Founders

I geeked out analyzing "THE AMERICAN DYNAMISM 50: AI EDITION" by Andreessen Horowitz, so I figured I'd post it in a thread here where it won't get buried by the algorithms. I'd love to hear other folks' takes on this as well! https://a16z.com/american-dynamism-50-ai/ @pmarca

I’m an entrepreneurship addict doing strategy & business development for startups.

Overall takeaways: very narrow, male, conservative definition of “natl interest.” Is our ability to navigate space more important to natl interest than fixing black maternal mortality rate? Or preventing the nearly 50K gun deaths per year? But there is some edtech/health here, so it's not just a narrow definition

Anyway, moving on. Definition/choice noted. 
Here’s my reaction to the companies they do talk about:
1. Aerodrome: So cool & so helpful in practice, though in theory so Minority Report & so easy to see how this could go badly.

2. Agility Robots: cool re: last mile delivery! Lol Amazon take note since apparently you’ve burned & churned your way through most of the human labor force :/

2. Agility Robots: cool re: last mile delivery! Lol Amazon take note since apparently you’ve burned & churned your way through most of the human labor force :/ https://www.vox.com/recode/23170900/leaked-amazon-memo-warehouses-hiring-shortage

& upgrading all that equipment to modern comms equipment as per FAA’s NextGen initiative would require major investment coordinated between a ton of folks? This feels like tinkering in the margins if flight time is one of your main KPIs.

https://www.gao.gov/products/gao-24-105254#:~:text=NextGen%20is%20FAA's%20multi%2Ddecade,satellite%20navigation%20and%20digital%20communications.

4. Ambient.AI: Minority Report again. And you’d have to be SOOOO careful about that training data & how you avoid bias (because studies show that excluding #demographic data *increases* #bias because basically the training data is likely to include historical bias,

& ML algos can create proxies by triangulating other factors to predict demographics [& it reproduces/exacerbates historic bias]).

& ML algos can create proxies by triangulating other factors to predict demographics [& it reproduces/exacerbates historic bias]). https://hbr.org/2023/03/removing-demographic-data-can-make-ai-discrimination-worse

5. AMP Robotics: interesting & important development in #recycling and #wastemanagement. Though the founder’s last name is Horowitz - any relation? (No shade but seems like something you should disclose? And/or address?)

6. Amperon: k cool. #ElectricityGrids, #machineLearning, yada yada (with presumably significant economic impact). I’d also be curious about #Porters5 here (seems replicable?) but if Andreessen Horowitz is writing about you you’ve probably got some excellent political capital & can beat others in the space out.

7. Anduril Industries: #yikes. Scary bomby #drones. 
8. Applied Intuition: k, cool. Better tools/software for #selfDrivingVehicles

9. Built Robotics: k, cool, #robots building #solarPower stuff. But isn’t an inch kind of a significant difference in construction applications? If memory (and my laughably limited woodworking knowledge) serves, my dad measures his projects by 1/16 inch. I doubt commercial #construction is that much sloppier?

(I mean you’re definitely doing cool stuff and heading in the right direction. But seems like a ways to go before you’re ready for prime time/sales?)

10. Chef Robotics: k, cool. Seems pretty straightforward.
11. Covariant: cool, good stuff. :-p more helpful #tech for Amazon as it continues 2alienate the human workforce.
12. Databricks: K. Basically looks like it helps companies build LLMs from their own internal data in a private/more secure/governance-friendly way

13. Dexterity: AI-guided, autonomous robots for manufacturing/warehouse applications. K.
15. Figure: cool. Robots for dangerous/undesirable/understaffed jobs.
19. Hive AI: cool. AI-based online/community moderation. 
22. LeoLabs: don’t crash your satellite.

14. Farm-NG: super cool robots for all kinds of farms in theory, though in practice its commercial success depends heavily on economics. I don’t think of farmers as having particularly high margins, and therefore I would be surprised if small farmers had the capital to invest in this kind of thing.

(Unless by “small farms” you mean vanity projects of rich people.) In the #DevelopingWorld you’d also have to be careful considering the availability of replacement parts and trained [robot] mechanics to fix them if/when they break when estimating market size/sales plans/projections.

16. Flock Safety: hm. License plate reading technology. …how is that different/better than the existing cameras that decided I didn’t stop at that stop sign even though the time stamp/distance they recorded would seem to indicate that I did??? (I’m not bitter at all, I don’t know what you’re talking about.)

17. Floodbase: …now I’m super curious about what flood data governments & insurance companies already collect. Because my understanding of this type of thing is that government (and insurance companies’) lack of response to this kind of thing is a function of (a lack of) political will, not a lack of data.

Like when you buy a house there are maps & they tell you if your house is in a floodplain, which is part of how your insurance company determines coverage in the first place. So how is your data different/better than their data? (And different enough to change the politics of it)

18. Gecko Robotics: This gets into a question of how much of energy company’s lack of repairs are a function of a lack of knowledge vs. a lack of institutional will(ingress to spend money). Massey Mines & BP's (pre-accident) #OSHA violations indicate it's a lack of willingness to pay.

20. Kobold Metals: AI for mineral mapping. …oh the places we’ll go. Super curious about the geopolitical implications of *this* product. Cuz like…yeah. I wonder where they’ll find stuff.

