Been awhile since I rapped at ya. Like a year. The puppy we got last summer is now an adolescent, wild and carefree. I kept meaning to write in the blog and never did. Life on this planet moves fast.
Anyway. I kind of look at this now as a fresh “relaunch” of Intellectual Dandruff. Less fronting, more posting. I’m going to try and cover the shiz I said I would back on my first post. You know, actually follow up on that.
One of my non work focuses lately has been AI- largely LLMs and the metaphorical shit stain situation that revolves around it. For a couple years now we’ve been hearing how LLMs will lead to AGI (Artificial General Intelligence, essentially real thinking machines) and we’re all at best, out of work, or at worst, fucking dead. And most of us at our place of employment over the last year have been encouraged or flat out directed to use Claude, Co-Pilot, Chat GPT, or some other LLM, to either become more efficient at our work, automate our work entirely, or I guess justify management having paid some dollar amount for the service.
Ed Zitron is a far bigger expert than I on all this- so check out his site and podcast if you have not/are not- https://www.wheresyoured.at and Better Offline respectively. Ed’s got the goods that your gut has been telling you. This is a cult selling you a Ponzi scheme. AGI is no where NEAR happening, and for almost all work, LLMs can’t find a good use case and/or a profitable one. If you have tried to use say Co-Pilot for anything you probably already know that it’s a lot of smoke and mirrors, a whole lotta squeeze and almost zero juice. It can help you write emails! It can automate your inbox! It can take notes for you in meetings so you can “pay attention”! I believe if you can’t run a meeting, take notes, and pay attention, we probably should relieve of any responsibility. Maybe we can let you wrangle the carts from the parking lot. Maybe.
This is not to say though that there aren’t a small handful of things that AI is doing moderately well. From what I understand through Ed and others, including my own experiences with Claude, is that mathematics, coding, and some other big data tasks are seeing some improvement by using LLMs. Makes sense. As they get better at understanding plain language and how to translate that into parsing data I can imagine we’ll make some progress in areas that need data analysis. And as Ed notes, along with others, that while the coding harnesses on top of the LLMs can spit out code, the quality is not always fantastic. I wrote a iOS app, with 0 Swift knowledge, in 4 days. So far it’s working great, but 1) I don’t know if it’s shit code and 2) even if it isn’t, were I to go ahead and put it in the Apple Store, how would I support it if there is an issue? So even for use cases where LLMs can be additive, it’s far from perfect and it’s not going to eliminate entire job functions overnight.
Part of the reason I created the app was for fun and partially because I paid $35 for a year subscription to an app that promised the functionality II wanted and failed at it (for specifics, this is a collectibles app- I wanted to inventory all my dorky action dolls in one place because I am anal retentive. The app used AI to determine the action figure and provide specifications like brand, scale, name etc. and created an entry for it. Problem was, it was REALLY inaccurate. out of five action figures it only got 1 right. So I thought “There has got to be a better way” and well, shit, I have oodles of free time so maybe I can see if Claude wants to help me build the app).
I started using Claude Code but soon abandoned that in favor of just the LLM so that Claude could say “do this, add this here”, etc and I would write or cut and paste things in Xcode. Ends up, I have learned a little bit of Swift code. It’s been a fun way to see how things are structured, what calls we’re making, and other structural things. This is not to say I am now a iOS developer BY ANY STRETCH. I think that one use case for AI is learning/teaching. It was a LOT more fun to build an app than to watch video after video, then maybe do a lab, like I did when I was thinking of getting an Azure Dev certification. I like hands on learning and assuming the LLM is a decent coder, it feels like a way to do some fun, cheap(ish) hands on learning.
Of course, this micro example ignores the weird economic realities that Ed and others have documented about the actual cost of AI. If I am understanding it, the main players (Anthropic and Open AI) will need 1.1 Trillion! in revenue by 2028 in order to meet the compute reservations they have made. TRILLION. And this from companies that so far are losing tens of billions of dollars each year. And you might say, well, so what if they don’t? So those contracts get cancelled and they (Anthropic and Open AI) will pay the price. As I understand it, there is a lot more than that as these companies that are providing the data centers and compute, whether built now, being built, or being planned, then will lose that money and they will be financially impacted. And then other companies will be. It’s a chain of events. And it’s coming.
Add to that the general fear mongering (Ed’s podcast had a great episode on that earlier this week) around hey- this might actually make humanity worse or even kill us- and you find that the every day impact of AI has been terrible. Imagine if Ford said (and I am kinda stealing this example from the episode) “Hey, uh, there is a GOOD CHANCE that F-150 will rear up and decide to KILL ALL HUMANS but it might not and so we should press forward with it…”. The general population would laugh you out of business. Yet the big dogs in tech and the AI companies in particular are doing just that.
I was on a call today where someone mentioned that a member of the board of directors for our company said recently that ALL our coding should be done with AI. Seriously? I looked the board member up- they are a big wig at a venture capital firm, so… yeah. That tracks. Cause VC has pumped BILLIONS into AI and if they don’t get something back for their investment… well, hopefully they did not invest your 401K or Pension, because you may soon be taking a loss there.
Sigh.
There is so much more to unpack with LLMs. It’s truly a crazy time. Tech, Politics, the USMNT winning their group at the World Cup. Down is up. Up is down. Nothing lasts forever, even cold November rain. The best we can do is stay vigilant, and skeptical. When someone is selling you a bill of goods that seems too good to be true, it probably is. Best you can do is survive.
So, keep your heads up folks. World will keep turnin’ and you’re on it for the duration of your life. Stay kind, stay humble. The AI overlords might soon be not much more than a footnote, like Google Glasses or “the Metaverse”. We can only hope for now.
Intellectual Dandruff 