Reading this article its just clear the author hasn't used the current generation for real work.

> LLMs can wing it for tasks that are related to natural language (e.g. writing social media posts, reports, articles, etc.) but when it comes to code, the same engine that struggles to count number of R’s in “Raspberry” or suggests a walk to the carwash, also exposes other logical fallacies

Weirdly none of those things matter when writing code and actually LLMs fail at social media posts and articles to anyone who has seen enough of it can clock it's AI straight away, yet everyone who's used these models properly has solved harder problems than walk to the carwash with them, neither of the problems he's claiming are code were proposed as code problems or tested as code problems.

A lot of what's said just comes across as wishful thinking and being out of touch with the level of output current models can do, and I mean hard problems too.

▲

rhdunn 4 hours ago | parent [-]

They matter when implementing the business logic of the application, when writing tests for a character counting function, etc.

	▲	whywhywhywhy 3 hours ago \| parent [-]
		They literally don’t because these “problems” don’t arise in that situation, and the bias in this article misses that. This is a writer hearing problems LLMs have and inventing an imagined reality where those cause coding issues not because they understand how LLMs manage to code but because they think these problems are part of what coding is. The actual reality is LLMs have no problem writing code for anything described here as shortcomings because writing code to do something is different enough from doing the thing. Proficiency in counting Rs and proficiency in writing code that counts Rs are different and you can fail one and excel at the other it turns out.