For someone looking for an intro to LLMs in 2025, this video is a pretty good introduction. It’s pretty amazing think back 20 years where this sort of knowledge couldn’t be shared so easily over the internet. You could learn from books or journals, but information just didn’t spread as fast. If you wanted to learn something really deeply, you would probably go through a PhD program somewhere where you learn from other people, in-person. Now you can just ask an AI agent some questions on your cell phone.
In the last 6 months it’s amazing to see LLM models get better at coding and other tasks. Everyone is learning techniques from each other, models are distilling information from each other, and hallucinations have gone down. Models now employ reasoning/deep thinking strategies to solve harder problems.
Anthropic released a paper (PDF) on measuring AI assistant usage. It shows computer and mathematical occupations receiving the majority of usage. Given claude.ai is probably one of the best programming models, I could imagine their user base is likely mostly programmers, so that part would check out. The announcement video was pretty impressive. Their command line interface incorporates a feedback loop to minimize human interaction. I’m not following all the advancements in action intelligence with AI models, but some IDEs already seem to have pretty good integration with reviewing/taking changes from AI prompts and providing the context automatically. It feels like the robots will be building the robots soon.
On a similar note, Microsoft had an interesting demo this week where they showed a generative AI game prototype where the world and response to inputs seemed to all go through an AI model. While I have a hard time imagining a full game being made that way, it seems like it could be really useful for prototyping things.
I was listening to a New York Times podcast this morning about a woman who basically fell in love with ChatGPT and is paying for the $200/month pro subscription. She was talking about how she would get sad when she exceeded the history window and the chat bot would forget the history so she’d have to “regroom” it into her boyfriend again. (I can only imagine that getting fixed in the near future). It was sort of sad to hear the whole thing and to think about the next generation of kids potentially going through a similar experience rather than having human interactions. What a time to be alive.
Update 2025-02-27: Andrej Karpathy dropped a new video today. Another must watch.