Claude's Shmuck is Bigger!

Maybe that isn't what matters most

When Anthropic announced the release of its large language model Claude 3, they highlighted a chart comparing its test metrics to various LLM competitors. It is as if Claude showed off a picture of its shmuck and yelled, “Mine’s bigger!”

Many industry pundits care about who has the biggest shmuck. I differ.

The pundits look at advances in machine learning as a race to build the ultimate Answer God, or Artificial General Intelligence. They treat AGI as a shmuck and proclaim that size matters. In my view,

intelligence is not a thing at all. It is an ongoing process.

I think that the most likely business scenario is not One Model to Rule Them All. Instead, I foresee multiple application-specific uses for machine learning. In medicine alone, there will be applications embedded in robots, applications for aiding in diagnosis, applications to aid in record-keeping, applications to aid in processing insurance claims, and perhaps more.

The superpower of LLMs is their ability to communicate with humans in ordinary language. This will make it possible to build applications faster and with a much more intuitive user interface.

Under this scenario, an LLM will not require a gigantic base of training data in order to be effective. Instead, it will need training data and human reinforcement that hone its skills in a particular application.

A medical diagnosis app does not need to “know” about military history or astrophysics or classical music. It does not have to be trained to give answers that are politically correct. It needs to be trained to give answers that are clinically useful.

I am old. I can remember times when the consensus view of where an industry was headed turned out to be wrong. I wish these cases were taught in business schools and referred to in the popular press. In the rest of this post, I am going to give two examples.

The Financial Supermarket

In 1981, the retail giant Sears acquired Dean Witter Reynolds, a major stock brokerage firm. According to Wikipedia,

At the time of the Sears acquisition, Dean Witter Reynolds had a retail broker force of over 4,500 account executives in over 300 locations with over 11,500 employees in total.

According to ChatGPT4,

The strategic rationale behind the merger was based on the concept of a "financial supermarket," where consumers could access a wide range of financial services under one roof. Sears aimed to capitalize on its massive retail presence and customer base to cross-sell financial products. This was seen as a forward-thinking move at the time, given the nascent state of financial services cross-selling.

That is how I remember it. At the time, I was an economist at the Fed, working in a section that studied the financial industry. We read The American Banker daily. for that industry newspaper, the Sears/Dean Witter merger was perhaps the most important story of the year.

The financial industry at the time was undergoing rapid change. It had entered the 1970s highly fragmented, with interstate banking illegal and a strict regulatory dividing line between commercial banking (taking deposits and making loans) and investment banking (marketing securities on behalf of corporations).

In the 1970s and 1980s, the themes in the financial industry were consolidation and deregulation. Regulators were giving financial firms permission to invade one another’s territory and business specialty.

Pundits and industry experts touted the emergence of a “financial supermarket,” a single institution that would meet a customer’s needs for payment processing, savings and investment, insurance, and more. When Sears acquired Dean Witter, many pundits in The American Banker and elsewhere thought that this merger represented the future of financial services. But a few competitors derided the combination as “stocks and socks.”

The result? As ChatGPT4 put it,

Over time, it became clear that the synergies envisioned were harder to realize in practice. The financial supermarket concept did not resonate as strongly with consumers as expected. By the early 1990s, Sears began divesting itself of its financial services operations. Dean Witter was spun off in 1993 [other sources say 1992], eventually merging with Morgan Stanley in 1997.

In fact, no one else built the “financial supermarket.” Contrary to the pundit consensus circa 1980, you do not go to your mutual fund company to buy insurance. You do not buy stocks at your bank. The financial firms that grew up with the Internet, like PayPal and Lending Tree, are specialists, not generalists.

Content Plus Eyeballs

In 2000, Time-Warner merged with America Online. At the time, this was the largest merger in history. Pundits gaped in awe. Time-Warner, like Disney, owned the rights to movies, music, and other media. America Online was dominant in the provision of Internet access to consumers. The combination of content plus eyeballs was supposed to be unbeatable. As ChatGPT4 put it,

The expectations surrounding the merger were high. Analysts and executives from both companies touted the potential for leveraging Time Warner's vast content library across AOL's extensive internet subscriber base. The vision was to utilize AOL's digital platform to distribute Time Warner's content (which included news, entertainment, and media assets from CNN, HBO, Warner Bros., and more), thereby tapping into new advertising and subscription revenues. The merger was presented as a strategic play to dominate the emerging internet landscape by combining content with online distribution.

As it turned out, “content” is not a fixed lump of copyrighted material. Each year since 2000, as much content has been created on the Internet as existed in all previous years. Compared to what is on the Web or on YouTube, what Time Magazine and Warner Brothers have to offer is trivial.

AOL’s franchise was built for a world of desktop computers and landline modems. Smart phones and broadband Internet made AOL’s ubiquitous CD-ROMs obsolete.

As ChatGPT4 put it,

By 2003, it was clear that the merger had failed to achieve its strategic and financial goals. In 2009, Time Warner spun off AOL as a separate company, effectively undoing the merger. The AOL-Time Warner merger is now widely regarded as one of the most ill-fated mergers in corporate history, serving as a cautionary tale about the risks of merging companies with vastly different cultures and business models, especially during the height of market bubbles.

Conclusion

I used ChatGPT4 to write up the history of these two mergers because it was quick and easy. But I do not think this represents a powerful or important use of machine learning. The really compelling business uses, including obvious business opportunities, will be for specific applications, not for retrieving general knowledge.

ChatGPT and Gemini and Claude can go on competing to be the biggest shmuck. If I had venture capital to deploy in the machine learning space, I would look for companies building specific applications.