This is a transcript of a chat with some friends.
What I’m about to say doesn’t mean that one can’t make a lot of money out of LLMs, yet I have some doubts that there are many sustainable multibillion-dollar businesses based on them.
Several years ago, long before Satoshi Nakamoto, while speculating with a friend who now is a prominent crypto personality about a possible fully P2P decentralized coin, I argued that I could not clearly see a possible way of sustainably making substantial revenues because if there are no asymmetries in the playing field (like in pure P2P), no player has a competitive advantage over others and hence a clear way of extractin value and making money. But if there are some asymmetries, granting someone some degree of control (for example to manage inflation) then that someone could clearly be the target of law enforcement and the system could have been blocked/limited, pretty much like it had just happened to qqcoins and linden dollars. (I could see unsustainable ways of making substantial amounts of money, though…)
Furthermore, I had no clear view of the underlying sustainable value of an alternative payment system, as, if there are significant improvements over the existing infrastructure, the mere existence of such a system would have put pressure on the existing payment infrastructure, forcing it to evolve. (now we have realtime, free/almost free payments throughout europe).
Now I can see underlying value in crypto: a global system based on people who globally trust it, can be more trustworthy than a local currency in some states with unreliable governments. Furthermore, the amount of technology and innovation that the initial speculation has spurred is astonishing and cannot be uninvented; I think it’s just a matter of time before these technologies become part of a well functioning, regulated system.
A couple of years ago, speculating with a friend who held prominent positions in SV in the innovation and VC sectors, we argued about Tesla risk of having no sustainable competitive advantage in the medium-long term.
I will use stark words even though reality is more nuanced, but just so to make the argument: There will be no self-driving cars wandering around our city centers, fully without human intervention; I will likely not see in my lifetime the moment in which car rental business outweigh sales. Thus, it will not be “autonomous self driving cars as a service” that will solve cities’ traffic problems.
Musk has charged customers 15k to sell them a vision that I doubt will ever happen. Actually what is legally sold is different of what normal people understand of what has been communicated, over and over again. (We’ll see the outcome of eventual legal rulings; in a worst case scenario, a class action could eventually have a dramatic impact).
This hasn’t barred Tesla from becoming the most valuable car maker in the world. But other car makers have catched up on technology (the difference in PE ratios is apalling) and eventually, it might end up being bought by someone (merged, with Musk losing control).
This is because the world is too diverse, complicated, and unpredictable (except for niches) and there are many bad people, so a self driving system is vulnerable. Hence, at a second level, the issue of liability and penal responsibility is difficult to resolve. Let’s wait for the first rulings on this as well.
Let’s come to LLMs: they are beautiful.
They produce texts that-..-wow!
But they are not at all trustworthy. We all know the examples.
This is in a large part because they don’t “understand.” A subset of understanding is the capability to explain. Explainability is a necessary (but not sufficient) precondition to ensure consistency in answers (an alternative is narrowing the domain).
Let’s forget medicine-like domains, also for accountability and responsibility issues.
Legaltech can be fine, provided human oversight. But I doubt productivity improvements for legal companies to be that much more significant than today’s best practices.
Voice interfaces ? There are huge privacy issues and moreover, are we sure they are that efficient ? Alexa has been defined “a colossal failure”.
The mere idea of contending the business to Google because of a possible search engine evolution seems to me a very long shot. Google’s business is not search, but rather the 2-sided advertising ecosystem market, of which search is a piece, and I deem Android an unbreakable asset under present market and legal conditions.
It will be fun to see AI systems promoting content based on “engagement” determined by content interactions done by AIs on texts written by AIs…
Bottom line: IMHO a lot of money can be made out of LLMs, more on long term evolutionary rather than disruptive terms, more on expectations than on long term results.
But maybe I am missing something, maybe there’s some domain where we’ll see short term disruptive results brought about by LLMs.
What’s your view ? Can you envision some ? I’d love to read your comments.
3 thoughts on “A (somewhat) contrarian view on the promises of LLMs”
…la seguo da tempo con interesse. Suggerisco di chiarire – ogni tanto, non sempre – il significato degli acronimi. Credo che anche gli specialisti apprezzerebbero.
I see the actual quality&easy of use level reached by GPT&others (let’s not forget about at least 10 good players in the field)… is similar to what happened in the mobile industry when the touch screen got the sufficient quality to be not distinguable from magic.
That tipping point has been reached by Apple at the time and is now clearly reached by any system, that can listen to natural language to do some kind of task. My daughter Camilla (10yo) actually plays with no difficulty with the NLP pre-trained systems.
On the matter of “understanding” that you feel is missing, try more ChatGPT, you’ll see that in many cases it will get the meaning of the question in order to propose actual answers.
The game that will interest me more is about ethics in proper source quoting (just like at the beginning of the internet!!).
On Blockchain I still have to see a proper real-world killer app that couldn’t be made with other technologies… before I can really understand if there will be an impact. Even in El Salvador I see that Bitcoin adoption is still struggling.