21. LabelBox: …k. Um. Is this AI literally just creating labels? Like you make it sound a lot fancier than that, and maybe you’re talking about very sophisticated labels…Yeah I’m curious what kind of hardware we’re talking about here, because this mostly seems like software…

23. Liminal Insights: (see my reaction re: Lumafield, sounds like they’re in a similar space) Same question re: the nominal cost of ultrasound equipment.

24. Locus Robotics: Robots as a service (for warehouses/logistics). Isn’t the economic advantage of robots that if you spend the upfront investment to buy the robot it’ll decrease your labor costs in the long run? Does renting robots save $ or are you straight up stealing jobs?

25. Lumafield: CT-ish scanning for product design quality assurance. What are the economics here? What products are so individually expensive to produce that it’s cheaper to rent a special CT scanner than to cut one open?

It says they use this for medical devices, batteries, electronics, etc. but I wouldn’t think the economics of those products would justify this expense?

26. Modern Intelligence: AI for defense stuff. And…Cutlass…didn’t I see another similar company with that product name?
27. Mythos AI: …okay if the port is the bottleneck, how will building “marine highways and AI-based maps” help? Like…aren’t you speeding up supply into the bottleneck?

28. Osaro: yet more warehouse robots. 
29. Outrider AI: …how many trucks/trailers do #logistics companies lose??
30. Pano AI: AI-based forest fire detection. Cool.

31. Parallel Systems: okay so autonomous green trains that avoid switchyards…so are you building your own rail infrastructure to avoid switchyards? I have no idea what bottlenecking looks like in the train (vs. trucking) system.

32. Prepared: Smartphone data access for 911 (or 112 in Europe), seems like they have significant market penetration already). I'd love to see them leverage AI to assist with the lack of 911 staff & wait/hold times, but this is cool too. I'd also want it to stop sending data/cuts off access at some point, cuz privacy

33. Primer AI: oooh, suuuper creepy & messed up. Minority Report for the military AND Fortune 500 companies. Cuz Fortune 500 companies’ influence on military operations has never been problematic, so let’s give them those capabilities directly. Cool cool cool.

34. Rain Aero: K. Cool. Cameras & AI & stuff to optimize & speed initial wildfire response, including autonomous helicopters. …if you have 1,000 cameras, what density & geographic area are we talking about?
35. Rapid Robotics: out-of-the-box robots for manufacturers. Cool. Seems like a good idea in theory at least.

36. RRAI: um. Open architecture for defense and commercial vehicles. …doesn’t open architecture make it easier to hack things? ...which with defense products would be a bad thing? 
37. Saildrone: Sounds like the movie Twister (1996 with Helen Hunt)
38. Saronic Technologies: k cool, ocean boat monitoring stuff

39. Scale AI: cool - helps companies build their own LLMs. So obviously the big question is what’s their position on DEI & eliminating bias from training data & rules & such.

40. Shield AI: military/government drone stuff. Wow, bragging about Israeli Defense Forces using your product in the aftermath of Hamas attacks 2locate “civilians and barricaded shooters.” Cuz IDF has been so careful about protecting civilians in Gaza attacks…Interesting marketing choice. SkyNet would be proud.

41. Skydio: cool. Camera-equipped drones with infrared capabilities for night flying for rescue missions, use in war zones, etc. I mean very disturbing as part of the Minority Report trend here, but this marketing at least makes it seem like their heart’s in the right place.

42. Slingshot Aerospace: cool. Cuz testing in space is expensive to the extent that it kind of defeats the purpose of testing.
46. True Anomaly: assembly lines for satellites. Is the demand for satellites that high? (I have no idea.)
48. WeaveGrid: AI for electric vehicle power grid/demand predictions.

43. Surge AI: cool. AI to help with the “critical stage of reinforcement learning with human feedback (RLHF).” …I feel like I’d need to have a much more sophisticated, code-level understanding of the competition in this space to evaluate this. Plus, you know. Would want to know their DEI/bias policies.

44. Swarm Aero: drone swarms for the military. Cool concept, but…my brother Dan Quinn teaches fluid dynamics like this at the University of Virginia. Is this science at the point of commercialization? (Also yeah, creepy scary bomby swarms now. Cool. Cool cool cool.)

45. Synthesis - hmm. AI-powered learning originally designed for Elon Musk’s kids. Hm…what level of influence do I want Elon Musk to have on my (hypothetical future) kids’ curriculum…hm. I mean I guess it’s good they don’t include history or government in the curriculum they teach.

47. Vannevar Labs: Google Translate + ChatGPT synopsis for intelligence analysts. Seems useful assuming you can get the LLMs & training data (and security protocols) right. (Though…not sure who all uses air-gapping in this space, or how LLMs work with air-gapped computer processing power…)

49. Zanskar: geothermal mapping for power stuff.
50. Zipline: delivers blood & vaccines & healthcare supplies via drone, including in Africa, Utah, Arkansas, & North Carolina. Huh I guess Andreessen Horowitz considers some people’s health important to the US’s national interest.

3. Air Space Intelligence: isn’t main problem in air travel efficiency that most planes are equipped with old fashioned radar so planes have to travel over a system of towers?