Chris Lengerich
@chrislengerich.com
How do we make scientists and engineers 100x more productive to solve problems that matter?
This feed is notebook margins - raw research notes to train a science hypothesis AI later.
Final essays go to Context Fund: https://www.reddit.com/r/contextfund
This feed is notebook margins - raw research notes to train a science hypothesis AI later.
Final essays go to Context Fund: https://www.reddit.com/r/contextfund
Turbulence in the confused middle is now a constant feature of the new plotline (smaller variance models ftw).
But you can really feel when the right plot clicks in.
But you can really feel when the right plot clicks in.
November 7, 2025 at 4:00 AM
Turbulence in the confused middle is now a constant feature of the new plotline (smaller variance models ftw).
But you can really feel when the right plot clicks in.
But you can really feel when the right plot clicks in.
Seems like it compiles and runs now. Nifty.
A friend said tonight, "Thanks!"
I said, "but I haven't done anything yet."
"No, but you gave me hope."
A friend said tonight, "Thanks!"
I said, "but I haven't done anything yet."
"No, but you gave me hope."
November 7, 2025 at 3:56 AM
Seems like it compiles and runs now. Nifty.
A friend said tonight, "Thanks!"
I said, "but I haven't done anything yet."
"No, but you gave me hope."
A friend said tonight, "Thanks!"
I said, "but I haven't done anything yet."
"No, but you gave me hope."
Finally. While we were all looking at AI, USDC on Solana became a credible tradbank replacement.
1-2 min, $0.50 (eth) -> ~15s, <$0.01 (sol)
NB: Transactions visible on chain (obv) and easy to fat-finger funds to the wrong chain (not ready for grandma yet), but still, for geeks, it just works.
1-2 min, $0.50 (eth) -> ~15s, <$0.01 (sol)
NB: Transactions visible on chain (obv) and easy to fat-finger funds to the wrong chain (not ready for grandma yet), but still, for geeks, it just works.
September 21, 2025 at 12:59 AM
Finally. While we were all looking at AI, USDC on Solana became a credible tradbank replacement.
1-2 min, $0.50 (eth) -> ~15s, <$0.01 (sol)
NB: Transactions visible on chain (obv) and easy to fat-finger funds to the wrong chain (not ready for grandma yet), but still, for geeks, it just works.
1-2 min, $0.50 (eth) -> ~15s, <$0.01 (sol)
NB: Transactions visible on chain (obv) and easy to fat-finger funds to the wrong chain (not ready for grandma yet), but still, for geeks, it just works.
Longevity metrics:
Rather than talking about expected lifespan, it's probably more intuitive/useful to think about marginal cost to extend lifespan by one year. Right now, it's cheap in early age, much more expensive in old age. If we invest together, the cost will get much cheaper in the future.
Rather than talking about expected lifespan, it's probably more intuitive/useful to think about marginal cost to extend lifespan by one year. Right now, it's cheap in early age, much more expensive in old age. If we invest together, the cost will get much cheaper in the future.
September 17, 2025 at 6:27 PM
Longevity metrics:
Rather than talking about expected lifespan, it's probably more intuitive/useful to think about marginal cost to extend lifespan by one year. Right now, it's cheap in early age, much more expensive in old age. If we invest together, the cost will get much cheaper in the future.
Rather than talking about expected lifespan, it's probably more intuitive/useful to think about marginal cost to extend lifespan by one year. Right now, it's cheap in early age, much more expensive in old age. If we invest together, the cost will get much cheaper in the future.
Have a 10-year bet running now to never read a newsfeed in person again (using AI summarizers to stay informed and newsfeed-blocking extensions/UI hacks to enforce).
Doable now, increasingly efficient over time.
Doable now, increasingly efficient over time.
September 3, 2025 at 3:33 AM
Have a 10-year bet running now to never read a newsfeed in person again (using AI summarizers to stay informed and newsfeed-blocking extensions/UI hacks to enforce).
Doable now, increasingly efficient over time.
Doable now, increasingly efficient over time.
Newsfeeds are probably dead (for human consumption).
Replaced them with AI summaries for ~a month (scheduling utilities/async flows like Lutra/OAI helpful here). Same events, but the world feels calmer/contextualized, like reading an investor newsletter.
End of an era, probably for the better.
Replaced them with AI summaries for ~a month (scheduling utilities/async flows like Lutra/OAI helpful here). Same events, but the world feels calmer/contextualized, like reading an investor newsletter.
