🍃 Rubicon Organics ($ROMJ.V) - new write-up
🔗 https://t.co/6EgjgwQhjo
Canadian cannabis isn't a pure commodity market. Under the Cannabis Act, brand equity is harder to build but more durable once established. Yet, markets are pricing this premium brand at commodity multiples.
0.5x EV/Sales. 20%+ growth. ~30% incremental ROIC
#Investing #FinTwitt
The entire moral architecture of this civilization has been rewired to detect the smallest linguistic infraction, while stepping over the bleeding body on the sidewalk
I think this idea is genius. Have you ever met a woman who knows what she wants? Leave it to the authoritative algorithm, marketed by her favorite influencers to figure it out for her.
Population boom imminent.
Wondering if you have any data on implied compute supply coming online each year. Given the relative performance gains of each generation, supply comes on multiplicatively, yet I haven't run the data on what this % is; it would give the required demand pull for gpu/hour rates to remain.
@michaeljburry 'Compute shortages' against frontier getting cheaper; empirical evidence from model routers of demand elasticity in price alone does not exist; all usage is a function of model intelligence benchmarking. Thus, real returns lag & compute is overbuilt.
How is this bullish compute? You're claiming demand for tokens is elastic; run data on this. Price has ~no explanatory power for token volume once you control for capability and time. There is no evidence demand is elastic in price at spot, under which your thesis implies routing to smaller models is bearish compute spend.
@bendee983 no it doesn't lol it just needs to stop subsidizing labs & capture higher rent against price-insensitive token demand with other hypers. Microsoft making an LLM would be a literal burning $, kind of like why markets hate $META.
@PresentWitness_ It's not even that. Compute costs are structurally declining, so you can have expanded token demand without greater relative capex intensity. Hyper capex is modelled wrong & assumes elastic AI demand.
Economic outcomes need to decouple from minimizing loss to the natural entropy of language. The payoff that would justify the disruption has to come from an axis orthogonal to language modelling (e.g. capability grounded in acting and being corrected by the world, not by text likelihood). The agentic boom is an example of this; without it, use stays bounded by what next-token prediction can express, which is not a perfect substitute for intelligent labor, and has diminishing returns by construct.
@drgurner Agreed. This is the first step to submitting to something you believe is greater than the self, which is what I believe to be a pre-requisite to happiness. 'Who you are' is a decision of will/soul, not of the self's emotions.