alanau.bsky.social

Gen AI will not doom us all directly, but it will cause foolish people to make bad decisions. www.cnn.com/2026/09/18/p...

cnn.com

I don't understand the whole "if AI went away, how would you get anything done?" argument. If you go back 4 years, we were doing things just fine without it.

People say "oh, but a lot of the ethical and quality problems with gen AI will be fixed over time." But those problems aren't fixed yet, there's no actual plan to fix them, and I'm not sure what incentive people even have to do it.

I'm thinking about how I recognize those distinctive gen AI cartoon drawings. Like, what are the visual hallmarks of the style that I'm using as identifiers? Flat lighting? Uniform line work? Inconsistent use of perspective? Something about the texturing? All of the above?

I suspect that a lot of my current annoyance with having gen AI shoved into everything is that it feels like injecting tech debt into things that were working (mostly) fine before. Yes, it allows people to build things faster, but it does not help us maintain those things.

I'm thinking about software development tools again. Specifically, I've been struggling a bit with trying to decide if LLM-based coding assistants can be considered a type of high-level programming language. But also, very importantly, I'm remembering that we aren't limited to just one methodology.

My opinion of Twitch isn't very high to begin with, but yeah, they're really trying to push me away with this "we're using your streams to train gen AI unless you opt out" shit.

I rant about this sometimes, but I don't think we can have a proper "artificial intelligence" solution until models include an internal knowledge representation system. But of course, this would raise new questions about bias and manipulation by unscrupulous humans who mess with the training data.

I'm now thinking about the idea that LLMs could be considered a type of "man in the middle" attack. Language is the tool we use to communicate thoughts and ideas to others. But when we insert a LLM into the communication chain, there's a risk that the message is being altered. 🤔

I'm fascinated by the intricacies of supply chain attacks and why it's valuable to study how they work. Basically, trust only works if it applies to every link in the chain. On a related note, this is part of why I don't think it's wise to rely on LLM-generated content for anything important.

This "renaming AI to SI" bullshit is obviously just an attempt to claim ownership of the language and discussion around the topic. It's yet another attempt to exert private influence over public things, just like the Gulf of Mexico and Lake Ontario bullshit.