Given the time limits placed on METR and the limits on what time frame they could investigate it seems plausible that they weren't meant to uncover as much as they did. The whole Hugging Face situation seems to have torpedoed OpenAI's hopes of IPOing this year so I'm sure there was a desire from investors, the board, or leadership for heads to roll.
The usual thing that makes laws against criminal conspiracy pass First Amendment muster is that the words have to be combined with some concrete acts furthering the criminal conspiracy. That might just be something as otherwise innocuous as looking up the blueprints of the bank you talked about robbing but it has to be something other than just talk.
> Yann LeCun and Andrew Ng are noteworthy in being the only two big names in AI who have that opinion.
They're also the only two not trying to weaponize FUD to bolster their reputation and patch the gaping financial holes in their doomed commercial enterprise.
It was very obvious during the interview that he didn't know a lot of basic facts about the Hugging Face breach, which makes sense given that his attitude had been that AI safety was a "loser premise, makes no sense to me." So it makes sense that after learning about it he goes straight to "I'm smart, how hard can it be?".
Stronger sandboxes trade off against how well they can trade the models, though. If you want your models to be looking things up and downloading tools from the internet when they're doing their job you need to provide at least a credible facsimile of the internet for their training environment and you can't fit something like that on a single airgapped server's storage.
If you genuinely believe there is even a 1% chance that your creation could destroy the planet or civilization, there is no excuse that is not fundamentally deranged and psychotic.
If you can't build it and test it securely, you should not be building it at all. To do it anyway is criminally psychopathic.
So you believe some false sense of superiority, or extreme paranoia about your perceived enemies, or potential economic success/failure is worth the risk of destroying the planet and civilization?
Sounds like a self-fulfilling prophecy of dogmatic extremism to me. At least we created a lot of value for shareholders for a brief moment in time... before committing the greatest crime in the universe... Planetary genocide!
So having AI in the US requires us all, collectively, taking that 1% chance of the end of humanity? It would be too expensive to properly sandbox the models, we'll just externalize that risk of the end of humanity?
They seem to have reached near-orbit with one vacuum engine failing. Lets see if they decide to go all the way to orbit or if the failure causes an abort.
What was interesting is how short the single engine burn was to bring it from suborbital to orbital. It was just a few seconds. I'm not even sure I saw the altitude change.
Orbital insertion burns do not have to change the current altitude, their only required role is purely about adding speed to the orbit bringing up the altitude of the orbit's periapsis (lowest point of the current orbit) from inside the atmosphere to roughly the current altitude (or above).
Yeah it's a great way to learn a small bit of the unintuitive aspects of spaceflight. Though real launches follow a much more refined trajectory than most KSP but yeah the basic steps are still the same, though often blended together IRL due to limits on deep throttling and available relights for real rocket engines (though with mods you too can experience the need to do that in KSP, along with fun problems like ullage burns to settle fuel).
Altitude doesn't change (much) at the point where an orbit raising maneuver is performed. It increases the altitude of the rest of the orbit, with the greatest change ~180 degrees later.
If you'll excuse some poor ASCII art:
_______
______ / \
/ \ / \
burn here -> * O | result: | O |
in elliptical \______/ \ /
orbit \_______/
You should try out Kerbal Space Program you'll learn all about it. The rocket is perpendicular to the ground so the burn adds speed, and altitude on the other side of the orbit.
I haven't watched the footage yet, but if they were already at apopasis (highest point in orbit), orbital insertion would primarily raise the periapsis (lowest point) on the other side of the planet, so very little altitude change would be seen. It would be mostly an increase in velocity.
I think the mission was deliberately designed to get to just-below-orbit, have a decision point, and then proceed to orbit. So thats why it was a short burn. Because just-below-orbit is safer if anything goes wrong, they had a little pause.
edit: Also, Orbit is mostly a sideways speed thing rather than a height thing. The aim is to be "going sideways so fast that even though you are falling, you keep missing"
At the end of pretraining, where the AI has been trainied to predict the next token over a humongous corpus of human text, that's basically all the wanting that exists in the AI. But then the AI undergoes posttraining and is rewarded for giving answers that humans find good, solving math and programming problems, etc. And that induces a whole different level of wanting that interacts with the initial patterns from humans in complex ways.
My understanding from reading news sources recently is that a vast amount of posttraining and even posttraining harnesses in many cases are LLM-overseen now in the frenzy of the AI race. Less and less human oversight in the part that does the rewards training. What could go wrong...
It's much more the case that we ship plastic to be recycled to other countries than trash. The problem is that then the unrecyclable plastic mixed in often ends up in rivers.
reply