Projects generally don't have a pre-existing webpage. You have a company or lab website and then you have for example a 100 page NIH R01, or maybe a 5 page whitepaper for a proposal. We help you create what we think is an engaging webpage that's easy to understand and scan for all the important points: What are you doing, what's the impact, cost and timeline, your skills. And then let you embed media and attach files (with managed access)
I think anonymous DMs are probably worse and lead to more spam. People are much less likely to send dangerous or low quality inbounds if they have to attach themselves to it. I think there's value actually for funders to also be able to DM a researcher and say hey your work is interesting and we're exploring this area; same for researcher to researcher. We enable you to block others (and if it becomes a problem could turn off your inbox). We also have public commenting for ppl to ask questions or provide feedback in open
How recommendations work today (assume we'll write better algos over time) is that we basically take everything on your profile and activity to create a vector representation of your interests and then match on vector similarity. We secondly also maintain a social graph so can run things like mutual follows (or followers), mutual donations etc.
I think journals will remain in the near-mid future as a way of providing credibility and filtering work so ppl in a specific niche know what's "important" to read. I think in that world it's important we try to reduce the massive amount of wasted time and effort it takes scientists to get their work published. In the meantime we hope people share their results on preprint servers like arXiv or share results on Catalyze for projects they previously posted.
I agree I really think we are quite far from saturation points in most areas. There are opportunities to spend trillions of dollars in LMICs on medicine, food, education, infrastructure etc. In other areas we will have increasing capital needs as low hanging fruit are picked. The biggest question isn't what is saturated (I think we are far from global utopia) but how do the marginal values per dollar shift and are we continuing to balance where the next dollar should go.
This is a great idea. I've tried to tackle this in two ways: 1) My company is working with several government agencies and large foundations to pre-identify who we want to apply and proactively reach out. That way we don't just wait and see who finds a solicitation online but directly go to people we think have the right skills and expertise to achieve what we're looking for to avoid wasting people's time
2) On the platform I operate (think Manifund but for all R&D areas) we let people pitch out their ideas directly and can have your page created off existing materials in ~1 min to minimize time investment. We use that to: a) immediately surface bilateral recommendations between you and funders or research collaborators where we determine there to be very high alignment b) make it easy to directly message each other, share files, and execute financial transfers to reduce time/cost burden c) use the enriched profiles to help our engagement work in #1.
We're also looking to deploy similar tech with academic journals for people to pre-apply and get a probability of how likely their work is to be accepted and what reviewer feedback is likely to be before tons of time is spend going through the review/revision process.
Would love to discuss - feel free to comment or DM.
I'm not sure I see secular declines in this graph. The largest signal for valuations is really just the state of the market -rather than how much progress is being made - are we in a bull market for VC or not. You'd have to look a lot further to see how much faster AI is making them.
Anecdotally I do hear revenue is ramping way faster than it used to - that has also led to a shift in valuation/round expectations i.e. $10M ARR is the new $1M ARR. AI drastically increases your ability to build and deploy software rapidly which allows both you and competitors to scale more rapidly. This should lead to a world where VCs eventually get to "buy" way more revenue for a given dollar, but we also are going to see declines in profitability, both because tokens decrease gross margins and because increased product competition will mean increased sales and marketing spend, service level requirements etc. YC is going all in on vertical/domain-specific AI e.g. find a niche, build something that probably a 100 others can, and then go use our platform to market harder and ramp distribution as fast as you can.
This makes a lot of sense. I will say though the $20m headline is at API price which is obviously inflated vs electricity/compute prices. I think a lot of similar work is already possible with open source models. It would be good to give a really hard problem to them and then basically create a graph of the speed, cost tradeoff with different numbers of agents in a swarm to calculate what the tradeoffs look like today.
Transfecting viruses or bacteria with external DNA is ridiculously simple. There are thousands of kids who do this in middle school research internships and hundreds of thousands who do this in college. Ordering oligo-nucelotides is also very easy to do online. The point of AI biorisk is that you can quickly design a thousand different versions of something many of which will be modifications on known pathogens, many of which might be totally new. You can recombinate those sequences into existing viruses or bacteria or fungi. The biggest piece here is that you don't need to be perfect. You nee only a few to hit - having a 1% or 0.1% success rate is OK. Moreover, they all mutate meaning that even if you didn't do a perfect job they can get more dangerous over time.
