More on AI
Over the past week I’ve written quite a bit about AI’s end-market.
My view is that the forthcoming AI slowdown that will hit the model/harness builders won’t hit NVIDIA at the same time because as business slows for Anthropic and OpenAI, both will continue to invest heavily in their core models.
Both Anthropic and OpenAI will work diligently to grow their model context windows and to expand model memory. Neither context nor memory is close to an “end-state”.
Reasoning has room to evolve under the current model construct, but I’m not sure how much innovation will occur in the near-term as the playbook seems to be to throw more compute at everything rather than engineer new approaches to model learning and inference.
Perhaps Andrej Karpathy having joined Anthropic this week will help accelerate Anthropic’s R&D cycle.
Tonight’s NVIDIA EPS call livestream with sentiment analysis is below. It gives you a flavor for what we are building. This product extension will be production-ready later this summer.
Too many elements are scored “neutral” at the moment. We’ll enhance the sentiment analysis with more training (Claude powers the sentiment analysis), make the scroll more legible, and will roll this out as part of Kilby with no incremental price markup in Q3 (Kilby is a 25% markup on the list price of the underlying Claude models - Sonnet 4.6, Opus 4.6 and Opus 4.7).
We will host more of these live streams during the June Q earnings calls - some with my live commentary. Tonight’s was more a product test to be honest. I like to show people what we are building.