Latest / Minds, Bodies, and Terawatts / OpenAI Just Built a Junior Banker. Goldman's Own AI Lead Says the Grunt Work Was the Training

Transcript

OpenAI Just Built a Junior Banker. Goldman's Own AI Lead Says the Grunt Work Was the Training · Minds, Bodies, and Terawatts

• 0:00
  [Host] This is Minds, Bodies, and Terawatts, the show where the hundred-hour week just got a product page. CNBC reported on September tenth that OpenAI unveiled ChatGPT for Financial Services, a version of its business product built to research companies, build financial models, and produce the pitch decks that the recent graduates Wall Street calls analysts have made for decades.
• 0:21
  [Guest] OpenAI built it with Morgan Stanley and Evercore as design partners, whose bankers' feedback will train the next versions, and it runs on Astra, the company's newest model. In the live demo it sized up a takeover target, pulled the numbers from the standard data feeds, and produced a slide deck in the bank's own house style. Asked which banks have actually bought it, OpenAI would not name one.
• 0:45
  [Host] A friend of mine did two years of that, hundred-hour weeks and a deck at three in the morning so a managing director could change the font at six. Why would anyone mourn that?
• 0:58
  [Guest] OpenAI's head of product, Nick Turley, made your point almost word for word. Asked by CNBC whether this shrinks the need for junior bankers, he said analysts work hundred-hour weeks, and this will do what Excel did, better analysis, faster.
• 1:12
  [Host] So the pitch is that the spreadsheet didn't kill the analyst, and neither will this.
• 1:17
  [Guest] The difference is what sits inside it. OpenAI's own post says it bundled the paid data feeds bankers live on, hosts them itself, and ties every figure back to the filing it came from. Excel was an empty grid, and this one arrives knowing where the numbers live.
• 1:35
  [Host] Fine, but the job is still awful, and if a machine eats the three a.m. deck the twenty-two-year-old goes home. Where is the loss?
• 1:43
  [Guest] A Goldman Sachs partner running one of the bank's flagship AI projects answered that a few weeks before the launch. Chris Churchman said on Goldman's own podcast there is a huge danger we outsource our reasoning to these models and end up with, his phrase, cognitive atrophy.
• 2:00
  [Host] Cognitive atrophy from not making slides?
• 2:02
  [Guest] His example was junior traders pricing client requests by hand. Automate it, he said, and do you still get senior traders who fully understand? He compared it to what GPS did to our sense of direction, and said you learn this business by doing.
• 2:18
  [Host] So the hundred-hour week wasn't the product. It was the tuition.
• 2:22
  [Guest] And the bank paid the student. First-year analysts at the big banks start around a hundred and ten thousand dollars a year in base pay, so the bank was paying people to be bad at the job under supervision until they were good at it. The decks were the by-product.
• 2:40
  [Host] Are banks actually cutting the classes, or is this still a conference-panel worry?
• 2:45
  [Guest] The McKinsey partner who runs its AI consulting arm said this summer that banks are cutting junior analyst classes by as much as two-thirds, shrinking them, not ending them. And about sixty-two percent of the people banks move into AI jobs came out of those same classes.
• 3:02
  [Host] They are shrinking the pool they hire their AI people from.
• 3:06
  [Guest] And the bill does not come due for years, because that is how long it takes to turn a graduate into someone you put in front of a client alone. The manager who cancels the class books the savings now, and the shortage lands on a successor.
• 3:22
  [Host] Then the senior bankers who already learned the hard way are the safe ones. They have the judgment banked.
• 3:29
  [Guest] That is the part the launch does not touch, and the closest evidence we have says no. A study in The Lancet last year followed experienced doctors doing colonoscopies after their clinics added AI detection. On the exams they still did without it, their rate of finding precancerous growths fell from about twenty-eight percent to twenty-two.
• 3:49
  [Host] Nobody took the skill away from them.
• 3:52
  [Guest] They just stopped being the one looking. Churchman's atrophy is not a warning about interns, it is the mechanism, and it works on partners too. The bank is cutting off the supply of new judgment while the judgment it already has wears down.
• 4:07
  [Host] So who is left to catch it when the model in the deck is wrong?
• 4:13
  [Guest] The product ships with citations so a human can check every figure. Checking only works if the checker has built a wrong model before and knows what one feels like. That is the skill nobody is paying to build anymore.
• 4:27
  [Host] I came in thinking this was good news for every twenty-two-year-old sleeping under a desk. I still think the hours were indefensible.
• 4:35
  [Guest] Both can be true. The misery was doing a second job nobody named, manufacturing the people the bank will need in ten years. OpenAI automated the first job, and nobody had written the second one down.
• 4:48
  [Host] In Unscarcity, Patrick Deglon moves the apprenticeship out of the quarterly firm and into what he calls a Mission Guild, where mentoring the next apprentice is logged as contribution, the same as the deck itself. A bank has no reason to fund a banker it will not need for a decade, and a guild has no other move.
• 5:09
  [Guest] For more on who trains the next senior when the machine does the junior's job, head to unscarcity dot ai. The book is on Amazon, the audiobook is on Spotify. This has been Minds, Bodies, and Terawatts.