End of an era, probably for the better.
August 28, 2025 at 5:28 PM
Newsfeeds are probably dead (for human consumption).
Replaced them with AI summaries for ~a month (scheduling utilities/async flows like Lutra/OAI helpful here). Same events, but the world feels calmer/contextualized, like reading an investor newsletter.
End of an era, probably for the better.
Replaced them with AI summaries for ~a month (scheduling utilities/async flows like Lutra/OAI helpful here). Same events, but the world feels calmer/contextualized, like reading an investor newsletter.
End of an era, probably for the better.
Would be neat to see a leaderboard re: implicitly personalized AIs (those with memory-based adaptation methods) on common benchmarks?
Pre-training still the biggest factor of AI intelligence by far, but how much does implicit prompting (via memory) change the intelligence between users?
Pre-training still the biggest factor of AI intelligence by far, but how much does implicit prompting (via memory) change the intelligence between users?
August 28, 2025 at 5:22 PM
Would be neat to see a leaderboard re: implicitly personalized AIs (those with memory-based adaptation methods) on common benchmarks?
Pre-training still the biggest factor of AI intelligence by far, but how much does implicit prompting (via memory) change the intelligence between users?
Pre-training still the biggest factor of AI intelligence by far, but how much does implicit prompting (via memory) change the intelligence between users?
Why do biotechs fail?
TL;DR - scale matters, hedging matters, efficiency of a trial matters. Often the science fails first, leading to cascading management/funding failures.
A lot of knobs here that can be tweaked (some obvious, others less obvious).
www.librariesforthefuture.bio/p/why-do-bio...
TL;DR - scale matters, hedging matters, efficiency of a trial matters. Often the science fails first, leading to cascading management/funding failures.
A lot of knobs here that can be tweaked (some obvious, others less obvious).
www.librariesforthefuture.bio/p/why-do-bio...
Anatomy of a Biotech Failure
Eight Case Studies in Recent Shutdown History
www.librariesforthefuture.bio
August 28, 2025 at 5:01 PM
Why do biotechs fail?
TL;DR - scale matters, hedging matters, efficiency of a trial matters. Often the science fails first, leading to cascading management/funding failures.
A lot of knobs here that can be tweaked (some obvious, others less obvious).
www.librariesforthefuture.bio/p/why-do-bio...
TL;DR - scale matters, hedging matters, efficiency of a trial matters. Often the science fails first, leading to cascading management/funding failures.
A lot of knobs here that can be tweaked (some obvious, others less obvious).
www.librariesforthefuture.bio/p/why-do-bio...
Re: AI markets, would much rather see new markets for new asset classes and geos, rather than new markets for new actors (AI agents, humans w/ AI tools). Even with algos running 75% of trades in the market now, few markets have been created only *for algos.*
August 9, 2025 at 9:18 PM
Re: AI markets, would much rather see new markets for new asset classes and geos, rather than new markets for new actors (AI agents, humans w/ AI tools). Even with algos running 75% of trades in the market now, few markets have been created only *for algos.*
Typically, new ideas come from (different) fringes of the network where diverse threads start to intersect, then travel towards the center to get power (but may pick up competitors in the process).
August 9, 2025 at 7:55 PM
Typically, new ideas come from (different) fringes of the network where diverse threads start to intersect, then travel towards the center to get power (but may pick up competitors in the process).
For how similar they are in world models, VC and founders diverge quite a lot in actions.
VC - horizontal, power accumulation, knows everyone.
Founder - vertical, deep, iterates, knows users, team and funders.
Growthco - horizontal x vertical (a la OpenAI/Google/Alibaba).
VC - horizontal, power accumulation, knows everyone.
Founder - vertical, deep, iterates, knows users, team and funders.
Growthco - horizontal x vertical (a la OpenAI/Google/Alibaba).
August 9, 2025 at 7:05 PM
For how similar they are in world models, VC and founders diverge quite a lot in actions.
VC - horizontal, power accumulation, knows everyone.
Founder - vertical, deep, iterates, knows users, team and funders.
Growthco - horizontal x vertical (a la OpenAI/Google/Alibaba).
VC - horizontal, power accumulation, knows everyone.
Founder - vertical, deep, iterates, knows users, team and funders.
Growthco - horizontal x vertical (a la OpenAI/Google/Alibaba).
