LLM output makes everything stupider, even LLMs.

Sadly, often.

It is sticking the most common content into a blender and dispensing the result according to a few tweaks. I have seen a lot of people eat that up.

Same reason why there’s so much money in low-effort mobile gaming.

It’s one of the things that really sucks, people with creative talent that would otherwise be rewarded and have their vision shared for lack of too many other options being drowned out by a flood of no-effort slop that the masses are just fine with.

For reference, that was the year Vista came out with a minimum memory of 512MB, with 1GB suggested.

AI can do any job.

It can’t do every job, or else it would have happened. There’s a lot of limitations that prevent it from accomplishing even the most basic facets of some jobs.

Ironically, especially the manual labor jobs not already replaced by robots are some of the toughest nuts to crack with AI. For example, any task where a human starts to have a hard time just by putting on some gloves, we don’t have ways to feed the AI the usual set of inputs that the human is acting on that is needed to properly imitate the behavior.

On the front of dealing in fact based work, LLMs don’t really deal in ‘facts’ but facts are very highly correlated to the narrative, so it can frequently do a passable job, but it is factually incorrect a fair amount of time and if the facts actually matter, then it’s a tricky scenario. If you can hook it up to a feedback loop where mistakes are both harmless and fed back into the model for a retry, then you can significantly mitigate this drawback, but it’s not always feasible to do this.

Now perhaps it can ‘do’ every purely fictional output of text, audio, images, and video, which would be the topic of the post. The stuff we were kind of hoping to be freed up to do as fun and rewarding art. Instead, it’s the first thing that is ‘nailed’ by AI as far as corporates are concerned freeing up people to be those plumbers and HVAC technicians instead of doing art, and we get stuck with the most milquetoast slurry of slop to consume instead.

We had a lot of weird all at once.

Lots of death and isolation. Economy all screwed up with a flood of cash to try to keep things level. Trump’s sore loser behavior validated just the worst behavior (it is always there, but usually leaders moderates it when it rears its ugly head, Trump fanned the flames instead). Putin continued his aggression against Ukraine, except this time Ukraine held its ground.

Every big event didn’t require COVID, just a bunch of long burning issues bubbling up at once.

… I would like to subscribe to your newsletter.

Allow me to introduce the incels.

Hard to get that experience when no one wants to be around you at all, let alone naked.

For example, everyone knew the gallows were for Mike pence, but that doesn’t mean that Mike pence was doing the wrong thing… In that occasion…

Everyone thinks AI can do everyone else’s job except their own.

I personally don’t think so, and will not disrespect people by using it. It’s just amazing how the fad at work went with everyone posting how they were going to automate away every other job responsibility but theirs. The crazy part is how eager they shared their concepts with the very people they were basically saying were obsolete.

“resounding success” but given up as a bad idea.

The fact that it could possibly “work” was not really in question, it was whether it would be worth the costs and drawbacks. It was not.

But either way you are right, space even worse than water with no benefit compared to ocean.

Question is to cover rural communities as-is, do you and up practically replacing every car on the road with a bus? Bus is generally only better if you have significantly fewer of them to scale.

If you didn’t pay for roads, but you wanted public transit, I don’t see how that works.

There’s plenty of room to improve cities without having to bemoan that rural people are living wrong. About the only place I could see is making rural people switch over to intra-city transit instead of trying to have the cars wiork to get within a block of any given destination. That’s probably about as far I think to inconvenience rural people to meaningfully improve things.

Mass transit is only efficient if you have a density of people going the same way at the same time. A bus is only better when you have significant occupancy, and that’s not going to happen in very low density living.

You have to remember that the current investment level assumes that somehow everyone is going to use the AI stuff that much. It won’t do if the only heavy usage is code generation.

My experience is that prepping a code review with LLM is a safer bet than having a person review LLM code. None of this should replace static analysis or human code review, but it can provide a rapid feedback about something that looks off either to the original developer or helping a reviewer catch a detail that might have escaped their attention span. LLM missteps don’t matter so much, since the audience is a human (an exceedingly high false positive rate is annoying, but not a disastor). Also why it makes for a handy facility for security, mistakes don’t matter, missing something is no worse than before, false positives can be ignored.

