Chris Lengerich

30+ commit days - fun, but kinda fuzzy.

One of the more magical experiences I've had w/ AI has been live-coding Android apps on real debug devices (via OpenCode + GLM5.3-Flash). adb has the tooling to do file pushes, snapshots, UI presses, so it feels like a synchronous pair programmer. Minimal lag compared to other computer use agents.

30+ commit days - fun, but kinda fuzzy.

The fallacy of (transitive) order: The finite action space of contrastive decision-making (a vs. b) requires an agent to impose an order on the world (1,2,3). Yet we know from watching nature that games observed honestly are often non-transitive (rock-paper-scissors).

(this is also why I've been basically 2005-Amish (ie. froze most of my tech at 2005, when airgapped/offline/pull flows were normal, not always-online push culture)

The fallacy of (transitive) order: The finite action space of contrastive decision-making (a vs. b) requires an agent to impose an order on the world (1,2,3). Yet we know from watching nature that games observed honestly are often non-transitive (rock-paper-scissors).

(also cf our friendly Kelvin waves)

The fallacy of (transitive) order: The finite action space of contrastive decision-making (a vs. b) requires an agent to impose an order on the world (1,2,3). Yet we know from watching nature that games observed honestly are often non-transitive (rock-paper-scissors).

And perception tech that projects the wrong ideas about order onto a powerful manifold, invisibly, fits into the lower right quadrant of things to worry about - both powerful AND invisible (the other example being climate change).

The fallacy of (transitive) order: The finite action space of contrastive decision-making (a vs. b) requires an agent to impose an order on the world (1,2,3). Yet we know from watching nature that games observed honestly are often non-transitive (rock-paper-scissors).

But in reality, the operating system of US democracy 1.0 (selecting a transitive order from a non-transitive environment) was broken by social media (along with the complexity of abstraction layers) and it happened 10 years ago. Misdirected outrage is part of that problem, not a solution.

The fallacy of (transitive) order: The finite action space of contrastive decision-making (a vs. b) requires an agent to impose an order on the world (1,2,3). Yet we know from watching nature that games observed honestly are often non-transitive (rock-paper-scissors).

Even some in VC (which are supposed to be curious) lock into "best of Twitter". Order in that case ~= funding order, and if they can turn investment $ into ARR, it works. Still, many of my most consistent returns came from highly technical founders with minimal SM clout.

The fallacy of (transitive) order: The finite action space of contrastive decision-making (a vs. b) requires an agent to impose an order on the world (1,2,3). Yet we know from watching nature that games observed honestly are often non-transitive (rock-paper-scissors).

And yet, "the current thing" mechanics of social media select for ideas that reinforce the social media projection surface (AI is better or not than humans, for example).

The fallacy of (transitive) order: The finite action space of contrastive decision-making (a vs. b) requires an agent to impose an order on the world (1,2,3). Yet we know from watching nature that games observed honestly are often non-transitive (rock-paper-scissors).

Non-transitivity is especially true in high-dimensional spaces. Plenty of places for non-transitivity to hide. In those environments with a lot of corners on the gradient, shorter step lengths coupled with close observation often is better rewarded.

One of the more uncomfortable things with investing is that to make 100x, you have to be comfortable with some investments being down 50% a couple times (or worse, no updates).

AI bubble tracker (echoes of 2022): 1. Friends are getting $100M offers to stay at bigcos. 2. Landlords raising rent 20% or more. 3. No one - investor, researcher, founder - is incentivized to short it. 4. ~9 months of runway before many startups need to raise billions, or become walking dead.

The dollars are not the same - companies need to shovel in VC money every year, 4-year contracts are, well, contracts.

AI bubble tracker (echoes of 2022): 1. Friends are getting $100M offers to stay at bigcos. 2. Landlords raising rent 20% or more. 3. No one - investor, researcher, founder - is incentivized to short it. 4. ~9 months of runway before many startups need to raise billions, or become walking dead.

6. Some AI startups have healthy revenues, but not as many as you'd expect given valuations. For now, the music plays on, but how much of fast growth is just a fast business cycle (up and down) due to interconnected networks in 2026, not an indicator of long-term value?

AI bubble tracker (echoes of 2022): 1. Friends are getting $100M offers to stay at bigcos. 2. Landlords raising rent 20% or more. 3. No one - investor, researcher, founder - is incentivized to short it. 4. ~9 months of runway before many startups need to raise billions, or become walking dead.

It's not just NLP-based coding and writing more READMEs/verification documentation for your agents, what the software goal is itself is changing based on 10x faster coding.

The abundance problem isn't really a new thing, it's actually quite old (cf. asabiyyah cycle). Just that it's happening in intelligence now vs. other products in the past. And almost as a rule, societies have not managed it well.

RIP Scott, one of the brilliant. 68 is far too young, wish you could have seen how far we'll go.

xcancel.com/ScottAdamsSa...

Types of economic models: 1. Discriminative (distill a couple variables for a simple decision boundary, model-based) 2. Generative (distill a generative process for more complicated model, model-based)

3. Ecological (just describe the niches and complex feedback loops that emerge, without imposing a model, wolves of Yellowstone-ish, model-free).