Why can’t AIs generate for the “existing, mature ones?” Like the other commenter said, I’m not sure I get the “this is totally for AI” marketing. Why can’t AI use the existing ones and why can’t humans use this?
I went through a similar phase decades ago with Common Lisp. It takes a week or two. Now, it’s quite a natural syntax and I see the parens as a huge benefit. I like Clojure syntax even more than CL and Scheme because of the map and vector literals.
To be sure, social media is a problem for our children. No doubt about it. But the solution is to take screens from the children, not set up the foundation of a police state.
Okay, but that still has a limit, right? Do training costs have a limit? Everyone is in the frothy stage of this technology wave and they continue to buy more, but training the next model requires exponential increases in model sizes to get the same sorts of model performance increases, which suggests exponential cost increases, too (even ignoring temporary cost factors such as RAM price increases). You say your company will double or triple what they are paying today; how far are they willing to go? At some point they are going to have to cut developers to fund it (e.g., cut half the developers and give the survivors each an AI assistant with $180k in token budget, captured from the salary savings), but that also presupposes the productivity gains are there to support it.
I believe that OpenAI is in trouble. Too many high level defectors. Altman is clearly in hype mode. Is Amodei? Yes, I’m pretty sure he is at some level. He might also believe it partially (it’s not totally binary).
It’s all about timing. This is tech bubble 2.0, Dotcom Boogaloo. If you’re able to flip it quickly, you’ll have generational wealth. If not, you could be holding a lot of worthless paper.
Perhaps, but as the AI analysis becomes part of the release process (or even the CI process as prices fall), you’d expect those new issues to be caught before release and fixed. We’re seeing them caught post-release for now because the code is older than the AIs, so we’re catching up.
There is going to be a flurry of this sort of stuff as the AIs get smart enough to find them. It will naturally die down as the legitimate ones are fixed. Yes, there will always be some level of this, but I’d expect it to be low and the exploits found to be increasingly complex. This is a time of transition.
So sad. I first met Om in 2001 or so. I pissed him off because I wouldn’t meet with him to do an interview for our startup. He always loved getting the early scoop and we weren’t ready for any publicity. In later years, we would laugh about it and I gave him the early scoop on the next one. During those years he became a friend and we would sometimes grab lunch and chat about all manner of things, from tech to family. I dished on some of what I watched go down in the dot-com bubble for his Broadbandits book. Later, I would go on to write contributed articles for GigaOm. Goodbye, buddy. 60 is too young. You were one of the best. Maybe you’re getting the early scoop in a different way.