I'm rooting for open source
Industrial policy, investor updates, my new CRM, Pangram
Open Models
Oligopoly is bad for startups. If intelligence is expensive and the returns/margin accrue to the foundation model labs then we will basically see the current productive base accelerated but not transformed, which is good but only as a difference of degree.
But if intelligence is abundant and free, we can see a difference in kind - the most interesting AI businesses are predicated on some combination of model diversity, local models, free intelligence, etc.
Open models are better for the broader (more decentralized/long tail) innovation ecosystem and especially for the app layer. And the vast majority of companies are “app layer” in that they consume and use inference, intelligence, (language) models rather than develop models or manage infrastructure.
The pro-market position is obviously to encourage competition on price and quality, not regime coziness.
America should have an open source model ecosystem and the returns on intelligence should go to its consumers/users, not just its creators.
Open models are only decelerationist (discouraging investment and recursive improvement) insofar as the leading labs have structured/capitalized themselves on the basis of oligopoly and protectionism. A fiercely/maximally competitive market for intelligence is better than the captive world we’re otherwise moving to.
The most pro-social, pro-innovation, pro-market outcome is for model development and infrastructure to be a fiercely competitive commodity business not a (gross) margin rich oligopoly propped up by government mandates as quasi-GSEs.
Lots of the biggest VCs will reflexively be open models them because 1) Trump is against them 2) the American foundation model labs are against them - most investors reflexively kowtow to the powerful/root and side with winners. To their credit, a16z seems to be saying the right thing here.
The best version of venture should be an index of the entrepreneurial innovation economy, not civilization clash of kings.
Oligopoly is also bad for our politics/political economy. We are seeing state capture by the foundational models and business - a new unitary body at the further detriment of our democratic market system.
If our labs become state backed and controlled they’ll no longer really be “American” in any meaningful sense - and neither will we.
The worst case scenario of a ban on open source (chinese) models is that LLMs/intelligence winds up like energy. As the rest of the world moves to the electric stack and lithium ion supply chain, America doubles down on oil, losing an opportunity for global leadership to enrich regime-proximate-toadies. We’re going to fall behind on energy abundance, fail to develop our own critical industries/supply chain, and ultimately wind up losing soft power to China.
That can’t happen in AI.
I made a new CRM
I spent the last week and ≈$2k building a pretty slick replacement for a $10k SaaS product with Fable while consolidating together a bunch of vibe coded side projects that I’d built to supplement the SaaS we’re now churning from. All told, I’m gonna end up:
saving a few shekels/year (savings TBD because there’s now variable spend)
using a product much more tailored to what I actually need with flexibility to change it
Will I regret this? Maybe! I definitely wouldn’t want to have to support multiple users or try to commercialize what I’ve done.
Nevertheless, this was a lot of fun and I might write more about it since I’m using a ton of different services and tools.
Nathan Made a Crime Simulator
He started working on this at Slop Con and am glad he picked it back up at Slow. He’s been interested in the idea of “performative software”
Not all expressions have to necessarily be intentional, because personality is not. My point is that software has finally been added to the list of mediums for us to use. So create more performative software, and perhaps you’ll figure out what it meant later (if ever). The worst thing that happens is more people catch on.
Fun ideas - funny execution. Let’s see where it goes.
Why you should (maybe) send investor updates
If you want my opinion, advice, or help (and you may not!) I need context. Context is very hard to acquire on a point in time basis. I need time series data (quantitative and qualitative) to have a opinion beyond “Yoni’s general frameworks and shit-shooting”. This goes for all your investors too, obviously.
Investor updates are obviously useful for pushing to the top of mind/arming me to advocate for you whether that’s to downstream capital or to potential hires (and again you may be justified in not caring).
You can preempt lots of anxious and irregular but frequent Q&A by sending information. If you have a cap table to manage, you’d much rather send one email than do lots of 30 minute meetings just to transmit info.
Finally, investor updates are a good way to measure for yourself if you’re closing your books well, on top of your numbers, and hitting your plan. You may know your numbers and have that muscle already (great, in which case sending an update is easy) or not care to develop it (bad, in which sending an update is not necessary because your company probably won’t work).
If you are gonna send updates, here’s how to do it well:
Don’t do a whole long thing. Short is good. Timebox yourself on how long you spend writing If it’s really hard to get your numbers together that’s a separate yellow blinking light which itself is worth reporting.
Do it consistently. Monthly is better than quarterly but weekly is too much without good reason.
Numbers are better than words wherever possible.
Don’t market to your insiders. Things will be messy and hard. Don’t self flagellate but be honest.
You’re allowed to kick people off the updates if they are annoying and create more work for you with follow up questions or if you suspect they’re not good stewards of the information.
Elsewhere
Phoebe acquired Alden’s assets and customers. This is what it looks like when a winner starts emerging in a category.
There is a real/seeming dearth of recruiting firms focused on non-technical hires, especially slop cannons (high tool use generalists) - or at least I don’t really know them. I’ve had multiple companies ask me for help in finding these people. Send me recs and/or start a boutique.
Pangram is integrating with Substack for AI detection. There’s a great opportunity to create an “AI safety and research lab” that builds self improving bot-detection-defeating-agents in order to protect us from bots which defeat bot detectors. And it seems like an obvious objective benchmark to hillclimb for anyone willing to spend some money on Pangram tokens and inference. I’m mostly kidding but also I would fund this.