Forgive me nostr, I have just learned about Gresham’s law and Thiers law; it’s all very clear now

“Employee productivity is sky high but company revenues (outside chips supply chain) are roughly stagnant.” This can easily be explained: Revenue is usually demand-constrained, not output-constrained, and Amdahl's law shows us that speeding up a non-bottleneck step doesn't speed up the system itself. It’s REALLY, REALLY hard to identify isolated tasks that directly block revenue for most companies. And even if they’re output constrained (rare), any new AI productivity gains are just competed away… For everyone involved, regardless of what type of constrained you are, net-new AI productivity gains typically show up as costs, not revenue. Token spending is a labor-to-vendor transfer: part of the payroll that would have grown now flows to labs and cloud providers. Andddd thats where the revenue is lol In other words, the productivity gain is being captured UPSTREAM rather than created. Where I don’t have the data but would like it (couldn’t find anything reputable + recent): gains should appear first in sectors where demand is elastic and output is the bottleneck and last where demand is fixed or capacity-limited by physical or regulatory constraints. Examples of such markets where latent demand is priced out by labor cost (Baumol sectors), and lower prices would unlock quantity rather than just cut cost: - Legal, accounting, and compliance for small businesses and individuals - Tutoring and coaching - Health/diagnostic-adjacent - Bespoke software Even there, the gain accrues mostly as consumer surplus, with lower prices and more quantity. Recapping bc it’s late and I’m tired and idk if I went off track: very few industries are output-constrained in a way that converts productivity into revenue. AI's effect is mostly deflationary at the task level, and revenue growth requires new demand, which tends to come from new products rather than cheaper existing ones.

Il faut cultiver notre jardin

Haven’t been online much these days, but something I firmly stand by, but see very little discussion of, inside/outside of nostr, is bitcoin used as a human verification layer. It’s the best antidote to slop cannons i can think of. Nothing novel here, just reiterating what we all know, but needs to be repeated: Pre-AI, digital information output was a function of energy involving humans. Post-AI, digital information output is a function of energy that does not involve humans. It’s pure energy/compute. As tokens/inference/compute gets cheaper, slop will only get worse! If internet activity was paired with micro transactions, the costs would be ~0 to the avg person. Bad actors wouldn’t be so fortunate… Bitcoin is the best form of “digital proof of energy expenditure” we have, which becomes expensive at scale to any bad actor deploying agent swarms. The incentive to pump the internet full of slop, propaganda, etc at a meaningful enough scale is low to an individual, leaving these tasks to nations/states/regimes/organizations who have the financial means. This could make it very easy to spot coordinated efforts as well, or at least REDUCE such efforts to whatever is humanly possible (Russia/china have been doing this already for years). Again, my focus here is on human verification, not verifying how “genuine” the activity is. Humans make slop too, just in lower amounts. Perhaps we don’t even need to “force” these transactions, and they could be made voluntarily, attached to our activity. Proof-of-work literally becomes proof of work. I’m proving I did this work, I’m proving I produced this information. Not including such transactions in digital activity could become a negative signal, a “bad actor” signal, or simply just low-status, which could be the most impactful of all. (Would this be the first time a btc user would ever be considered high status to the general public? Lol, you get the point) But yeah, we sure don’t need a “world ID” where a private company scans my retinas, that for sure I know. Micropayments cut the low-margin slop (not all!) and add a traceable cost layer. They do NOT stop agent-driven persuasion/scams/ etc (those have real ROI here). Who’s the recipient? No idea, maybe this only works on nostr. I need to think more about that - very important - part. But I think this is a good discussion nonetheless

Trying to make sense of this chaos. Staying offline for the time being.

Max Sheler put into words a feeling I’ve had about the world, but could never quite explain myself. When you understand his description of Ressentiment, you see it everywhere.