Agree they'll get lower in variance + smarter, but I also sometimes don't need that much smarts. Models like GPT 4.1 Mini/Nano are quite useful for small tasks where I don't want reasoning.
Some good predictions here. OpenAI’s productisation strategy signals they’re doubling down on eyeballs and understanding the wetwear behind them. Memory, search, and detailed buyer profiling will be packaged as seamless social experiences rather than developer tools. Some existing big tech lunch will be eaten.
I expect moves in the next 12 months to socialise interactions: integrated forums, shared chats, AI-mediated social functions (eg model mediated dispute resolution, meeting facilitation…). This prioritises consumer stickiness over enterprise flexibility. Are OpenAi betting that owning the social layer compensates for losing developers who want compositional control? Versus Anthropic?
This might create a bifurcated market. Consumer platforms (old Google, old Meta) will face genuine disruption from OpenAI’s integrated features. But enterprise/developer segments may fragment differently. As noted, OpenAI’s abstraction layers push application builders toward open models precisely because they need “just the model.”
Open model economics will require revenue to come from providing : specialised inference services, enterprise APIs, compliance-friendly alternatives.
If product integration pushes away enterprise customers, open alternatives may gain faster adoption. Especially if decent small open models can make their way onto phones and laptops.
Lunch will be eaten selectively - creating new opportunities in segments OpenAI implicitly abandons.
Good post. Although I think reasoning models are going to get lower on variance so not sure it’ll be long before that point changes.
More investment is also just going into reasoning models too, and therefore will be smarter.
Agree they'll get lower in variance + smarter, but I also sometimes don't need that much smarts. Models like GPT 4.1 Mini/Nano are quite useful for small tasks where I don't want reasoning.
Some good predictions here. OpenAI’s productisation strategy signals they’re doubling down on eyeballs and understanding the wetwear behind them. Memory, search, and detailed buyer profiling will be packaged as seamless social experiences rather than developer tools. Some existing big tech lunch will be eaten.
I expect moves in the next 12 months to socialise interactions: integrated forums, shared chats, AI-mediated social functions (eg model mediated dispute resolution, meeting facilitation…). This prioritises consumer stickiness over enterprise flexibility. Are OpenAi betting that owning the social layer compensates for losing developers who want compositional control? Versus Anthropic?
This might create a bifurcated market. Consumer platforms (old Google, old Meta) will face genuine disruption from OpenAI’s integrated features. But enterprise/developer segments may fragment differently. As noted, OpenAI’s abstraction layers push application builders toward open models precisely because they need “just the model.”
Open model economics will require revenue to come from providing : specialised inference services, enterprise APIs, compliance-friendly alternatives.
If product integration pushes away enterprise customers, open alternatives may gain faster adoption. Especially if decent small open models can make their way onto phones and laptops.
Lunch will be eaten selectively - creating new opportunities in segments OpenAI implicitly abandons.