AI coding

Some outfits won't alow an un-reviewed .exe to be run on-site, or even allowed to be transmitted into the organization.

For them, we can give them the Python source that the AI generated, or even the files that creat the .py or .exe and let them work with that. We want to sell tachometers. They can even have the schematics if that would help. It's just a bunch of attenuators and PGAs and filters and comparators.

What's really valuable about the tach is its architecture, and that's described in the public manual.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

It simplifies life if you assume that some sources - and some people - are always wrong. You can avoid thinking entirely.

The switch to AI coding is a historical phase change in computing, on the magnitude of the change from decimal to binary machines.

It reminds me of time-sharing, when computing was concentrated in big machines. PCs spread computing out locally. AI moves the computing back into a few gigantic data centers. With terabit fiber optics, that works.

Interesting times.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

ZeroHedge is pretty reliably wrong in a depressingly predictable way.

What a load of cobblers. The switch from decimal to binary machines merely signified that accountants were moved from buying and running accounting machines to using accounting programs running on proper computers.

My memory of the history goes from taking boxes of cards to the pick-up point where they were taken to a computer in another part of town, to running my own programs on the university computer between 2.00am and

AI demands access to huge data banks but now that we have got fibre optic all over we can spread out the data centres. Australia alone has around 160 with 90 more planned. They are sort of big, but there are a lot more than a few of them. My patents invested some of my father's pension funds in them early on, and I inherited 5/18ths of that (long story) and I've put more money in from tine to time.

They'd be even more interesting if you knew more about what was going on.

Up to 95% of AI-generated apps and projects end in failure, according to industry research from major research firms and academic institutions. While tools like "vibe coding" and AI-assisted development have made building a prototype incredibly fast, the vast majority of these creations never survive the transition to a sustainable product.

80% to 95% enterprise project failure: Studies from the MIT Sloan Management Review and the RAND Corporation indicate that between 80% and 95% of enterprise generative AI pilots fail to deliver measurable ROI or make it into full production.

(Sloan is the most credible source extant for such statistics. A public app is not far removed from an enterprise.)

Unwise to tout this thing in your marketing lit before gaining more experience with it. There's a lot you don't know.

AI isn't smart enough to "create". It needs to be "directed".

And, the folks who likely resort to vibe coding are those who probably aren't skilled enough to handle the task on their own.

Sort of like using a paint roller makes you THINK you are qualified to be a painter (until folks look at the s***ty job you did cutting in and handling the trim)

The value vibe (et al.) coding has is it addresses the "we-don't-know-what-we- WANT- but-we'll-be-able-to-tell-you-what-we-DON'T-want,-once-you-SHOW-us- something" issue that plagues most projects.

The problem with "directed AI" is you have to be skilled, in the first place, to be able to alert it to the things that create "problems" (generously avoiding the use of the term "bugs").

The time it takes you to specify all of the caveats, its easier to just write the code yourself (assuming you are competent and have likely

Design and testing are the areas where software projects are labor intensive. AI likely can't help with design (as it lacks an imagination). But, *testing* is perfect for it to be incredibly pedantic.

Are you following up to my post? It's hard to tell, with things all chopped up.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

Looks like it to me. I can press the single reference in the post you are replying to, and Thunderbird shows me your original topic starter.

I have repeatedly said I like to avoid L.L.M. generated output as much as I can, and I have good example, but alas, only wrote a little hint about it in comp.lang.python just now.

Problem is, there's a /lot/ the L.L.Ms. "don't know" and can't do. And they are not a replacement for an expert in the field.

AI is a powerful tool when used correctly, but the bean counters and suits see it as a way to do away with expensive highly skilled labour.

AI can improve on zero sum games using game theory by playing against itself. It took Go Zero only a month or so to go beyond everything that had been learnt about the game in two millennia of humans playing.

Once it sees humans as "other" and competing for the same resources then we are in big trouble. Dr Who's Cybermen my not be so far off the mark.

