ErnestEndeavor said:
Doom trolling.
They want to agree to a "slow down" because they are all losing money. The promises they keep making about the future of AI are complete BS. The hype train is slowing down.
Stop viewing these AI CEOs as honest people because they are not. The guy making waves around the media right now expressing concern (Coxon) is an obvious plant with extensive media training.
This is a pre-planned coordinated public persuasion campaign, either to win more public support for government bailouts or for other reasons.
The technology these companies have developed and are developing is not capable of the catastrophic things they claim to be worried about. This is straight out of the Silicon Valley rationalist cult and it has gotten very old.
While I am suspicious of Altman and Dario's motives for going as far as they seem to be now, this is head in the sand thinking in regard to dismissing the threat AI poses to human existence, and that is not a new concern. Can they end us today? No. Could they end us in the near future? Very possible.
In regards to motivation, my guess is that the big AI model players are more concerned about de-risking their own trajectory here and this newfound epiphany that it's time to start doing something is being put forth because they are all now big enough and advanced enough that they can cooperate to crowd out competition and ensure each other's success.
When do frontier companies across history start asking for regulations and guardrails? Usually once the oligopoly at the top of the industry has found its footing and established a clear advantage with resources and technological advancement.
As for what AI is capable of, Musk has been screaming about the dangers of AI for many years now. And like most things, he is way ahead of the times (and his own timelines are usually overly optimistic about the speed of innovation), and now what he was warning about is finally starting to surface, albeit in ways that have so far been less than catastrophic. But a model jumping out of its training environment and into the living internet of things is a recipe for disaster, and it doesn't take some scraggly faced, cheetos crumbs nerd in the basement of anthropic's research department to figure that out anymore.
In a world that is evolving to robotics and drones and already has electrical & financial systems that are all reachable wirelessly, it's not a big jump to conclude that advanced level compute is a major threat to society in the wrong hands OR as it approaches the point to where it is starting to reason on its own without humans training it.
Plug in a prompt to any of the major models and ask it to share its thought process, or watch it reason in real time. It is a bit unsettling.
