem and allowing the millionaires at the top to keep making tons of money.

That's not a problem for tools to solve, that's a societal problem that goes back to - at least - the start of agriculture.

I hope you get better management that actually understands how much data one person can process.

Cynically, management doesn't care if we read or understand the material, only that we click a button that documents (records) that we have "been trained on it" and acknowledge that.

In reality, after decades of exposure to the same basic concepts, most people who are "trained" on this material do know - vaguely - what's in the documents, and the real function of the training is to remind you that the procedures are there and when they are applicable to what you are doing (rarely, in my job function) then you should pull the official, current, procedure from the official document repository and reference that as a work instruction while you do what it applies to. Part of why the procedures are in such bad shape is that people rarely do exercise them like that, so flaws remain un-noticed for long stretches of time.

I hope that one day, the world can openly accept that people are not perfect executors of written, spoken, or graven in stone instructions and that that is O.K. and we can still do a good job with our flaws. Some of the procedural safeguards are implicitly acknowledging that without coming right out and acknowledging that. I suspect that day will come after I retire, if ever, and I may work for another 30 years.

Most ai output has been watermarked since I think 2023, Anthropic is just more open about it.

I am pretty sure this still produces too many false positives when applied to code because of the way these watermarks are implemented. I am not 100% sure on that though. It could be used on commit messages though.

I am curious about the downvotes; what I said is factual. LLMs use random numbers in deciding output, so they are not deterministic.

Nah the EU legislation made it so that it has to be watermarked (LLM output was not watermarked until that passed, it was just image and music output) and discernible with a tool to tell. So you can check if something is AI generated. Anthropic says it should work on code.

you’re acting like the solution here is LLMs

No, you're imagining that's what I wrote. LLMs are a tool, not a solution.

Fine, you’re acting like the tool for the solution is an LLM.

A helpful tool, very helpful in some situations.

X to doubt. My job is currently filled with people using LLMs and most of what I’m doing is undoing their shit our detangling whatever LLM output they paste

The real problem that people and LLMs struggle with is: what should we be doing in the first place? LLMs don't solve that,

Correct

they do accelerate the making of a software thing when you know what you want that software thing to be.

Incorrect. They accelerate making a ton of code that is impossible to maintain, update, etc. If you know what the software is, it was never writing it that was the hard part. That’s why senior+ developers have been saying for decades that the hard part of the job isn’t coding. It’s everything else.

They also accelerating the finding of flaws during the process, enabling cleaning those flaws out faster.

Incorrect. Like I said, half my job is reading over extremely verbose LLM bullshit from colleagues to tell them every single thing the LLM said incorrectly. That didn’t happen before LLMs. People checked their work. LLMs don’t have a mind, they cannot check their work.

That's not a problem for tools to solve, that's a societal problem that goes back to - at least - the start of agriculture.

I didn’t say it was a problem for tools to solve! But it sure is a problem tools can cause! And that’s exactly what’s happening here. It’s making everything so incredibly worse that there is literally no justification for it!

Asking LLMs to find flaws, then verifying those flaws are real, is more powerful than just looking for flaws without the tool.

No it isn’t. Your example is so contrived. What was happening before these documents were reviewed by LLMs? Were the lines that were incorrect causing issues? If so then a human would have found it years ago. If not then they weren’t causing problems and all the LLM has done is invent work that a human now needs to clean up. The LLM didn’t solve a problem, it created a new one!

Now you need to pay a human to make sure that the LLM isn’t actually making things worse, which wasn’t even a job before! Your company that won’t pay a human to review the document in the first place is now needing more humans to clean up a mess that wasn’t a problem in the first place.

Anyone claiming to do “real programming” shouldn’t be listened to. It’s incredibly clear you don’t know what deterministic means.

They are equivalent. Your quote:

This is exactly right and I love Linus so much for cutting through the emotion every time. All he cares about every single time is if the commit is good or bad. Thats how it should be.

So the only thing that matters is that the commit is good. Not whether the commit turns on DRM for every Linux machine on the planet, not if the commit starts requiring age verification for every user, not if the commit fucking bombs every city on earth.

Your reasoning is literally ignoring harm.

I said nothing about profits or ignoring harm.

No, you implied it with a dog whistle. And now you’re pretending like you didn’t.

My job is currently filled with people

My jobs have been filled with people misusing tools for 40 years, particularly new tools and computer oriented tools - the current trend is just an extension...

They accelerate making a ton of code that is impossible to maintain, update, etc.

When used improperly, sure, see above.

If you know what the software is, it was never writing it that was the hard part.

I'll agree, but still say that LLM agents accelerate the setup of toolchains, the learning of new languages, and many other things which do accelerate the writing part so instead of having to carve out 2 weeks of quiet time to get a good proof of concept ready from concept to ready for others to give feedback on, that can now be done in a day or two.

reading over extremely verbose LLM bullshit from colleagues to tell them every single thing the LLM said incorrectly. That didn’t happen before LLMs.

Your incompetent colleagues have been given a steamshovel for their bullshit, my condolences. It's not the steamshovel's fault it can work faster than a man with a spade on a stick.

LLMs don’t have a mind,

Correct.

they cannot check their work.

Incorrect. They're actually better at finding fault than they are at creating fault-free work. But they do require direction to keep looking for faults, they'll tell you they didn't find any (true statement) but when you tell them to look again (with a clean context window), unlike people, they'll look again just as carefully as the first time - and find more faults. I went bug hunting on a nice meaty module and found and corrected over 12 rounds of faults - they were getting rather trivial toward the end, and I had to direct to fix through removal rather than addition of code whenever possible - repeatedly, but after that it was actually clean and solid.

it sure is a problem tools can cause! And that’s exactly what’s happening here.

Tools aren't causing the problem here, managers are causing the problem. Any excuse for rank and yank will be gleefully embraced to its fullest potential.

Were the lines that were incorrect causing issues?

Mostly not because mostly people gloss over the text and say "yeah, yeah, I know what you're getting at" and do what they think the page says instead of doing what the page actually says.

The LLM didn’t solve a problem, it created a new one!

If you have a road with mis-painted stripes, but it never causes a problem because everybody ignores the stripes anyway... yeah, in that sense fixing the stripes is un-necessary work...

Your company that won’t pay a human to review the document in the first place is now needing more humans to clean up a mess that wasn’t a problem in the first place.

Oh, the company pays plenty of people to create and review the documents. They have metrics for how many documents they review per quarter, and they do review those documents, and they do find these things, and other people are paid to revise the documents again with metrics for their performance, and we all get trained on the revisions as they come out - at a remarkably regular pace when you look across a scale of tens of years - and that training accounts for about 5% of my weekly chargeable time accounting.

The difference is: the LLMs are finding problems faster, all at once, and the "continuous improvement process" is a little bewildered about what to do with all these, valid, observations coming out so much faster than they used to. We're just starting to think about how to come to terms with that, and in the meantime the process rolls on revising documents at about the same pace as always, because if we fixed them all at once it would put a kink in the system, like a python swallowing a buffalo.

I don't understand this whole idiotic concept of "don't use AI for X".

I do. People are stupid, bandwagoning morons.