Wharton professor Ethan Mollick argues the "capability overhang" from existing AI models will reshape society even if development stops

I know this is a common thing to mention in some corners of X, but there is still so much focus on future model development & innovation bringing transformation that people are really overlooking the degree to which the transformation is already baked in, no AGI needed.

@emollick I've felt this way since GPT-4, you were the only person I remember talking about this back then. The amount of change in the system is already enough to completely remake the world, and we are now moving faster than we ever have been.

@AndrewCurran_ @emollick

@emollick Agree 100%. People keep asking why we need so many DCs.... I say until I can call Walgreens Pharma and get some actually useful information without waiting for a human. How often are you interacting with AI as a daily consumer? It's still pretty rare...

@emollick We don’t understand what we didn’t evolve to understand. None are so blind as those who cannot see.

@LokiJulianus @emollick Yeah that's the stuff.

@emollick Agree, I see no sign that AI development will stop.

What is now non-zero probability is that access to future AI models will be embargod/prohibited by administrative fiat.

That was not something I considered a possibility a few weeks back...

@emollick Silver lining: If development proceeds fast enough, one way or another the social integration problem goes away?

@emollick @AndrewCurran_ Yup, we got slowed down, but it’s enough to build

@emollick agree

@emollick Every single day I discover new ways to use the current AI tools to do something cool, big or small. So much opportunity for transformation and innovation.

@emollick The focus so far has been on fields with quantitative benchmarks (coding, math, knowledge...). Real impact requires more than that: deep understanding of the field, reliability, deployment, adoption... Moving from invention to innovation takes years if not decades

@emollick True, but it will be challenging to overcome the resistance of people and bureaucracy. We need to do our best to use these new powers for good and offer solutions that benefit everyone.

@emollick 100%. Most of the economic benefits will come from scaffolding/harnassing around existing models. It’s interesting that the us labs have ceded this to open source mostly. (Most scaffolded usecases I am aware of need sonnet level intelligence at low prices).

It's more than I can imagine. The upcoming AI chips have performance multipliers that boggle the mind:

2018: Alessandro Volta chip (V100) - 1× 2020: Ampère chip (A100, A30) - ~3× 2022: Grace Hopper chip (H100, H200) - ~16× 2024: David Blackwell chip (B200, GB200) - ~160× 2026: Vera Rubin chip - ~400× 2027: Vera Rubin Ultra chip - ~800× 2028: Richard Feynman chip - my guess: 1,500–3,000×

@emollick the weird part is the models don't even need to get much better for a lot of people's habits to change completely.

@emollick and all value should flow from legacy companies to ai-native companies.

@emollick I think this is true too. Everything is changing even if the models stay exactly where they are. It's just a matter of how fast it disseminates.

@emollick There are unfortunately some signs that despite development continuing internally, the release of better models might be starting to slow soon and undergo more and more serious scrutiny

@emollick I disagree. AI may deliver major gains in specific, narrow domains, but today's models still lack the breadth, reliability, and autonomy needed to make large-scale changes to work and society inevitable.