"Everybody would say yes, but we kept moving the goalposts," Ghodsi said in the discussion, which was published Tuesday."
""Since we kind of achieved it, let's come up with something even bigger," he added, referring to the push toward superintelligence - AI that can reason far smarter than humans."
"AGI "has everything we need to be able to automate and build the agents," he said. "We just need to do the boring work," he added."
""It's getting harder and harder to get value out of the next pre-trained giant model," he said."
AI chatbots now meet a decade-old definition of artificial general intelligence capable of human-like reasoning. The pursuit of superintelligence moved goalposts beyond practical capability and remains far from achievable with current techniques. Building systems that outsmart the world’s brightest minds is unnecessary for most company needs. Existing AGI capabilities suffice to automate workflows and create agentic systems, with the remaining effort focused on engineering and integration. The recent era of rapid gains from model scaling has slowed, and extracting incremental value from ever-larger pre-trained models is becoming increasingly difficult.
Read at Business Insider
Unable to calculate read time
Collection
[
|
...
]