Regarding the frontier models, there’s a idea called Karpathy’s Funnel which notes the massive difference in experience with AI models. It funnels the population of the world like this: (borrowing a lot from a Karpathy tweet with some paraphrasing)
6B people (~75% of the population) have barely come in contact with LLMs.
1-2B (~20%) are casual and infrequent users of free-tier ChatGPT-like products. This group sees derpy chatbots and AI slop, to this part of the world the whole thing feels crazy still.
20M people (0.2%) see frontier-grade LLMs, complex projects that used to take them weeks/months can now be completed by agents with a prompt. You typically need a budget to get this experience. There are multiple levels in this tier as well from those that use ‘open-weight’ models to expensive monsters like Claude Fable. Models are Powerful tools at this level.
5k People (~0.00006%) have access to frontier-grade systems internally inside AI labs, they have the budget to put the frontier on automatic, spin up hundreds of agents, and burn their huge salary’s worth in tokens.
At these lower tiers of the funnel, they believe in the future the AI companies promise because they’ve literally bought in enough to actually get the experience.
This definitely lines up with my experience. I’m in the 0.2%, where work pays for us to have access to the latest frontier models via a CLI that lets the model essentially use our computer as if it were us. I’ve seen several colleagues transition from the 20% to the 0.2%, and it’s always amusing seeing their excitement as they realise what’s possible.
I remember that moment for myself too, and with the excitement came frustration. Until then I was convinced AI was always going to be slop, Will Smith Eating Spaghetti, the next corporate metaverse push to try and sell something people didn’t need. It was very easy to hate, along with the environmental, social, cultural and financial problems it was causing. But suddenly my world view was in conflict. This technology got good. Really good. Like being given superpowers 25 years into my career. Suddenly I was a hypocrite, and it took me a while to get used to the idea that these problems weren’t going away, and we’d need to solve them.
Regarding the frontier models, there’s a idea called Karpathy’s Funnel which notes the massive difference in experience with AI models. It funnels the population of the world like this: (borrowing a lot from a Karpathy tweet with some paraphrasing)
At these lower tiers of the funnel, they believe in the future the AI companies promise because they’ve literally bought in enough to actually get the experience.
This definitely lines up with my experience. I’m in the 0.2%, where work pays for us to have access to the latest frontier models via a CLI that lets the model essentially use our computer as if it were us. I’ve seen several colleagues transition from the 20% to the 0.2%, and it’s always amusing seeing their excitement as they realise what’s possible.
I remember that moment for myself too, and with the excitement came frustration. Until then I was convinced AI was always going to be slop, Will Smith Eating Spaghetti, the next corporate metaverse push to try and sell something people didn’t need. It was very easy to hate, along with the environmental, social, cultural and financial problems it was causing. But suddenly my world view was in conflict. This technology got good. Really good. Like being given superpowers 25 years into my career. Suddenly I was a hypocrite, and it took me a while to get used to the idea that these problems weren’t going away, and we’d need to solve them.