Hedging: $, portfolio, iteration, expertise, (sometimes) specialized contracts.
Key insight into founders is how they react uncertainty - when you encounter uncertainty, do you raise more to hedge (portfolio hedge), or do you raise less, think more (expertise hedge)?
Key insight into founders is how they react uncertainty - when you encounter uncertainty, do you raise more to hedge (portfolio hedge), or do you raise less, think more (expertise hedge)?
August 9, 2025 at 6:50 PM
Hedging: $, portfolio, iteration, expertise, (sometimes) specialized contracts.
Key insight into founders is how they react uncertainty - when you encounter uncertainty, do you raise more to hedge (portfolio hedge), or do you raise less, think more (expertise hedge)?
Key insight into founders is how they react uncertainty - when you encounter uncertainty, do you raise more to hedge (portfolio hedge), or do you raise less, think more (expertise hedge)?
The Internet right now looks like a global market on the surface (anyone can read!), but it's actually a bunch of local markets disconnected from each other, unless you're already shipping at scale. What are your short growth paths to resource and user communities?
August 8, 2025 at 6:42 PM
The Internet right now looks like a global market on the surface (anyone can read!), but it's actually a bunch of local markets disconnected from each other, unless you're already shipping at scale. What are your short growth paths to resource and user communities?
1. The economic value of prediction scales with your (real) action space.
2. What you predict about the world also depends on your (simulated) action space.
Explains a lot about value systems in the modern, hyper-scale world vs. the 150-person village.
2. What you predict about the world also depends on your (simulated) action space.
Explains a lot about value systems in the modern, hyper-scale world vs. the 150-person village.
August 8, 2025 at 6:15 PM
1. The economic value of prediction scales with your (real) action space.
2. What you predict about the world also depends on your (simulated) action space.
Explains a lot about value systems in the modern, hyper-scale world vs. the 150-person village.
2. What you predict about the world also depends on your (simulated) action space.
Explains a lot about value systems in the modern, hyper-scale world vs. the 150-person village.
The line between investment and cognitive labor is starting to get a bit blurry.
Using AI agents to do work directly, yielding $, is not that different from investing money in another human verifier to do the same thing, except ownership fraction and control (which does, of course, matter).
Using AI agents to do work directly, yielding $, is not that different from investing money in another human verifier to do the same thing, except ownership fraction and control (which does, of course, matter).
August 6, 2025 at 1:10 AM
The line between investment and cognitive labor is starting to get a bit blurry.
Using AI agents to do work directly, yielding $, is not that different from investing money in another human verifier to do the same thing, except ownership fraction and control (which does, of course, matter).
Using AI agents to do work directly, yielding $, is not that different from investing money in another human verifier to do the same thing, except ownership fraction and control (which does, of course, matter).
High-variance techs often lead to a multimodal future (not necessarily, but often).
July 28, 2025 at 3:01 PM
High-variance techs often lead to a multimodal future (not necessarily, but often).
If you can with >70% confidence, build the vertically integrated thing (definitely optimistic).
Otherwise, build the personal tools that upgrade the tech era, which makes the vertically integrated thing possible, and be more indefinitely optimistic.
Otherwise, build the personal tools that upgrade the tech era, which makes the vertically integrated thing possible, and be more indefinitely optimistic.
July 26, 2025 at 4:09 PM
If you can with >70% confidence, build the vertically integrated thing (definitely optimistic).
Otherwise, build the personal tools that upgrade the tech era, which makes the vertically integrated thing possible, and be more indefinitely optimistic.
Otherwise, build the personal tools that upgrade the tech era, which makes the vertically integrated thing possible, and be more indefinitely optimistic.
Arguably enough money earmarked for AI companies now that we can start talking about momentum vs. fundamental strategies in VC.
True for the AI labor markets too.
The music will probably stop for some big names in 1-2 years in dramatic ways, but the winners do have 10x-100x growth left.
True for the AI labor markets too.
The music will probably stop for some big names in 1-2 years in dramatic ways, but the winners do have 10x-100x growth left.
July 17, 2025 at 8:02 PM
Arguably enough money earmarked for AI companies now that we can start talking about momentum vs. fundamental strategies in VC.
True for the AI labor markets too.
The music will probably stop for some big names in 1-2 years in dramatic ways, but the winners do have 10x-100x growth left.
True for the AI labor markets too.
