[haswell]

Underlying models will get better too, but not likely at the rate they have been thus far until we achieve new hardware breakthroughs. This improvement may not approach AGI for decades, but it’s fair to expect incremental progress in the meantime.

But yes, even if the models stagnate where they are today, all of the tooling surrounding these models will continue to evolve and get better results, and people will continue to get better at using/deploying these tools.