re anthropic's ipo:
my biggest question is how should we think about gross margin for model companies?
from an accounting perspective, traditional COGS doesn't work anymore. anthropic isn't selling intelligence, it's selling intelligence gap over free intelligence (i.e. chinese models)
accordingly, their true COGS isn't just the cost the serve their model, but the cost to maintain that gap. their 85% gross margin figure fits nicely into today's GAAP rules but practically speaking should include *total* training cost - not just cost to serve, but also other compute, data, maybe labor
that said, the base model intelligence gap is probably larger than people think. anthropic and openai's benchmarks are the only faithful benchmarks out there and everyone else is more behind than the benchmarks imply
3
70
we've achieved something close to agi, and unemployment rate is... down?
some theories:
1 enterprises are slow to adopt things
2 people hire and fire based on economic vibes instead of operational need (and right now the economy is good)
3 people are much lazier at work than it seems (automating takes upfront work)
what am I missing?
40
before Uber went public, we took for granted how cheap and fast it was. after they went public, prices shot up and service quality went down
Anthropic's IPO is a few months away. everyone who's complaining about Claude limits today is about to feel what Anthropic chasing profitability actually looks like (and it won't be pretty)
1
144
openai is doing something really interesting with luna that people aren't talking about
luna is a gpt-5.5 tier model priced at $0.10 input. it's both cheaper and more performant than any chinese model
openai's strategy must have shifted from being frontier-only (i.e. competing with anthropic) to also competing on value (i.e. against chinese models)
the prevailing thought has been that US models will differentiate through superior IP, at the frontier, and capture $$$ margins, while chinese models compete on the low end and rely on scale
so luna's pricing strategy comes at a shock, and it makes me wonder what openai's underlying strategy is. do they have some thesis on how to win on the low end in the long term? or are they temporarily selling at a loss to slow down chinese labs?
that said, chinese labs seem to be more effective at "marketing" their models than openai has with luna. if luna benchmaxxed the way chinese models do, I doubt these chinese model launches would be capturing many headlines at all
2
96
Charities that are measurable + cost ineffective have a strong incentive to stop being measurable
1
3
306
Philip of Macedon had a brilliantly simple insight that was obvious in hindsight: make longer spears. His 16-foot sarissa hard countered the traditional 8-foot Greek spears, gaining military dominance overnight.
Similarly, the US military's carrier-driven strategy has ruled for decades, and I wonder -- what'll be the hard counter to our strategy, and will we adapt quickly enough to stop it?
1
3
308
American Airlines to brand agency: we'll pay you $X million per word you come up with
Brand agency: introducing the American Airlines AAdvantage Barclays Aviator Red World Elite Mastercard
1
2
325
AI agent idea: ElonBot
Asks “What did you get done this week”
Asks follow-up questions based on response
Integrates with your HRIS and auto-fires people
You can't control who it fires
3
151
Open offer #1
Will pay name change fees if you change your name to Ptolemy XVI
2
3
144
My 80-20 list on how to lower your LDL:
- lots of psyllium husk*
- plant sterols* (~2g/day)
- berberine + fish oil*
- daily movement (walk/workout)
- cut fried food & processed junk
- 16/8 fasting
*if you take these all at once in the morning, it shouldn't cost you any time/energy











