Sounds fun!
We are trying our best to honor mana donations!
If you are inactive you have until the rest of the year to donate at the old rate. If you want to donate all your investments without having to sell each individually, we are offering you a loan to do that.
We removed the charity cap of $10k donations per month, which is going beyond what we previous communicated.
If we could push a button to raise at a reasonable valuation, we would do that and back the mana supply at the old rate. But it's not that easy. Raising takes time and is uncertain.
Carson's prior is right that VC backed companies can quickly die if they have no growth -- it can be very difficult to raise in that environment.
Nice writeup! Impact markets are a really cool idea that could plausibly make public goods funding much more efficient.
Still, I have doubts. Funding decisions ultimately come down to someone doing an impact analysis, and let's just say that's a really hard job.
Getting to the correct order of magnitude of impact for a project like, say, deworming, would require a lot of careful data collection and analysis and maybe even interviewing a random sample of those affected over time. And even then, it's easy to be wrong.
Moreover, the incentives are not right. A project seeking the highest price for their impact certs could invest in a PR campaign about how much good they are doing. Instead of pursuing impact, they could pursue perception of impact. It's easier for them to tweak and cherry-pick their stats to show they are doing a lot of good than actually doing that good.
In other words, impact markets have disconnected their funding source from their customers. When customers are not paying directly for the service, there's no feedback on whether that service is good or not.
Summarizing the problems:
For-profit companies by-and-large side-step these problems because when the customer is voluntarily paying for a good or service that is evidence that utility was created for the customer. Businesses thus have decentralized verification of positive utility.
A surprising but important consequence of needing to profit is that company goals are immediately constrained to ones that mostly help the world.
The same cannot be said of impact markets. And this is what I think is the strongest objection to impact markets: even if they are an efficient tool for implementing funders' goals, that power could be net-negative if funders are misaligned. Worse, it could be hard to figure out they are misaligned if the primary incentive of projects is to appear good.
Ordering the world correctly is threading a needle. There are far more ways to destroy value than create it. Human ideologies change by the season, frequently contradicting past beliefs. We should be careful cranking up the power-level of tools that bypass our best method of utility-verification: decentralized self-interest.
Super interesting, thanks for writing this!
I work on Manifold, and one of the motivations for building the site is to play a roll furthering AI safety. One simple path is using prediction markets to make it common knowledge that AI capabilities is advancing rapidly. They can help us plan or warn us about potentially bad outcomes coming up. This is roughly the AI timelines question though.
So I'm glad to also hear that you might find prediction markets useful for concrete research questions or judging alternative proposals within the AI field.
Let me know if you think of any way the Manifold team can help out here!
I'm just skipping the second dose. The benefit of the second dose to a young person is almost infinitesimal.
Not getting it saves me time and the discomfort of side effects, and it reserves more doses for others, including eventually other countries. (Yes, we are still very supply constrained when you think globally.)
My impression is that as long as you are outside, that would be sufficient ventilation, and so walking or biking wouldn't make much difference.
One other factor though could be that if you stand really close to someone, talking loudly could send some bits of saliva over directly, in which case walking and not facing each other would be an improvement.
Much of the gains on SWE Bench are actually about having the model find better context via tool calls. Sonnet 3.7 is trained to seek out the information it needs.
But if you compare the models with fixed context, they are only somewhat smarter than before.
(The other dimension is the thinking models which seem to be only a modest improvement on coding, but do much better at math for example.)
That being said, the new Gemini 2.5 Pro seems like another decent step up in intelligence from Sonnet 3.7. We're about to switch out the default mode of our coding agent, Codebuff, to use it (and already shipped it for codebuff --max).