If your scenario is well served by TDD with absoutely comprehensive test cases that can grade with absolute perfect coverage of all the scenarios, then code gen can thrash on that and make progress, though it’s a bit token heavy because there’s a fair amount of guesses and reguessing. But if you AI gen the tests and the code, then you might get into weird states. Even if you didn’t mean to have the AI mess with the tests things can go south. A couple weeks back I tried to prompt up a change that I thought should have been reasonably within the capabilities of the model, along the way it failed a long standing test and fixed it by changing the test, making the determination that the test case needed correcting to let it pass, instead of trying to pass the test case. The test case was, in fact, correct, it screwed up.

Part of the problem is ‘writing code’ is about as diverse as ‘writing text’, and some folks are working in territory that the LLMs can chew through competently and others are places where the LLMs struggle, and then there’s the person lulled into a sense of security by being in the former category until one day suddenly they are in the latter unexpectedly for no apparent reason.

often instead losing their livelihoods.

Even more severe, the owner class got the government to classify property damage as punishable by death. They were actively having the government kill Luddites who had not (at the time) even so much as injured a person.

“Ai” has mostly wasted my time.

Particularly other people’s use of AI has wasted my time…

Out of the few that said AI made their productivity better, most of them couldn’t code.

Note this point makes business leaders so excited. They hear “cheap labor can be productive”. Missing the point that the output is supremely dubious. They are “productive” because they don’t even know if the result is good or not, and so they can accept what is handed to them.

Currently, have a feature I thought we should have done, but management said one of the sales reps vibed up the answer, so we don’t need to do it. I used it and pointed out that the result was actually incorrect, but management thinks I’m just some luddite because they can’t understand the nature of the problems either.

One thing worse than AI PRs is AI bug reports.

One is a user that experiences a symptom that could be described in a quick copy/paste with maybe a sentence. Instead they have “AI” dig in and generate a dissertation on what it thinks is the issue and the original issue is lost in the AI diagnosis. Since the user doesn’t even have it wired up to actually test the hypothosis, it is just the AI first guess, which is frequently wrong. Wasting my time and failing to provide data.

The other is someone letting AI run over code and report “bugs” as in “it disagreed on some detail in the code, even though no person has actually experienced misbehavior, nor is the particular issue a security issue, but it might be an issue that some person one day eventually might hit”.

But I totally feel the coworker that had questionable code before that is now just utterly hopeless thanks to AI. I have nothing to do with his project but occasionally got called in to figure out and explain why his code would flip out. It hasn’t passed test in over a month as it’s now eternally buggy. Any attempt he makes to fix bugs inevitably lands with more bugs. But it’s too far now, he can’t turn back (CodeGen rewrote his codebase in a language he can’t follow, so he can’t hand edit, and he’s too afraid to roll back to when he almost understood his code). Hilarious one time within a week where he reported he could work on his project no more because he had exhausted his tokens, and without the tokens he can’t do anything anymore. Framed this as a failure of management to fund his needs adequately.

I have been shocked at just how many people, people long in the business seem to actively hate writing code. How much additional work they are willing to go through to claim victory in development through prose. Spending hours trying to get CodeGen to fix a trivial CSS issue in a web page that would have taken less than a minute if they just hit f12 and looked at the offending element. I can’t fathom how they could stand their careers for sometimes two decades before CodeGen came available if they hated it just so much.

This likely refers to the famous fictitious quote from Heny Ford that wanna-be self-important “innovators” like to trot out to show how they are smarter than the customer when they put out some unasked for bullshit.

"If I had asked people what they wanted, they would have said faster horses.”

People like to imagine the car as some miraculous thing no one but a genius could have imagined wanting and the genius descending from on high and delivering. Truth was people could easily imagine and want it, the genius was in knowing how to make what people knew they wanted.

Instead, Henry Ford said: "If there is any one secret of success, it lies in the ability to get the other person’s point of view and see things from that person’s angle as well as from your own”

So essentially the exact opposite of “screw the customers, we know better!”

It sounds like how students think professional sounds.

Well, also some executives that I can’t stand that, if I didn’t know better, would swear were ChatGPT for decades…