That myth has absolutely caused harm. That you can't see that is a personal failing. Try thinking before blindly reacting.

It's a silly toy, not a tool. If you have any sense, you'll realize that in time.

You're certainly producing more code, but that doesn't mean you're more productive, only that you're producing more code. What a lot of people are discovering is that they're getting a lot of 'dead' code and redundant code, particularly when using agents. Here is one guy's experience, if you're curious.

The writing is on the wall. The promised improvements are never going to materialize, as I've been saying for years, due to fundamental limitations of LLMs. Using AI might make you feel more productive, but all you're really doing is making your code base larger, more fragile, and less secure. It will cost you more in the long-run than you think you're saving in the near-term.

In 10 years time, you'll deny you ever believed such a ridiculous thing.

Well, it used to be anyway...

We covered more CS content in my intro course than kids get today in 4 years. The average CS undergrad program is little more than a glorified programming bootcamp these days.

Python is nothing like BASIC. The comparison is both misleading and, to beginners, actively harmful.

As you've probably already figured out, Python is a terrible language for beginners.

Python obscures many basic concepts. For example, I've also found that the whitespace rules obscure blocks, making it very difficult for beginners to understand control flow. Even simple for loops are needlessly burdened. BASIC makes these things so simple and obvious that they almost don't require any explanation.

No question. Dijkstra was wrong about BASIC.

If you haven't already, Papert's Mindstorms is worth a read.

It's not like we haven't seen that countless times over the years. It always ends the same way. It turns out that specifying the kind of program you want in sufficient detail for a computer to generate it is just programming.

Sure ... except that the bots are both expensive and can't actually do the job.

I've never once regretted ignoring the latest programming fad. The time, effort, and money saved over 40+ years is incalculable.

The difference between "AI coding" and the other tools you mention is that those are accurate, consistent, reliable, and inexpensive. LLMs are not.

That said, AI coding agents do provide a competitive advantage ... to the shops that have figured this out already and aren't using them.

You're the only one talking about Searle, dipshit.

Prove it.

We don't know how brains work, you drooling moron.

You're such a fucking joke!

If you believe otherwise, you've got to overturn 100 years of computer science.

I'm sure that you, someone who randomly brings up and still fails to understand Searle's argument is surely capable of such a feat!

What a fucking joke!

You can stick your head in the sand and ignore reality if you want, but you're only hurting yourself and any unfortunate person who takes your advice.

Takes some time to actually learn about what LLMs do and how they work. It will very quickly dispel whatever ridiculous science fiction nonsense you inexplicably believe is science fact.

Again, these things can't actually summarize text. The "sources" are just the pages that had some of their content included in the context when that part of the response was generated.

That's a nonsense solution. You can ask for "well sourced material", but odds are against you actually getting that. The model has no concept of "well source material". You can ask to "include the relevant snippets form the sources", but odds are against whatever it produces actually existing in the source documents. (Ask it to quote relevant snippets from something that you know doesn't exist and it will still produce quotes.) You can ask it to check the sources support the conclusion, but you'd be better off flipping a coin. That'll get you the right answer more often.

The problem here is that LLMs objectively can not do those things. Remember that they don't operate on facts and concepts, nor can they actually follow instructions (this is provable). All they do is next-token prediction. It's only through careful training and framing that we're able to create the illusion that they can do those things.

These are all perfectly reasonable use cases. After reading the insanity above, this seemed praiseworthy.