kvee

No, we should make it actually aligned and unbiased

Fun loss of control story to pair with this: https://x.com/joshua_clymer/status/1887905375082656117

https://archive.ph/NDzW4

On the other hand, it may be that "Alignment likely generalizes further than capabilities." - https://www.beren.io/2024-05-15-Alignment-Likely-Generalizes...

https://archive.ph/z5MMg

https://archive.ph/sOkH2

Future of Humanity Final Report from Anders Sandberg here:

https://static1.squarespace.com/static/660e95991cf0293c2463b...

and Google Doc version: https://docs.google.com/document/d/1jgl2KqtiJ6lLkpoZ1I_VeniP...

https://github.com/robertvacareanu/llm4regression

It does seem actually to be "feature working as intended" if you read the initial Gemini paper.

Some relevant concerning stuff pulled from it here: https://twitter.com/psychosort/status/1760849091171352956

https://twitter.com/psychosort/status/1761044625307963445

Paper itself here: https://storage.googleapis.com/deepmind-media/gemini/gemini_...

I think I can see how you interpret it that way.

I certainly didn't interpret as him disagreeing with his statement "I mean I have like lots of selfish reasons for doing this"

It's the "as you've said" part of "as you've said I get like all the power of running OpenAI" that would make me inclined to think what you wrote here.

But I do think there's a greater chance that he is saying that he does like the power.

There's also another quote either in this video or the other one I shared I think where he's asked why he's doing this, or what motivates him, or something like that, and he responds with something like "I'd be lying if I didn't say I really like the power"

Just went to the first video and got the following from 1:36, here's a link which starts at that point: https://youtu.be/3sWH2e5xpdo?si=bmum-8B02FLoVkWj&t=96

"I mean I have like lots of selfish reasons for doing this and as you've said I get like all the power of running OpenAI, but I can't think of anything more fulfilling to work on and I don't think it's particularly altruistic, it would be if I didn't already have a bunch of money, yeah, the money is gonna pile up faster than I can spend it"

Some other fascinating and relevant stuff in that video too.

Here are a couple I could find in notes I took while listening to podcasts, though there are more -

“I get like all of the power of running OpenAI” “I don’t think it’s particularly altruistic. Like it would be if I didn’t already have a bunch of money. The money is gonna pile up faster than I can spend it anyway.”

Those I think are either from https://www.youtube.com/watch?v=3sWH2e5xpdo or https://www.youtube.com/watch?v=_hpuPi7YZX8

He has said in podcasts he is motivated not by the money but by the power he has at OpenAI

You may be right here.

However, there seems to be a decent amount of evidence that Sam has done exactly what you're talking about.

He manipulated and was "not consistently candid" with the board, he got all the OpenAI employees to support him in his power struggles, he made them afraid to stand up to him (https://x.com/tobyordoxford/status/1727631406178672993?s=20), he exhibited delusions (though I guess they were correct) of grandeur with pg with a glint in his eye making clear to pg that he wanted to take over yc, he did little things like made it seem that he was cool with Eliezer Yudkowsky with a photo op but didn't really chat with him, etc.

Again, I am not sure this perspective is necessarily right (and I may be convinced just because he's such an effective psychopath).

In any case, I think this is a pretty good explanation of this perspective: https://x.com/erikphoel/status/1731703696197599537?s=20

In this case I think I just mean more effective at seeming good to others.

I think they both believe they are good and doing good.

People tend to be more suspicious of Mark Zuckerberg's motives than Sam Altman's.

Sam Altman himself even said he can't be trusted but that was ok because of the company structure and then, when he needed to, overpowered that structure he claimed was necessary: https://x.com/tobyordoxford/status/1727624526450581571?s=20

pg talks about how Sam Altman is the most powered person he's ever met. Seems we have a super powerful psychopath running perhaps the most important company in human history.

I do think he legitimately believes he's doing the right thing though all throughout, which maybe makes it more scary.

Sorta like how Mark Zuckerberg seemed to truly believe in Facebook's mission and wound up having all sorts of negative externalities for the world. Mark Zuckerberg just isn't quite as effective as Sam Altman, and it's easier to be suspicious of his motives.

Not to say that psychopaths are necessarily bad. Peter in Ender's Shadow turned out great!

