The Ledger
Every statement that passed quotation and attribution checks. Mix any filter with any other: certainty 1/5, debate potential 5/5, or both at once.
why aren't all 46 resolved? a statement only gets an assessment when the public record can support or contradict it. opinions and what-ifs never can, and 1 checkable ones are still open, waiting for their date. predictions held up or didn't; assertions are supported or contradicted. on every card: ▮▮▮▮▮ certainty · ▮▮▮▮▮ debate potential. speakers are clickable
Genuine AGI safety concerns are 10 to 20 years away
“I think this is likely what we're doing with AI and AI safety and stuff like that, that we could foresee in 1020 years AGI, where there's actual safety issues, but we are so far from that today. That's, I think, actually getting close to the peak of inflated e…”
Kolicich: Expect an AI crash and market retrenchment within two years
“I believe in two years where everyone's gonna say, oh, you know, we thought it would solve everything and it's all a very limited number of use cases. It's complete hype. And, you know, we'll probably see a bit of a crash and we'll see a retrenchment in AI.”
Kolicich: Open-source AI iteration speed will surpass closed source
“Cause you'll see the case of iteration at open source is going to eclipse close source.”
Kolicich predicts wave of VC firings over failure to adapt to AI
“And so I think you'll see over the next 10 years, a lot of retirements, a lot of firings, perhaps because people won't make that transition,”
Competitors will catch up to OpenAI's o1 reasoning model within six months
“O1 just came out. We haven't fully benchmarked on our side yet. It's definitely an improvement. And so open AI is ahead of everybody else. It's an improvement in reasoning which is good. We can talk about that later on, but I expect people will catch up within…”
Incumbent software systems will capture AI value through existing data gravity
“So I think if that happens and that's what the world looks like, then a LLM looks kind of like a query processor in a database. You give it context, it does its work, you get the answer out. But the point of leverage is the place with, at which the context Liv…”
Nvidia is not trading at dot-com bubble valuations on an earnings basis
“There's a lot of AI hype, but if you look at companies like Nvidia, they're actually not that expensive on an earnings basis. Like it's not, we're not in bubble territory, you know, Cisco, 2000 timeframe.”
AI startups should aggressively raise capital every three to four months
“If you are in AI or AI adjacent, you should be raising every three to four months, in my opinion, and you should be doing it aggressively.”
Early-stage 100x AI multiples are unsustainable artifacts of mega-fund capital oversupply
“There are rounds happening that are 102 hundred X revenue in the early stage markets, and even there you have to assume, you know, you're gonna have 10 X Multiple compression in the later stage rounds. So I don't think those are sustainable for any period of t…”
Venture capitalists voluntarily returning fee-bearing capital will never become a trend
“Money managers returning fee-bearing assets will never be a trend, in my opinion.”
The volume of venture-scale startup opportunities has declined 60% since 2021
“My estimate was, I think the amount of venture scale opportunities has gone down 60, 70%.”
LLM scaling exhibits decreasing returns, requiring 10x compute for sublinear improvements
“If you look at it today, though, we are, the exponential in LLM performance is so far working against us that you need Order of magnitude increases in compute, 10 X increase in compute to have relatively sublinear increases in performance. So you're at decreas…”
AI will hit a trough of disillusionment as revenue lags infrastructure buildout
“And that kind of starts to tell you that we make it close to the trough of disillusionment at some point where all this, you know, the infrastructure that was built and the idea is build it and they will come. You know, the people actually don't come. The reve…”
AI acqui-hires are specifically designed to bypass Lina Khan's antitrust scrutiny
“On the M&A side, I do think it has a lot to do with regulation and Lena Kahn, you know, threatening to block most mergers and acquisitions, and that's why you see these acqui-hire inflection character AI adept type acquisitions where it seems like they're doin…”
The capped defense budget will yield only tens of new primes
“I think we probably think it's a little overhyped in general, where we think it's real for sure, but it's not going to be an opportunity the size of software or AI or something like that. The defense budget is naturally limited as a percentage of the economy. …”
The volume of pre-IPO leadership departures at OpenAI is historically unprecedented
“I don't think we've ever seen pre-IPO so many leaders leave the best, like, crown jewel company in Silicon Valley. It's common, like, you IPO, everyone gets rich and they leave. Okay, that's normal. But before IPO, right in its, like, heyday, you don't see the…”
Kolicich: Venture capital will shrink over five years unless AI takes off
“I don't think it's gonna, you know, the space is gonna crash or everybody leave overnight, but I think just slowly over the next five years, unless AI really takes off, you'll see a general shrinking of venture capital, I think”
Kolicich: Big Tech's Low AI Pricing Will Hurt Competing Startups
“Big tech is doing a good job of anchoring the price of all these features pretty low, like co-pilots, 20 dollars a month per developer. It's incredibly cheap relative to the value that it does add. And that really anchors, you know, value in that market pretty…”
Kolicich: Big Tech Subsidizing Open Source AI Is Hard for Startups
“They're really subsidizing the space, a lot of big tech, which I think is positive for open source and positive for users, hard for startups.”
Kolicich: LLMs won't collapse vertical software pricing to zero
“And just the process of defining that software system such that an LLM can generate it would cost you hundreds of thousands of dollars. And so even if it were free to generate a new software system, you would probably still use one of the incumbents as long as…”
Kolicich: Open-source AI models lag proprietary models by one year
“Some people have said it's six months from state of the art. I think it's more like a year away from the state of the art, but these models, those models that keep coming out are getting better and better and better.”
Mega-check foundation model investors have largely avoided write-offs via returned capital
“If you're an AI investor doing very big checks and foundation model up, foundation model layer up, you haven't really had many write offs. You've in your losers, or I wouldn't say losers in the companies that haven't broken out, you've made your money back.”
