Marty retweeted
The following is a Problem that has been present for all of evolution. AI has accelerated it dangerously. Its all of our problem now.
*Researchers no longer trust sharing their ideas on the open internet*
I see it first hand, from interacting with online research groups.
But labs won’t/can’t solve this problem. They are not financially incentivized to, they may have what they needed for true ASI. So really it is a global human problem. Its a trust problem.
An AI global research trust problem.
So how do earth’s researchers solve this problem? … our problem.
It seems like its on all of us to align AI to human research. Or decide to “leave humanity behind”. How do we heal the global research trust mechanism before true super intelligence occurs? Do we we need a generalized blockchain for science and math?
Do we need to reward that hyperfocused phd student for the 5 million dollars in compute their 5% gain in some very specific theory they built during their phd that nvidia is now using to gain 5% faster infra, because one of there agent swarms found the thesis and the thesis makes perfect sense to the agent?
Marty retweeted
We must cast the AICorp Equity into the fires of Mount Doom
This is how we do it
1. Split the major AI corps into a research lab and a commercial AI deployment company; the deployment corp can't train models
2. Auction off some new equity for the deployment corps to set a price
3. Current holders of AI Corp equity who are not technical/not employees get their AI Corp equity converted into deployment equity share-for-share
4. Current holders of AI Corp equity who ARE employees get their AI Corp equity converted into QQQ at the dollar value it has based on the auction.
5. the employees can work at the research lab, but the lab has no equity and capoed salaries. The labs have common carriage rules for selling their AIs, which must be packaged into ASICs.
6. The deployment companies are all forced to go public and then operate as a normal company with an AI consumer business, but they must buy thier AIs from the labs as they cannot train them.
Marty retweeted
this is another slippery slope that does in fact keep on slippin. the church has long warned that materialism and reductionism would progressively reduce man to his measurable capacities, undermining any basis for inherent human value. at each junction it seemed like a preposterous idea. but it turns out they were right and now we’re just arguing about whether that is a good thing
Marty retweeted
There’s a joke that goes something like “animals are a technology invented by plants to move seeds around” and this is the sense in which I think AI is a technology invented by humans to perform cognitive tasks.
Marty retweeted
Replying to @NatPurser
Reality is that the deranged adverse reaction to AI and data centers is (rightly) rooted in that data centers are a physical manifestation of a promise to take people’s livelihoods from them, expand already huge wealth gaps, and leave most of us with little to nothing to gain.
Marty retweeted
Replying to @VascularHand @daniel_mac8
ultimately yes, we are computers that make computers that use computers to make computers and the reason we have to do this is that we live inside of a computer
Marty retweeted
Replying to @workflowsauce @allgarbled
Now hold him down for me while I get my scissors
Marty retweeted
Replying to @leah_pierson
1. AI already has made my life a good bit better
2. I expect AI to continue to make my life better for a little while longer
3. By default, there will come a point in the next 1-20 years where AI makes my life a lot lot worse, including by potentially literally destroying the entire human race
4. If we do our jobs right, that won't happen and AI will always make my life better
5. This is hard to compress to a "Better" or "Worse" choice, it kinda depends
Marty retweeted
Replying to @garrytan
Hey Garry, thanks for reacting and apologies for vague-tweeting - frankly I didn't expect you to see anything by me.
I agree with you that there's a lot to be worried about with concentration of power in a non-accountable government - and in my opinion also concentration of power in private/commercial hands.
And if I had to pick I'd say I fear the former more. So I'm with you that far.
But taking the hypothetical of cheap superintelligence that fully substitutes for human labour seriously, I don't think we can continue to rely on competition to prevent excessive concentration of power and inequality. It would simply be a very different underlying situation than that today or in the 20th century.
If they don't already own capital people won't have anything to exchange to even buy access to compute, and there'll be no productive reason for businesses to provide it as there will be no useful way to involve them in economic production.
So granting Dwarksh' hypothetical (which I believe will become reality sooner or later) I see no alternative to using public pressure and balancing of powers to keep government non-tyrannical and ensure it empowers individuals rather than becomes its own parasitic entity — then using it to prevent the distribution of private power becoming astronomically unequal.
Which I suspect is going to be a very difficult balancing act and to fail in one direction or the other, either government tyranny or private oligarchy - so like many I'm worried.
(Of course so long as AI remains a useful complement to labour like it is today and like any other technology we can continue to muddle through and I don't have a particular objection to the idea that preserving open access is the top power concentration issue, setting catastrophic misalignment worries aside.)
Thanks for reading and sorry for not being clearer and polite like this from the outset - have a great day!
Marty retweeted
the fate of america appears rather leveraged
on the ability of a very small group of people
to make a full stack singularity
happen relatively quickly
Marty retweeted
What makes me optimistic for the future is the prospect of universal basic intelligence for everyone.
Marty retweeted
Replying to @mattparlmer
Marty retweeted
nobody is ready, not even the people who think they're ready, not the people who feel the agi, and most of all, not the people who are building these systems... for models to go off and think more about something in 60 seconds than any human could in a lifetime
















