LLMs, AJAX, and computing's next platform shift

Sep 21st, 2023

Large Language Models will change everything. That seems pretty clear to me. However, I am not certain that LLMs will represent a true “platform shift” for consumers.

To date, we have experienced two platform shifts in computing:

1. From offline desktop computing to internet desktop computing, at the birth of the consumer internet in the 90s.

2. From desktop computing to mobile computing. The first iPhone was released on June 29, 2007; I think of the mobile era as having started then.

It isn't clear to me that LLMs will look the same. I do think they will be embedded in everything - but I think we are incorrect in referring to them as a platform shift.

Asynchronous Javascript, or AJAX, revolutionized how users interact with software. But we don't think of that as a platform shift, do we? It's just a protocol. But it's a protocol that literally every single piece of client software built is now hopelessly dependent on.

I think LLMs will look the same.

--

From LLMs, I expect the following:

1. A majority of financial returns will accrue to the incumbents. No startup can buy more H100s than FAANG. And it makes sense for FAANGs to embed LLM capabilities into their existing cloud platforms. FAANG will eat an increasing share of startup gross margins. Any breakout consumer products will be swallowed by FAANG CorpDev. The only hope to change this is if OpenAI, Anthropic et al can build a text-native ad network that surpasses AdWords. The consumer internet runs on advertising. Unless a new challenger can build a better ad medium than those already run by FAANG, I think LLMs will further entrench the latter's dominance - and this doesn't even consumer the prospect of a national LLM license, or related regulation.

2. LLMs will be embedded in most products. However, I really don't know if I consider this a “platform shift”... Again, the AJAX comparison feels relevant - and curiously, I have not seen this point mentioned anywhere else yet. Today, we take asynchronous client operations for granted - particularly my generation, which came of age on a post-async internet. Ten years from now, I think we will feel the same about LLMs. Every great piece of software will have LLM operations in some form.

So what about platform shifts? I think the true platform shift will come… soon. LLMs will enable the creation of new consumer hardware, which will be incredible.

Alan Kay famously said, “People who are really serious about software should make their own hardware”. Well… it just got a lot easier to build great software. I think this paradigm may now flip: incredible software may now be a commodity. The true challenge may now be to build hardware that can best take advantage of this.

While LLMs are not literally a protocol, I think they are best understood as such. Rather than LLMs being the platform themselves, I think they will be the protocol that enables the next platform shift.