LLMs and the Way Back Machine*

LLMs and the Way Back Machine*

Image by Pete Linforth from Pixabay

First, a bit of history. At the dawn of the Web 1.0 era, everyone felt the need to have a presence on this new information superhighway.  Something. Anything. Businesses/corporations started putting up websites, which by today’s standards were placeholders, for which they paid millions to early web-focused ad agencies/web dev shops. But consultants to whom they paid thousands/hour advised them that this was what they needed to do, or their businesses/corporations would become irrelevant in this new tech age. For context, HTML coders were commanding salaries well into six figures. A lot of money was being thrown at a lot of youth and inexperience – web shops where the founders knew nothing about business, luckily, working with clients who knew nothing about the web. If the young founders walked into a client meeting with a palm pilot, they were clearly members of the digerati and you needed to go along with anything they said.

These young companies were renting way more office space than they needed, hiring way more employees than they needed, and were running out of money, so they’d throw a party, get some press, and get acquired by a large company/corporation. Who’d learn too late that they’d acquired little more than smoke and mirrors. But what they really bought was the hype.

Which is a large part of the reason why the Web 1.0 bubble burst.

We mention this as here we are in the Age of AI, which promises to again change everything. And it will, but it’s still early days and what man cannot remember, he is doomed to repeat and thank you once again, George Santayana. Again, too much money is being thrown at any startup focused on AI, the new aphrodisiac for investors. No matter that many are features, being acquired by large companies gripped by FOMO, as we witnessed the first time around.

There were precursors to social media as we know it today back then. Six Degrees, Geocities, MySpace. Standouts in their day, but gone and long forgotten as Facebook and LinkedIn et al rose to take their places. AOL comes to mind. There was a time when AOL was synonymous with the web. To most Americans at the time, there was no way to access the web without an AOL account. There was a time when Yahoo! owned search. Is there really such a thing as too big to fail? What man cannot remember…

We’re still in the early days of LLMs, although how long has the media contended that it’s going to take over jobs? Decimate entire employment verticals? LLMs are currently little more than a tool, to be used selectively, and not a replacement for human experience and expertise, and in case you missed it, the new version is not only expensive, but potentially prohibitively expensive. As Shelly Palmer reported, “The new models require significantly more computational resources, with operational costs estimated up to 10 times that of GPT-4. One company we work with is spending about $60,000 per month on GPT-4. To do the tasks with o1 would cost approximately $3,000 per hour.”

Not that the hallucinations are gone either, mind you. And as always, why let a good opportunity go to waste, and this just in: How an AI ‘debunkbot’ can change a conspiracy theorist’s mind, said Popular Science. One example used in the piece: “Nearly a third (29%) believe voting machines were manipulated to alter votes in the 2020 presidential election,” and never mind that Hacking Convention Finds Voting Machines Were Easy To Hack, BuzzFeed News reported, and we’ll leave it at that.

So, it would seem that someone did find a way to put hallucinations to good use, and if it has the power to ‘deprogram,’ keep in mind that it also has the ability to program and down the slippery slope we once again go? Just an observation, and a heads up to founders: having a useful and preferably even monetizable use case is always a good idea.

As we said, LLMs are still in their infancy, as was the Web itself in those early, relatively innocent days of Web 1.0. We were told that the web would change everything and indeed it did, considering the Age of Social Surveillance. LLMs seem to be holding the same promise, and history is always a great teacher, so tread carefully, and do keep in mind the mistakes of the past. We do realize that to err is human, but in the words of Oscar Wilde, once is an experiment, twice is a perversion. Onward and cautiously forward.

*Way Back Machine

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.