Put a price on breakthroughs
alexwang.ai"We’ll acquire any team that matches our last-generation model on this eval suite using a tenth of the training compute, for $2B."
I'm not sure how the team expects to get paid unless they actually hit. And its a rat race too by the structure so not only are you not getting paid you are incentivized to work 7 days and not sleep.
OP has it backwards really and is also missing an important step with biotech. Biotech is also not doing what they describe. If you are some tiny drug discovery startup, I'm sorry, you aren't working on moonshots either. These are risky investments, you have to prove to the investors that your idea has a snowballs chance in hell. This is how we end up with a dozen plus obesity related drugs and little interest comparatively in rare disease. They very much don't want moonshots, they want solid bets.
The moonshots you have to take a step back into academia. That is where they actually happen in biology. Grant funding agencies put out a similar call for research, but they don't just expect people to work on it for free and first one to get there wins. You apply for them with a proposal. The grants fund you while the work is actually being done.
But of course, these are low odds investments. This is why the government has to pay for it and there isn't appetite in private investment to pay for this in such structures already in biotech.
>This is why the government has to pay for it and there isn't appetite in private investment to pay for this in such structures already in biotech.
But most colleges have a technology sharing agreement so that any discovery made at the college by a student or faculty becomes the property of XYZ™
Socialized costs, private gains.
There are ways around this and people routinely spin out startups in academia.
I don’t know, I feel like there’s plenty of money to fund a lot of risky bets in AI right now, across the research-commercialization spectrum. Not sure there is a need for a new form of funding the field?
Any time I hear "neolab" "neocloud" I immediately place the writing and author into the "larper-adjacent hype dealer" bucket vaguely on the same tier as "oh so this is where all the growth hackers and SEO types landed"
Pump those share prices, get those valuations up, everybody in my group chat is winning !!!
Do you prefer a different term?
The business model proposed here is complete nonsense.
Why commit 2 billion dollars for an outcome that may never arrive?
That also applies to the teams attempting this. By definition, the cost of discovering the breakthrough must be smaller than the risk adjusted return. Aka, if there is a 1% chance you can do it, you may at most invest 20 million USD into an attempt.
But this is where things get stupid. The frontier labs could just spend that money themselves. They can just host kaggle competitions with million dollar prizes and then extract a much bigger producer surplus or just straight up hire the winners to keep their mouths shut and block the competition.
As others have said, this scheme requires putting capital upfront with a high risk of failure and a non zero chance of foul play.
If I was OpenAI, I might dangle the carrot but then at the last moment say, well enough time has passed that we already have solved this problem in house ourselves and you only get us a 10% boost, we are willing to pay a discounted prize out to you. The team ends up doing what amounts to a fire sale because they have zero leverage.
Why? Clearly it seems like neolabs are willing to take the risk without it, there is enough capital to go around, whats the incentive for this structure?
Thoughtful piece that understands economics
No.