Fucking banger of a tweet.

Which will never happen.

Leaders of AI companies know this better than anyone. They understand what is at risk here, and what they stand to gain. But for them they are caught up in the race. Everyone is racing to get there first, to be the one to birth a super intelligent god. The stakes are too high to stop now.

People talk about UBI being a solution, however the revenue will be centralised, and taxed in one country, likely the US or China. So UBI could be a solution for their citizens, but digital intelligence has global impact. I worry about how this will exacerbate global inequality further.

When you have the ability to deploy digital intelligence, that is a capability that transcends political borders. These companies will have revenue streams larger than the GDP of many major economies.

AI companies themselves will be able to move faster than any regulation or policy making can reasonably keep up. There will be a major issues from AI agents failing or being misused. Hopefully the impact is just financial and not putting people at risk.

There will be a huge effect on the workforce, as well as the makeup of businesses. I foresee the risk in start ups dropping if you can quickly scale with digital agents before your first hire. Larger businesses slower to adopt will be challenged from all sides in ways that were previously impossible

At which point you have a land grab. Who can roll out the best agent for paralegals? Business strategists? Etc… I think domain specific agents will enable businesses at all scales to rapidly scale their workforce. Why hire a junior dev when a digital agent will work 30x quicker and be much smarter?

Once you have AI doing AI R&D, then you can scale it up in parallel, accelerating the time to new models and thus deeper intelligence. I don’t think this will lead to AGI, but it will provide the ability to have highly specialised models for domains other than software dev.

The gap is the novel research, if they can hit that goal before the investment bubble bursts, then we have some major shifts ahead. (And to be clear the anthropic model is one example we know of, I’m certain there’s more powerful ones behind closed doors.

Every investment in the tech world right now is investing in getting AI to hit recursive self improvement. That’s AI doing the end to end development as well as the novel research to new methods. Modern AIs can already do the end to end dev, anthropic have shown agents working for 30h unaided…

Couldn’t agree more. This past few weeks I’ve taken an approach of collab on an ADR with an AI, get it to build a plan to implement. Let it implement, and I act as code reviewer / pair partner. It works really well.

Astride of a grave and a difficult birth, we reload.
Down in the hole, lingeringly, the grave-digger picks up the chainsaw; the Arch-Vile ignites us once again.
We have time to grow old.
The air is full of our cries.

It becomes sensible at the ~10kW range, many satellites operate on these power budgets, but that still only gets you a rack in space. Launching 1000 of these seems possible, but you end up with a slow, impossible to maintain, distributed system in space. Sounds cool, but useless for the cost.

Huh, did some math and it turns out that the solar density out paces the need for heat dissipation. For a small 10MW data centre you’d need 40,000m^2 in solar panels and if running at 30C, about 25,000m^2 of radiators. Before you factor in radiation hardening, micro meteors or how you get it there!

Nor understands how hard it is to manage thermals in an environment where you can only bleed heat via radiation.

I love this idea, what proof checkers have you tried integrating it with?