jprete

39 ms·

691.
by jprete 3y ago
OpenAI literally kicked off an AGI race, which is exactly the thing you do not want to do if you’re interested in AI safety.
692.
by jprete 3y ago
Strategy disagreements are absolutely central reasons to fire executives.
693.
by jprete 3y ago
I think the board is required to be a majority non-equity-holders precisely because an equity-holding board will not keep to their non-profit mission.
694.
by jprete 3y ago
The question has enormous implications for OpenAI because of the specifics of their nonprofit charter. If Altman left out facts to keep the board from deciding they were at the AGI phase of OpenAI, or even to prevent them from doing a fair
695.
by jprete 3y ago
If OpenAI’s governing board is part of the nonprofit, their obligations are to the goals of the nonprofit, and “thriving” is not fundamentally the goal.
696.
by jprete 3y ago
I think LLMs and RLHF are inherently unsuited to producing aligned, useful, and fact-oriented AI. Something else is needed.
697.
by jprete 3y ago
AI’s case isn’t at all the same. Politician bullshit is always directed towards a goal, and that goal is usually re-election or popularity or similar. AI bullshit exists to satisfy an RLHF process that prioritized convincing humans that the
698.
by jprete 3y ago
OT, but did the author somehow not notice their AI-generated cow-robot cartoon has five legs? And some kind of ear-poop…thing…coming out the rear end?
699.
by jprete 3y ago
There’s no way to separate them. Education is influencing someone towards things that we believe are true. If an educational AI believes that propaganda is true, then education and propaganda are the same thing.
700.
by jprete 3y ago
The usability article [1] from yesterday suggested a similar problem - design-oriented people want to put their graphical mark on their GUIs, and it comes at a significant usability cost. GUIs are most usable when they are consistent - ever
701.
by jprete 3y ago
A lot of methods of expressing software ideas are also very inefficient. The actual interesting part, the entropy, is very small. In the demo it’s literally two sliders controlling two CSS attributes which is not a lot of bits of entropy in
702.
by jprete 3y ago
I get the idea, but the dystopian part is the students being continually isolated in their educational time with their only contact being an impersonal computer program (or something that fakes being personal, but is not actually free to be
703.
by jprete 3y ago
AI is trained off the intellectual output of the people who did Job X, so it seems 100% fair to me.
704.
by jprete 3y ago
If that’s what senior engineers have to do, I’m horrified to contemplate what everyone else would have to do.
705.
by jprete 3y ago
“Doing next-token prediction” isn’t a contradiction of “understanding” any more than it would be for a SO/friend who can complete your sentences. But it’s useful to remember that autocompletion of sequences is at the bottom of LLMs, an
706.
by jprete 3y ago
By X-risk person I just mean someone who thinks AI does have near-existential risks, and those risks are significant enough to merit being a top consideration in making decisions about AI. I don’t personally base my stances on Yudkowsky’s i
707.
by jprete 3y ago
I find Zvi Mowshowitz to be the best source available ( https://thezvi.substack.com/ ). He’s in the X-risk camp (and so am I) but seems to have a clear view of some AI things being exciting, others being dangerous, and still
708.
by jprete 3y ago
I haven’t seen it but I can believe it - parallel programming of any variety is hard, and to be successful as a vendor of such a system would require uncommonly good API design to get the kind of traction that CUDA has gotten.
709.
by jprete 3y ago
I’m confused as to why you need to pay an AI service to normalize audio volume? Isn’t this something basic audio tools would do for free?
710.
by jprete 3y ago
“@rcastlebury's activity is private”
711.
by jprete 3y ago
It’s a set theory problem, calling it division by three in the paper is a bit of a joke.
712.
by jprete 3y ago
A commenter on the thread mentioned that Apple once tried to one-day-ship them a laptop via Uber Eats and the driver vanished with the laptop. Apps don’t magically make things better.
713.
by jprete 3y ago
The job board has better vetting, because only actual truckers even know it exists. In the OP, at least “Tristan” was actually a trucker with a truck.
714.
by jprete 3y ago
You’re assuming a level of system coordination that doesn’t exist. The storage facility isn’t going to do a damn thing with security cameras unless the police order it. The police aren’t going to get involved unless there’s a clear shot at
715.
by jprete 3y ago
If you are really claiming that these are the questions that people are asking when they say ChatGPT isn't useful, then...that is an unbelievably blatant straw man.
716.
by jprete 3y ago
> I've tested both cases: correcting it when it was really wrong, and correcting it confidently when it was actually right. Both times it agreed that it was wrong and regenerated the answer it gave me. This is the peril of using wha
717.
by jprete 3y ago
Asking questions that it's not designed for? Which would those be?
718.
by jprete 3y ago
Windows 11 more or less forces OneDrive onto the machine. It takes active work to get rid of it.
719.
by jprete 3y ago
For what it’s worth, much of the cyberpunk skeleton is in place - good and bad - but it’s not clearly visible from a thematic perspective, because human life isn’t as cheap as in genuine cyberpunk environments. We’re also, frankly, moving t
720.
by jprete 3y ago
Religions act like governments - their first inherent preference is stability. SV ideology is explicitly anti-stability, exemplified by the phrase “move fast and break things”.

More ›