The
Usefulness of A.I.
Our conclusion is that A.I. is very useful indeed, provided
you ask the right questions with exercised judgment. We started out asking
Gemini on 10/8/26 four questions,. The first question,
to start off, was fairly technical, asking whether the
data that was input to the pattern-searching LLM model was of sufficient
quality to enable generalization. Here were the questions:
1) Is A.I. in a bubble, given the stringent requirements
of FAIR data?
2) Analyze the A.I. capex spending versus revenue
monetization gap across major tech companies and hyperscalers.
3) So, all this implies that it will be very hard to
make money on A.I.?
4) So, the criteria for profitability are the
traditional software advantages?
Click here
to see the detail. We have a small position in Symbiotic, which is automating
the Walmart warehouses. A 9/9/26
WSJ article indicates, not so easy.
Here is
Gemini’s conclusion:
“It is not hard to make money with AI-it is hard to
make pure, standalone software margin by selling raw, uncustomized
AI compute. As raw foundation models become commoditized, market value migrates
away from brute-force foundation models and towards hardware builders, legacy
platform with existing distribution and specialized vertical solutions with
proprietary data moats. While the fundamental drivers of tech profitability
remain intact (distribution, switching costs, moats, pricing power), AI changes
how those advantages are built and maintained. Traditional software rules still
apply, but they operate with lower gross margins, faster commoditization, and
higher ongoing costs.
AI hasn’t rendered traditional business economics
obsolete: it has heightened them. Winning in AI software requires combining
traditional distribution and workflow lock-in with a modern approach to data
moats and compute cost management.”
A very important value investor maxim is:
Do not invest in something you don’t understand. We
now think we sufficiently understand A.I. not to invest in it through the
S&P 500, yet.