GPT-4 is an upgrade over a search engine (on the 2010 internet, which was much easier to search than the internet today) and there is certainly opportunity in using it to chain together complex programming tasks fully programmatically, but we are stuck right on the cliff of a truly reliable and generalizable AI, and it isn't clear that we can just train an AI with the next generation of GPU's and more data and bridge that gap. Most of the truly high value add activities (fully autonomous programs and AGI that creates novel inventions) rely on a model more intelligent and more consistent than the current state of the art. So I think most of the valuation is in speculative potential.