As the Claude 3 family of models continue to impress, with new capabilities emerging as more users switch, pressure continues to mount on OpenAI to release GPT-5. This week Sam Altman has been busy hyping up the potential for the model, and some upcoming features (likely in the planning, autonomous agents and robotics spaces), but also dampening down timeline expectation.
Takeaways: The latest consensus is mid-year for GPT-5, and a major step up in capability, although what that really means will be a huge indicator of future AI trends. You can hear Altman’s interview on the Lex Fridman podcast here. Several quotes have emerged from Altman’s publicity activity this week that are worth considering:
- Firstly that ChatGPT to date has been a revolution in AI “expectation” but not in function… this seems fair in that much of what AI can achieve is still theoretical and there is a lot of integration to be done. But also suggests that the future capability steps will be more significant.
- He also suggested that companies should not be delaying innovation, and should learn and integrate now, so that they are ready for these significant future steps in capability as we near AGI. (Naturally we at ExoBrain would agree, but it also just seems like the most common-sense approach in an unpredictable world, which is about to be flooded with AI compute).