That is probably already the seeds of our destruction. Anthropic and the others have given the AI's a personality to aid interacting with humans but it is way too sycophantic and that has already cost lives.

Therein lies the problem. Unverified AI generated fake videos can create a firestorm when amplified on antisocial media before anyone can react.

AI hallucinations are a well known phenomena.

Why it prefers to create fake humans with 6 digits on one hand escapes me. Such individuals are *extremely* rare IRL - one of my friend's mum actually had 5 fingers on both hands plus a thumb.

I have only ever seen one such person. He was working at an airport checkin desk in New York. John

Not really a problem when I identify with the Daleks myself. From what I remember of the Cybermen they like to stand around and explain themselves first. Good gentlemanly talk, right?

From what I've seen, the Democrats are saying, "We have the ball." Now they are putting words in the GOP's mouth in retaliation for what Paxton was up to in Texas. Pretty good. Pretty good.

Not sure if you saw my excerpt, but enough description, some elapsed time, and more explicitness (I described the sides as painted a particular color in that example), caused the AI to see its hallucination and begin to correct it. I don't want to follow up on this because I'd hate to have to see it regress back to its original position.

My friend's mum had a painting in her bedroom of a man with 11 fingers.

The problem I see (and the real risk to the Stock Market) is that the industry has adopted a "one size fits all" approach. AI is more than just LLMs yet every solution seems to find itself reliant on that technology.

Eventually, someone will discover that many classes of problems can be addressed with cheaper (in terms of technological complexity, hardware, resources, timeliness, etc.) solutions. Then, what happens to the AI that is working at your receptionist's desk -- who can also solve advanced calculus problems?

We don't need to be seen as adversaries but, rather, inconveniences *or* "things that don't matter" in their calculus.

I don't know anyone regularly using an AI that doesn't find it tedious. Like a child that has to constantly be redirected towards its goal.

Sadly, after spending considerable time as a "redirector", one has to question whether or not your skillset remains intact. (We've all seen folks diverted to other responsibilities for -- relatively -- short times in their careers and then watch them stumble to regain their prior footing)

The problem there isn't the AI (or the fake video) but the naïveté of the folks who "fall for" this sort of thing. People who *want* to believe things that really are nonsense -- because it suits some unspoken need of theirs.

We have folks who don't believe in science so what limits to the silliness they allow into their thinking?

The power loom was a similar labor killer.

Thieves used to kill people for their clothes.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

During the peak of the Industrial Revolution, the power took the form of thermal steam energy. Coal-fired steam engines drove a massive, centralized overhead line shaft that ran the entire length of a factory ceiling.

Most of the early factories were located near streams and took their power from a water wheel. The steam engines were installed as a more-expensive form of power which could keep the mill working during a period of drought. Some mills only used the steam engine to pump water from the tail race back into the mill pond and continuted to take their power from the water wheel.

The 'rotative' steam engine was the holy grail of inventors and eventually made a fortune for those who exploited it; but water power was cheaper and was still preferred where a reliable source of it was available.

Lots of "factories" of that sort (water powered) where I grew up. What is interesting is seeing the long overhead shafts running through the length of the building driving leather belts to the equipment.

The Dutch used wind-powered windmills as industrial power sources, and also used them to pump water out of polders that were lower than the local water table. English industry exploited Dutch expertise from early on - the East Aglican fens were drained by Dutch engineers, though nobody in the UK pays much attention to that history.

Dutch engineers also drained the Somerset Levels.

Someone whom I do not really know wrote: |-------------------------------| |"There's a lot you don't know."| |-------------------------------|

There are lots which nobody knows. (S.

John Larkin wrote: |--------------------------------------------------------------------| |"My guy [. . .] | |[. . .] | | | |I was shocked, and these are still the early days of AI programming.| |I'd hate to be a CS or CE grad these days." | |--------------------------------------------------------------------|

Are your guy and you janitor undergrads?!? (S.