I have been trying to articulate these thoughts for a while.
What do you think is the main innovation behind LLMs ? I mean, really ? What makes them a revolution for everyone ?
In my (humble) opinion, most people are misunderstanding what are LLMs great at. Sometimes, people refer to LLMs as generative AIs. I think this term is misleading.
The strength of such algorithms is not that it generates text, code, images better than before; it’s that it understands what you mean, even if you are not very descriptive. It is very important, in my opinion, to make the distinction.
You want an algorithm that generates code that works 99% of the time ? We can already do that ! You know, using a compiler. But now, it takes less instructions to generate code that does more things. Not always correctly, but who cares ? More is better !
Therefore, I conceptualize LLMs more as interfaces, another way to transmit ideas from your mind to your computer. They are algorithms that ingest and infer; imitate more than generate. To make them actually replace any profession in full would result in failure for that exact reason. They are machines of understanding, not reasonning.
Seriously, take any of these AI companies that say it is pointless to hire Engineers, and go to their careers page. What will you find ? You guessed it, job postings for Engineering positions.
Taking Software Engineering in particular as an example, I would propose that the act of writing the code is really the easiest part of the job. The reasonning, following up on tasks, monitoring, knowing what normal system behaviour looks like (!!), debbugging, planning for the future; that’s what a Software Engineer actually does.
The act of writing the code is actually trivial. How do you know the decisions you make are not going to slow you down in 6 months ? That’s the difficult part. And LLMs can’t really help you with that.
So what is all the noise for ?
Comparing AI companies with more “classical” tech companies, there is one thing that is clearly different: the budgets required to make innovation happen. It’s really a number’s game, so you drive up the hype to get more substancial investments. Competition is therefore much much harder for new players, but this is besides my point.
Make no mistake though, there is a lot of money to make. But do you really think these companies will be profitable because they make fancy chatbots ? Of course not. AI algorithms are excellent at one thing: pattern detection, ingesting copious amounts of data.
And what is that veeeery good for ? Defense contracts.
Such models take away from the responsability of choice. How can you blame the judgment of a machine, of which nobody understands how it really works ? What we are really creating are algorithms that absolve any guilt from people that hold a lot of destructive power. Some applications are of course useful, some are just straight up terrifying. Meanwhile, tech bros make their money.
Pay no attention to that man behind the curtain.
It’s not all hopeless, of course. A lot of developments are extremely exciting and interesting. But I think the average person is fundamentally misunderstanding what LLMs and AI are really about. Moreover, it is my opinion that the tech CEOs that directly benefit from this misunderstanding actively keep misleading people on purpose. That’s maybe for another post, though.
Until then.
See you next time.