Big Three AI Labs Sustain Exponential Curve, Delaying Open Model Shift

In large part that’s because it’s hard to invest in a world where the exponential continues and open models never close thr gap. The value js just chips, energy, data centers, and the labs themselves.

@emollick How do you even measure this as exponential? I perceive it as following diminishing returns within my domain for the last 5 months (which is software engineering). I basically have seen the same issues with marginal improvement. Not near as large of a change as the 5 prev months

@emollick It could be less of it bending in principle and more of governments treating models beyond a certain capability threshold as potentially dangerous strategic assets

@emollick Any thoughts about quantum computing replacing LLM?

@emollick Not happening unless we get really bad regulation.

@emollick this people aren’t accounting RSI in Big Three that ongoing right now.

@emollick Not gonna happen

@emollick Scaling laws for Big Three models suggest compute efficiency gains are being offset by massive parameter growth in next-gen architectures.

@emollick The U.S. administration has shown that it will ban any model that's become too capable, so the money is well invested.

@emollick It's interesting, not having been involved in much hiring, how easy it is to justify spending a lot more money for somewhat better performance.

@emollick One imagines that real assets will retain value? Land, timber, roads, railways.

Those four areas are all connected as parts of the same ecosystem. The labs create demand for data centers, which in turn need chips and energy. In past tech cycles, when the conversation starts narrowing down to where the real value lies, it usually happens later in the cycle rather than at the beginning..

@emollick If AI is so powerful now, why hasn’t anyone yet developed a truly usable continuous learning paradigm for LLMs with it?

@billtheinvestor @emollick Join Unity Markets Signals Trade on Telegram today and unlock profitable trading opportunities. Thank me later

@emollick The weird strategic implication is that “wait for costs to fall” may be the riskiest plan. If capability keeps moving, the durable advantage is getting the workflow/runtime muscle now, then swapping models as the curve changes.

@emollick I think you're naming the real squeeze. if open models keep pace, you're buying commodity hardware that depreciates plus the real estate under it. the moat isn't the model anymore, it's operational scale and who can afford to run the thing. 🔌

@emollick ngl this take feels like hopium with a roadmap attached

weirder more slowly is a nice thought but theres no off switch