But it does seem dangerous for 1 person to hold so much power over the future of humanity.

Sam Altman's reasoning for him having all the power, I think, is that “short timelines and slow takeoff is likely the safest quadrant of the short/long timelines and slow/fast takeoff matrix.”

If you believe that and believe that Sam Altman having complete control of OpenAI is the best way to accomplish that, everything seems fine.

I'd personally have preferred trying to optimize for long timelines and a slow takeoff too, which I think might have been doable if we'd devoted more resources to neglected approaches to AI alignment–like enhancing human capabilities with BCI and other stuff like that.

If of interest to anyone here, my company has built an open source software library for transcranial focused ultrasound stimulation: https://github.com/agencyenterprise/neurotechdevkit

Its goal is to accelerate the development of agency-increasing neurotechnology and lower the barrier of entry for any developers to be able to solve open problems in neurotech without having to have their own hardware or human subjects. It's starting with ultrasound, which we find quite promising, and we hope to expand in the future to other areas as well.

https://archive.is/3LaJF

I suspect treating screen use as "treats" may not be ideal.

I also have a 2 year old son and am very curious about this.

It's tough, because I do know whatever I do won't be ideal, and we won't know till later when we learn more about how our brains work what we really should be doing with regards to screen use by kids.

And, at the same time, things are accelerating fast, so there's a decent chance tech is totally different by the time our children are in high school, or even elementary school.

I think that in an ideal world, we'd all use technology all the time and it'd be a natural extension of ourselves letting us be fuller versions of ourselves and more present with each other. It's a shame we're not in that world yet, and I'm hopeful that as we get richer humanity starts designing tech to be more agency increasing with this target in mind rather than losing agency to shorter term incentives.

With our 2 year old, I think the highest impact thing we do is try to strict about being present with him and not on our phones. I suspect this might be much more important than whatever rules are established for a child's own use of technology.

Reminds me of the pretty compelling poem "Children Learn What They Live": https://www.freepoemsonline.net/poems-htm/children1.htm

When NFTs were hot, we made one that was intentionally tasteless thinking that it could raise awareness for people to donate to effective charities (pretending it was NFTs for needy children, but actually the proceeds from NFTs would go to the Against Malaria Foundation), but never could really get the whole thing right so decided not to launch it.

If anyone has an idea to actually execute that original idea right and do some good net net, here's the website: https://nftsforstarvingchildren.com/

I think the best thing is to do more research to understand more rather than rejecting the scientific method entirely. It has and will continue to get better and better over time, with occasional regressions that may seem meaningful in the short term but are tiny blips when you zoom out. The history of science is one false theory being replaced by another that’s (usually) marginally closer to the truth.

Also recommend open sourcing the code so everyone can see that API keys aren't being stolen

So what's the best thing to do with lots of Brex rewards now?

"Brokerage customers who opt in to the brokerage cash sweep program (the IntraFiNetwork Deposit (IND) Sweep Service), will have their uninvested brokerage account cash automatically “swept” or moved into deposits at a network of program banks. There are currently six program banks that participate in our brokerage cash sweep program. Cash deposited to these banks will be eligible for FDIC insurance up to a total maximum of up to $1.5 million (up to $250,000 per program bank, inclusive of deposits you may already hold at the bank in the same ownership capacity). While the FDIC insurance coverage limit at each bank is $250,000, $2,000 is reserved for accrued interest."

-https://robinhood.com/us/en/support/articles/deposit-sweep-p...

For the record, that's not what the article is saying.

A better TL;DR might be:

There's a success sequence which makes people not be poor (97% effective) regardless of IQ. Elites hide this sequence from the masses perhaps because they believe it will make more people blame people for being poor. The article argues that the success sequence is easy enough that substantially more people would complete it and live better lives if they were made aware of it, and especially if it were tied more to social status rather than other things being tied to social status which decrease their probability of success in life.

Short term we may see the price drop a lot, but long term it's never a great idea to bet against Elon Musk

Starting from this URL instead allows you to see the full thread: https://twitter.com/mtaibbi/status/1598825121992638465

They do use a lot less water

I thought it was pretty interesting to read from the guy himself on his own site.

This one is quite well done: https://peaksalvation.com/socioeconomic-bends

https://darkhn.herokuapp.com/