The future of software is human-in-the-loop symbiosis, not fully autonomous AGI
“And so that would mean the future of software looks more like man-machine symbiosis. Then it does AGI, you know, the computer does everything for you.”
Aggregate AI hardware capex vastly exceeds actual AI deployment revenue
“Aggregate capex on AI hardware. It's like vastly out exceeding revenue generated by AI companies in the actual deployment of AI.”
Cloud giants must host foundation models even if they become low-margin commodities
“Microsoft doesn't care if the margins and foundation models are 80% or 30%. They can't lose that layer, right? Even if it's a commodity, it needs to be a commodity on Azure. Same thing with Amazon. Needs to be a commodity on AWS.”
Traditional defense primes excel at explosives but fail at software and AI
“Defense primes are good at aeronautics. They're good at explosives. You know, I tend to, I always say, like, they're good at boom. Don't compete on boom. What they're bad at generally is software, is AI, is autonomy.”
8VC partners initially feared office protests and reputational ruin for funding Anduril
“We literally had conversations internal at AVC with our partners where we were worried about our offices getting protested. And we gave a like small chance that, you know, actually maybe our reputation is completely ruined because at that time, everyone was ag…”
Successful defense tech startups are rarely founded by industry outsiders
“Most of these companies I mentioned that we invested in are started by people who were actually at one of these successful companies, Endural, Palantir, or SpaceX before, or who were in military service themselves, and they know sort of the landscape. It's ver…”
Software will shift from manual workflows to platforms supervising AI agents
“This sort of like supervisor worker type paradigm where it's not workflow software, it's supervisor software where you supervise the agents.”
Kolicich: Enterprise 1.0 software like Salesforce is just database wrappers
“If you look at enterprise software, we call it one. And they're very simple systems. In many ways, they're wrappers over a database. If you look at Salesforce, Salesforce doesn't help you make a decision, really. Maybe there's only for the product I do today, …”
Kolicich: Crossover funds exerted a 20x leverage effect on venture cap tables
“When A tiny girl or somebody else deploys money onto a cap table. They might say own five percent of that cap table, and they'll have, their dollars will have a leverage effect on the ecosystem as a whole. Because while they buy five percent of the company, th…”
Kolicich: Perplexity Synthesizes Information Better Than Google Does
“I think perplexity is actually a very good product that, you know, does a better job on synthesizing information than even Google does.”
Kolicich: GitHub Copilot Operates Like Autocomplete, Not Fundamental
“Co-pilot operates Much more like an auto-complete. So if you're writing an article on email or in Word, you automatically notice a sentence. It's pretty good. It does save you time, but it's not fundamental.”
Kolicich: 8VC is not interested in investing in model inference hosting startups
“Most major inference players, for example, i.e. People who help you host open source models or train new open source models are growing incredibly quickly. We're not particularly interested in investing in that. Aaron at the stack.”
8VC is buying traditional services firms to automate them into software businesses
“We're very excited about AI enabled services. And so can you pair the traditional private equity rollup model with sort of a software development team? Buy traditional services businesses, use an engineering team to automate a lot of the work, and turn what us…”
AI startups with $5M revenue are raising at $700M valuations
“Now we're seeing companies, you know, five million revenue in AI revenue, and they're raising at five, six, seven hundred million valuation. That's price to perfection, right? We're talking about a hundred X multiple sometimes in these cases.”
Fewer venture capital firms will raise opportunity or growth funds going forward
“I do think fewer people will raise opportunity funds or growth funds. I think that is likely.”
Institutional buyers will emerge for $500M single-position venture fund secondaries
“I think there's a big opportunity for someone to come in and say, Hey, I'll take this position. I know it's good. I'll, you know, buy half a billion dollars worth of it at some slight discount. That's a gap that doesn't exist in the market today, but I think s…”
Kolicich: AI will automate back-office paper pushing over software development
“I think you'll see a lot more deployment in back office jobs at industrials or any vertical Industry, you know, I mean, even if you look at healthcare, on the RCO side, there's like faxes and people faxing papers all over, like, the world still runs a lot of p…”
Kolicich: Software incumbents hold the advantage over thin AI wrapper startups
“Incumbents have the advantage. Where in most cases, you are a thin layer on top of open AI and that's okay. And in that case, if you have distribution to users and you make that value transparent to them, you already hold their data. That's going to be a great…”
Kolicich: Google's Defining AI Feature Is Reaction Speed Over Initial Errors
“And so just the defining feature will be how quickly Google can react versus, you know, mistakes they made at the beginning.”
Kolicich: More VCs are active today than in 2019
“I think the trend of headcount is lowered, but there's still way more venture calculus today than there were in 2019.”
Kolicich: 2024 Capital Deployment and Valuations Surpassing Early 2023
“Like certainly, 20, 24 already has been much more aggressive in terms of. Valuations and capital deployment than the beginning of 23.”
Public software multiples have normalized to roughly 6.5x forward revenue
“I publish our internal software benchmark and you see forward revenue on a software company six and a half times. That's about where it was for the last I don't know before five years ago, the last decade. So, you know, five years back, it's the 10 year averag…”
Defense startups can secure up to $30M early prototyping contracts from DIU
“You can get 10, 20, thirty million dollar contracts from the DIU very early stages to create prototypes to compete for some program”
Kolicich: 8VC allocates 50% to software, 20% biotech, 10% defense
“Half of what we do is software between vertical, horizontal, infrastructure, AI infrastructure, data infrastructure. 20% ends up being biotechnology because we think there's key innovations that are a lot enabled by computation, but a lot, you know, with gener…”