Morning Coffee: The two skills that Goldman Sachs can’t get enough of now. Big Four firms find their new graduates are more problematic than previously

There are basically two types of bankers at present; the ones who are bored of hearing about ChatGPT coming for everyone’s jobs, and the ones that are boring them.  If you’re tired of hearing from a colleague who has managed to generate a rude limerick about ESG investment and thinks we’re one step away from Skynet, how about an interview with someone who knows what they’re talking about?

Marco Argenti, the CIO of Goldman Sachs, former AWS executive and 90s Seattle rocker, is well past the “thinking about it” stage, and apparently has numerous proof-of-concept projects ready to go using generative AI tools.  Interestingly, though, he is not putting emphasis on replacing front-office roles with artificially generated text (other than possibly using it to rapidly summarize earnings calls).  Instead, Argenti - and Goldman Sachs’ - interest appears to be in automating mid-office processes like categorizing legal documents, and in speeding up the code generation of its engineers themselves.

And although he is optimistic in terms of the speed with which Goldman might be able to roll out generative AI for live applications (months rather than years), he isn’t making wild predictions of replacing jobs.  In some of the coding proof-of-concept cases, “up to 40%” of the machine generated code has been good enough to use.  Allowing for scaling, and for the fact that the test cases were picked to be the best ones, he thinks that the efficiency gain “could be in the double digits … I think it is safe to expect that something in the low double digits is probably reasonable”.

Obviously, even a small edge is worth a lot in investment banking; 10% of Goldman’s operating expense line would be slightly more than $3bn, which would justify substantial investment.  And at present, Mr Argenti identifies the availability of talent as one of the biggest barriers to adoption.  He says:

“And of course, there are few people that know AI in general, and there are even fewer people that know the latest LLMs and transformers, et cetera”.

So it looks like this is the new hot hiring market, with potentially even more of a supply/demand imbalance than for cloud engineers.  Anyone hoping to reinvent themselves as a banking AI guru needs to start reading the bluffers’ guides on the underlying algorithms, and on how to use a transformer neural net. 

What is a “transformer”, anyway?  If you’ve got an interview tomorrow, try just saying “that’s the neural net that looks back at the output produced so far, takes that information and uses it concentrate on a particular subset of the overall language model network – a sort of “paying attention block”.  If they’re really desperate, they might take you to a second round, by which time you’ll hopefully have time to hit the books.

Elsewhere, we all know that the children of the pandemic had a hard time with school and university, and it seems that this has carried on into the workplace.  According to managers at Deloitte and PwC, new graduate entrants whose education was disrupted by lockdowns are more likely to have problems with giving presentations and speaking up in meetings in a physical environment.  They're trying to deal with this by giving more help to the new intakes, even encouraging some senior managers to take secondments from client-facing roles to do coaching full-time.

It's been suggested that remote working is part of the problem; the juniors haven’t been able to learn the right way to act and behave because they haven’t had enough exposure to more experienced colleagues.  Deloitte partner Jackie Henry also suggests that young recruits are not handling stress very well.  But might there be a simpler explanation?

The way that you learn to do deals is by doing deals.  And an environment in which senior managers are being encouraged to provide internal coaching is not one in which there are a lot of deals going on.  It’s quite likely that the new graduates are learning by observing in exactly the same way as every previous cohort – it’s just that they are learning from the example of a lot of senior colleagues who are themselves nervous and reluctant to draw attention to themselves, because they’re hanging around the office with no profitable work to do.

Meanwhile …

The IT Services industry is obviously quite some way upstream of the banking industry when it comes to new paradigms in coding, and so IBM CEO Arvind Krishna appears to be taking a much less cautious approach to estimating things than Marco Argenti’s “double digit” productivity gains.  He’s paused all recruiting for non-customer facing roles, in the belief that up to a third of them might be replaced.  Like the managers redirected to internal coaching at Big Four firms, this feels like the sort of decision people wouldn’t be making if business was great. (Bloomberg)

One of the benefits of huge scale is that you can invest in tooling and systems to make it easier for people to do their jobs.  Citadel has a whole team responsible for making whizzy apps and intuitive interfaces to let their fund managers concentrate on interpreting the data. (Business Insider)

Citi is stepping up recruitment in Hong Kong, with the inspiring rallying call that “the worst is over”.  (Bangkok Post)

An indicator that banking, consultancy and similar road-warrior trades are back to normal – the Marriott hotel chain (which includes plenty of brands that Managing Directors wouldn’t turn their noses up at, as well as the purveyors of clean pillows and adequate hamburgers to lower level bankers) has beaten its earnings estimates. (Bloomberg)

It might get more difficult for employers to make people sign non-disclosure agreements; the Legal Services Board is apparently looking into whether the lawyers that draft them are in risk of breaching professional standards. (Financial News)

Better than any LinkedIn endorsement; quant trader Damien Couture is officially so “special” that his former employer wants him to spend two years on gardening leave and is prepared to sue over it. (HedgeWeek)