Ali Ghodsi said Royal Bank of Canada built agents with Databricks that begin working as soon as an earnings report is released.
“Royal Bank of Canada, uh, built agents with us that basically take as soon as an earnings report comes out.”
Person
Ali Ghodsi
1 episodes · 18 quoted claims
Ghodsi said the Royal Bank of Canada agent can publish an equity research report 15 minutes after an earnings call, compared with an industry standard of two hours.
“And it can get the equity, uh, report out in 15 min from the earnings call. Industry standard is 2 h.”
Ghodsi said Databricks customer Merck created a model called Teddy, which stands for transformer enabled drug discovery.
“customer Merck, uh, that in the life science space created a model called Teddy. Teddy stands for transformer enabled drug discovery.”
Ghodsi described the Teddy model as a transformer that predicts what is missing from a genome and learns the gene regulatory network, which he said helps with drug discovery.
“it instead can figure out which genome is missing if you remove a genome. So it really understands the gene regulatory network”
Ghodsi said 7-Eleven uses agents to automate its marketing stack, including audience segmentation and campaign content.
“seveneleven agents that completely, uh, automate the marketing stack.”
Ali Ghodsi said he came to the United States in 2009 and worked in the AMP Lab at UC Berkeley.
“I came to United States in 2009 at UC Berkeley, not far away from here. And, uh, I was in a AI lab, it was called AMP lab”
Ghodsi said robotic process automation (RPA) was rule-based and never learned, so it had to be given exact rules and rewritten when it made mistakes.
“It wasn't learning ever, so there was like zero learning. It was like, you tell it exactly here are the rules, and if it got something wrong”
Ghodsi said the frontier labs' superintelligence quest rests on the scaling-laws idea that whoever has the most GPUs and data will win.
“It comes from the scaling laws mentality, which is whoever has the most GPUs and the most data, uh, is gonna win the quest for superintelligence.”
Ghodsi named Rich Sutton, creator of reinforcement learning, and Yann LeCun as leading researchers who believe the superintelligence approach is wrong.
“Rich Sutton, who created reinforcement learning, which, you know, a lot of this stuff is built on. You have Yann LeCun, who's one of the three founding fathers”
Ghodsi said the skeptical researchers in his second camp believe superintelligence is about 20 years away.
“Those guys, by the way, they say it's 20 years out, so they're saying, hey, it's a physics problem and it's gonna take 20 years to get there.”
Ali Ghodsi argued that artificial general intelligence already exists, saying the goalposts for its definition have moved.
“I think we have AGI, I think we have artificial general intelligence, we really have it, we absolutely have it.”
Ghodsi used a gas-station analogy to argue that large language models are interchangeable commodities.
“when I took took econ classes, commodity was when it's interchangeable. Like you can get gas from this gas station, or you can get gas from that gas station”
Ghodsi said Amazon.com and Google already existed in 1998, using them to argue that today's AI infrastructure companies can survive new app waves.
“there was actually an Amazon.com already in 98, there was already, uh, Google actually existed already in 98 and so on.”
Ghodsi said Cisco, an early internet infrastructure company, is still around and worth roughly $300 billion.
“Cisco, by the way, is still around, and it's, you know, only 300 billion dollar company or something like that, right?”
Ghodsi said Zoom is well positioned to become a data-entry layer by capturing meeting conversations and feeding them into systems of record.
“a company that would be well positioned to do that would actually kind of be Zoom”
Ghodsi said Databricks' early attempts to automate software engineering failed because of how people were organized, not because of the AI.
“initial attempts at automate, a lot of the software engineering at Databricks kind of failed. Even there's nothing wrong with the AI, the problem is the humans and how we were organized.”
Ghodsi said Databricks' finance team moved from Excel to Python and machine-learning forecasting models built first by an external data science team.
“finance have kind of moved from Excel to Python, uh, largely at Databricks”
Ghodsi said some AI startups with no revenue are valued at $10 to $30 billion, which he called a bubble.
“I mean, there are startups with zero revenue, worth, you know, ten, twenty, 30 billion, that's a bubble.”