AI models' supersonic progress, in 120 slides
Add Axios as your preferred source to
see more of our stories on Google.
Illustration: Sarah Grillo/Axios
Generative AI has entered the mainstream faster than any previous new technology. But the tech industry hasn't yet figured out the best ways to build AI products — and fierce competition, along with rapid advances, means nobody stays at the top of the heap for long.
Those are some of the top-line findings of a new report on the state of AI foundation models from Innovation Endeavors, the venture capital firm co-founded in 2010 by former Google CEO and chairman Eric Schmidt.
By the numbers: The report (video), by Innovation Endeavors partner Davis Treybig, says that 1 in 8 workers globally now uses AI on a monthly basis. 90% of that growth took place in the last six months.
- New models regularly topple technical benchmarks, but the cost of training them is also ballooning.
- "The average duration of human task a model can reliably do is doubling every seven months," per the report.
Stunning stat: New models typically spend three weeks at the top of the usage charts and then drop off as newcomers emerge and open-source rivals absorb and commoditize their advances.
- Frontier models "depreciate on a 6–12 month timescale," the report says.
AI is "fundamentally disrupting" software development and collapsing the distinctions between programming, product management and design, the report also finds.
- "Therapy, life organization and learning" lead the list of general-interest AI use cases.
What they're saying: "Foundation models are no longer experimental, they're essential infrastructure," Schmidt said in a statement to Axios. "Every leader, entrepreneur, and policymaker must quickly get up to speed on the seismic shifts underway."