The music will probably stop for some big names in 1-2 years in dramatic ways, but the winners do have 10x-100x growth left.
Re-posting from Engage Bio (personal portfolio)
Exciting news from Engage Bio!
We’re proud to announce that the Cystic Fibrosis Foundation has made an equity investment in Engage to support development of durable, non-viral genetic medicines for cystic fibrosis.
Exciting news from Engage Bio!
We’re proud to announce that the Cystic Fibrosis Foundation has made an equity investment in Engage to support development of durable, non-viral genetic medicines for cystic fibrosis.
Exciting news from Engage Bio! | Will Olsen
Exciting news from Engage Bio!
We’re proud to announce that the Cystic Fibrosis Foundation has made an equity investment in Engage to support development of durable, non-viral genetic medicines for c...
www.linkedin.com
June 25, 2025 at 9:09 PM
Re-posting from Engage Bio (personal portfolio)
Exciting news from Engage Bio!
We’re proud to announce that the Cystic Fibrosis Foundation has made an equity investment in Engage to support development of durable, non-viral genetic medicines for cystic fibrosis.
Exciting news from Engage Bio!
We’re proud to announce that the Cystic Fibrosis Foundation has made an equity investment in Engage to support development of durable, non-viral genetic medicines for cystic fibrosis.
Objectively valuable milestones - a conversation that comes up again and again in hardtech. Hardtech cos/intrinsically motivated scientists often have complex intermediate goals and unique theory. The market, though, is (on average), not enlightened and not patient.
youtu.be/aG8_AeiwDPA?...
youtu.be/aG8_AeiwDPA?...
Celine Halioua: The Key to Longevity
YouTube video by S3
youtu.be
June 23, 2025 at 5:18 PM
Objectively valuable milestones - a conversation that comes up again and again in hardtech. Hardtech cos/intrinsically motivated scientists often have complex intermediate goals and unique theory. The market, though, is (on average), not enlightened and not patient.
youtu.be/aG8_AeiwDPA?...
youtu.be/aG8_AeiwDPA?...
Investment learnings from the last few years:
1. Contrastive distillation works better than most theories as a model of reasoning and writing. It's simple, it scales.
2. But data/experience increases precision. Hang out near the reefs. PhD-led startups with good coaches do unreasonably well.
1. Contrastive distillation works better than most theories as a model of reasoning and writing. It's simple, it scales.
2. But data/experience increases precision. Hang out near the reefs. PhD-led startups with good coaches do unreasonably well.
June 4, 2025 at 5:13 PM
Investment learnings from the last few years:
1. Contrastive distillation works better than most theories as a model of reasoning and writing. It's simple, it scales.
2. But data/experience increases precision. Hang out near the reefs. PhD-led startups with good coaches do unreasonably well.
1. Contrastive distillation works better than most theories as a model of reasoning and writing. It's simple, it scales.
2. But data/experience increases precision. Hang out near the reefs. PhD-led startups with good coaches do unreasonably well.
Only 24 hours in a day - focusing on flow with portcos and family matters more than one might think.
June 4, 2025 at 1:46 AM
Only 24 hours in a day - focusing on flow with portcos and family matters more than one might think.
Reposted by Chris Lengerich
Scientists we co-fund at @CambridgeUni.Bsky.Social have developed an algorithm that could pave the way for more personalised cancer treatments! It can pinpoint patterns of mutation in cancer cells and could help identify the patients who could benefit most from immunotherapy.
Read more 👇
Read more 👇
Scientists develop a 'metal detector' to find cancer's weak points - Cancer Research UK - Cancer News
PRRDetect, which was developed with our funding, picks out cancers that are vulnerable to immunotherapy by scanning for mutation patterns.
cruk.ink
April 10, 2025 at 3:36 PM
Scientists we co-fund at @CambridgeUni.Bsky.Social have developed an algorithm that could pave the way for more personalised cancer treatments! It can pinpoint patterns of mutation in cancer cells and could help identify the patients who could benefit most from immunotherapy.
Read more 👇
Read more 👇
This is probably right:
www.economist.com/finance-and-...
www.economist.com/finance-and-...
Why China thinks it might win a trade war with Trump
The country’s officials vow to “fight to the end”
www.economist.com
April 10, 2025 at 1:33 AM
This is probably right:
www.economist.com/finance-and-...
www.economist.com/finance-and-...