Yeah, less likely to be in the training set. I wonder if there's some kind of keyword you need to make it work. Or maybe more sort of chain of thought and planning?

Here's the thing. AI coding, well it works well enough to be massively more productive. Even if sometimes it fails, its still massively more productive. Businesses care about that more than anything. So it WILL be adopted broadly and there's nothing anyone can do to stop it. Money trumps all.

The thing is, its best when you're at the edge of your knowledge. Like when I'm doing web app stuff, I know roughly how these libraries work, but I don't care to learn the syntax for the 10,000th framework/library that's popular for this month. Just get AI to do that stuff.

There's an art installation light show in Lincoln over the weekend. Then another one weirdly just under a week later. You'd think they'd coordinate, but artists huh? :)

I've got a frightening number of interesting books to read over summer. Mostly art or AI related (often about how one impacts the other etc). I love books, but damn they remind you that you're time is way too limited and filled with useless admin tasks instead of fun and enlightening learning.

Trying to get academics to take AI coding seriously and they're just so reluctant to acknowledge how it will change things. I know the hype has been bad, but why is someone who is supposedly educated so against at least evaluating it for themselves? It will require massive changes.

And I'm fully convinced that the "programming language in the middle" paradigm will have to go. AI model talking to AI model through some well structured interfaces will take over eventually. Making an LLM learn to code is impressive but ultimately makes the programming language obsolete.

Post by Tim Sweeney over on X has me a bit frustrated. Talking about UE6 and all the changes to property systems and using new C++ etc. I can't express how clear it is that AI assisted productivity is where the gains will be. Structure your engine to support that and your customers will thank you.

Now that the latest copilot release has convinced me that AI-Assist coding will increase productivity by quite a lot. I'm really raring to get going on the backend code for my data capture. Already making good progress on the video annotation/transcription process. RL capture next.

Got to do my peer review of papers today. Then get on with some API examples for Thursday. Weekend is doing offer holder day activities. It's going to be a busy week.

I'm also thinking of writing a book about practical Reinforcement/Imitation/Curriculum/MARL learning for games specifically.

So it appears that the new github copilot release (using LLM's from the big providers) now does pretty well at basic Unreal Engine C++ classes. Haven't tried anything advanced yet syntactically, but hell, even that is useful productivity wise.

Thanks Mark, followed :) and to be fair, this was most of my list of follows at the other place too :)

Trying to figure out a good strategy to find interesting people on here. Because frankly I've lost the will to live on the other place :)

Anyone got anything interesting for me? AI stuff is good. Hyped AI stuff not so much. Feed me!

Michael Black's research students especially welcome :)