I listened to an interview with Grady Booch. If you've never heard of him, he has been a software engineer for a long time, developed UML, and worked at IBM and a number of other places. He has devoted his life to improving software engineering. He even told Bill Gates he didn't want to be the Chief Software Architect at Microsoft.

At one point he had a less than flattering description of AI LLMs. Politely, they are unreliable narrators. Less politely, he feels that they allow us to build at a global scale, unreliable BS generators. This is because the LLMs that are stochastic parrots, which can produce some coherent results. Primarily they allow us to navigate a very large lake in space, i.e. the Internet. I think that's true. I have found AIs to be pretty good search engines. Not perfect, but good.

Mr. Booch notes that LLMs, while interesting, are a shadow or a whisper of what humans can do. He's critical of those who think AI is going to compete with humans. Here are limitations on what the tech can do, and Mr. Booch thinks that the approaches people are taking, architecturally, are wrong. As part of his work, he studies more of how humans work and think as a way of trying to build better software architecture.

Caveat, he does think we can build more intelligent systems, but it's not with Gen AI/LLM architectures.

I tend to agree with him and do think that the LLMs are unreliable. They appear to be intelligent, but they are more predictive engines in many ways. They can be very helpful in many ways, but they aren't necessarily replacing smart humans. They might help smart humans replace some other humans, but they are likely to be better assistants than replacements.

It's a great interview and worth listening to. There's a walk-through where Mr. Booch touched on computing as well as a very positive outlook on the future of his career and the things that he is working on. It's long, (1.5 hours), but a fun listen. Give it a go and let me know what you think.