I don't think they will fund until there's very clear data that you can translate educational spend into DALYs. Trying to extrapolate studies from the US or other OECD countries to LMICs is probably a really bad idea. Besides spending on education in general we know that education isn'ta fixed commodity that's fungible and infinitely scalable the way pharmaceuticals are i.e. there can be differing quality between teachers and school systems and highly different spillovers based on local economic structure.
I'm happy to see them invest in studying the effects. I think it's important to know that goodwell shouldn't be considered the end all be all for how to give effectively - they operate under one very specific methodology and I'd view them as being a good source for maximizing dollars/DALY in global health vs how to think about all philanthropy.
Governments are making a huge risk by deprioritizing biosecurity when we're rapildy democratizing access to expertise (via AI) needed to created deadly pathogens
Thanks for replying - my responses in order
I agree I really think we are quite far from saturation points in most areas. There are opportunities to spend trillions of dollars in LMICs on medicine, food, education, infrastructure etc.
In other areas we will have increasing capital needs as low hanging fruit are picked. The biggest question isn't what is saturated (I think we are far from global utopia) but how do the marginal values per dollar shift and are we continuing to balance where the next dollar should go.
This is a great idea. I've tried to tackle this in two ways:
1) My company is working with several government agencies and large foundations to pre-identify who we want to apply and proactively reach out. That way we don't just wait and see who finds a solicitation online but directly go to people we think have the right skills and expertise to achieve what we're looking for to avoid wasting people's time
2) On the platform I operate (think Manifund but for all R&D areas) we let people pitch out their ideas directly and can have your page created off existing materials in ~1 min to minimize time investment. We use that to:
a) immediately surface bilateral recommendations between you and funders or research collaborators where we determine there to be very high alignment
b) make it easy to directly message each other, share files, and execute financial transfers to reduce time/cost burden
c) use the enriched profiles to help our engagement work in #1.
We're also looking to deploy similar tech with academic journals for people to pre-apply and get a probability of how likely their work is to be accepted and what reviewer feedback is likely to be before tons of time is spend going through the review/revision process.
Would love to discuss - feel free to comment or DM.
Can check out website here: rndcatalyst.com
I'm not sure I see secular declines in this graph.
The largest signal for valuations is really just the state of the market -rather than how much progress is being made - are we in a bull market for VC or not. You'd have to look a lot further to see how much faster AI is making them.
Anecdotally I do hear revenue is ramping way faster than it used to - that has also led to a shift in valuation/round expectations i.e. $10M ARR is the new $1M ARR. AI drastically increases your ability to build and deploy software rapidly which allows both you and competitors to scale more rapidly. This should lead to a world where VCs eventually get to "buy" way more revenue for a given dollar, but we also are going to see declines in profitability, both because tokens decrease gross margins and because increased product competition will mean increased sales and marketing spend, service level requirements etc. YC is going all in on vertical/domain-specific AI e.g. find a niche, build something that probably a 100 others can, and then go use our platform to market harder and ramp distribution as fast as you can.
This makes a lot of sense. I will say though the $20m headline is at API price which is obviously inflated vs electricity/compute prices. I think a lot of similar work is already possible with open source models. It would be good to give a really hard problem to them and then basically create a graph of the speed, cost tradeoff with different numbers of agents in a swarm to calculate what the tradeoffs look like today.
Transfecting viruses or bacteria with external DNA is ridiculously simple. There are thousands of kids who do this in middle school research internships and hundreds of thousands who do this in college. Ordering oligo-nucelotides is also very easy to do online. The point of AI biorisk is that you can quickly design a thousand different versions of something many of which will be modifications on known pathogens, many of which might be totally new. You can recombinate those sequences into existing viruses or bacteria or fungi. The biggest piece here is that you don't need to be perfect. You nee only a few to hit - having a 1% or 0.1% success rate is OK. Moreover, they all mutate meaning that even if you didn't do a perfect job they can get more dangerous over time.
I don't think they will fund until there's very clear data that you can translate educational spend into DALYs. Trying to extrapolate studies from the US or other OECD countries to LMICs is probably a really bad idea. Besides spending on education in general we know that education isn'ta fixed commodity that's fungible and infinitely scalable the way pharmaceuticals are i.e. there can be differing quality between teachers and school systems and highly different spillovers based on local economic structure.
I'm happy to see them invest in studying the effects. I think it's important to know that goodwell shouldn't be considered the end all be all for how to give effectively - they operate under one very specific methodology and I'd view them as being a good source for maximizing dollars/DALY in global health vs how to think about all philanthropy.
Governments are making a huge risk by deprioritizing biosecurity when we're rapildy democratizing access to expertise (via AI) needed to created deadly pathogens