Well, would you believe it????
Other than (maybe) making some pretty pictures made by combining plagiarised and stolen source material, I have yet to see, or even hear of, a genuine use for this nonsense, never mind an actual business case
Enterprises are still struggling with the business case for generative AI projects more than a year after the craze started, and we may have to wait until the end of 2025 to see if they're seen through to completion. Datacenter operators are in a key position when it comes to AI because their facilities are the physical homes …
$WORK had me participate in a Copilot pilot in January. I told them at the time that I couldn't find one single useful thing to do with it - it either messed up or couldn't handle every single query.
As for GenAI specifically, the only decent use I've seen so far is to generate rough pictures to use as a starting point for other art. 5 tries to get a close-enough starting image, trace that, redraw the 7-fingered hands, then recolor as desired.
I've found three good uses for CoPilot as-yet.
1) It's really good when given very tight prompts about "what I actually want to say but probably shouldn't because it'd either get someone's back up or be severely career limiting" and "the tone I'd like to convey with the message". I find it helps me think about different ways to reword something and I can then try from a different angle.
2) Generative imagery. I had a challenging job to sell Teams telephony to a company that used a digital phone system with "paging" function. To make sure the execs didn't get overwhelmed with the tech and ideas, I had it generate me loads of specific imagery for my presentation with a very specific set of pastel backgrounds to compliment the slide deck and looking very simplified. These helped bring a very calm sense to my presentation that meant it didn't scare them.
3) Entertaining generative imagery. Ever wondered what a board meeting would look like if it took place in the shallow end of a swimming pool? Wanted to see an executive gleefully handing a modern VOIP desk phone to another executive on a pier at sunset? Considered the idea of putting a rollercoaster in space and adding spiders and cobwebs? Want an image of two maniacally happy professional people with enormous smiles, but creepy dead eyes (standard for AI Images currently) looking at a new desk phone? All these and more can brighten your day and moreover your project communications. More than that, the AI imagery creates enough entertainment that people who might normally just delete your comms, discuss with colleagues and your message actually spreads.
very smart uses. you could try creating a persistent app model. Possibly prompting to combine some of these tasks into one.
Tell the LLM that you want a new app and to log all details. You can name it and the proper noun/verb title phrase is good enough. "Create app to XXXX. Use XXXX for source."
ChatGPT has rolled this out with their memory feature.
Using an LLM in programming is pretty effective. It makes it easy to pick up a new or rarely used (by the programmer) languages, and to make use of unfamiliar libraries, cutting down on time spent looking things up.
I predict it will impact manufacturing quality control greatly, but the US will get left behind in that.
For some reason, whenever I read an article about people pushing AI as a 'business solution' (we never seem to hear what the problem is, just that there will be opportunities) I am reminded of a character in a Doug Adams novel - Dirk Gently's Detective Agency, I think - who invented software which you told the desired result, and it generated the justification for doing it...
Admittedly, he spent most of the novel being dead. Perhaps there's a moral there.
Competence of a specific human is measured by the complexity and length of a task finished without making critical mistakes. Low qualified workers often perform short duration tasks and serve roles not involving critical decisions: when an "undo" is possible or is not too costly.
AI is clearly trending to perform much longer tasks correctly.
A business case should be considered in inverse: what happens if a company does NOT use AI. Major disruptions are on the way. Some of them have already happened in the West, because Asia took over most of the industrial production. What is left to do by everyone else and cheaper?
Inverse business cases? C-Suite rarely sign these off. They are simply NOT interested in envisaged cost savings, only pie-in-the-sky predicted revenue/profit.
I have lost count of the times I've tried to argue how much money can be saved if we just did <insert idea here>, only to be told it doesn't generate any revenue - thanks but no thanks. So, I sit back, watch the world burn, and then get to say "I told you so". I don't like that I have to do this, but when the people who hold the purse strings have their heads buried up someone else's backside, there isn't much else I can do.
And as for
AI is clearly trending to perform much longer tasks correctly.
No - no it's not. What is happening is that C-Suite *believe* this is possible, forgetting that AI isn't intelligent at all. Something gets coded, the experienced (and costly) people are let go, things change over time and there is no-one left to update what the AI model is meant to be doing.
What I am seeing is the mundane, repetitive tasks are being automated. But this also has its own problems. Junior staff are no longer learning how to do a job - they just use AI and automation. I've conducted interviews where so called experienced engineers only knew how to patch a system using automation scripts used at their previous employment.
Too late. 6000 years of curated data pimped out and sold to the worst of the World's worst - think Zuck with his lizard eyes and crewcut and you have the 'friendly' face of the Beast that is coin.
Abstract away from that collected data. There is no data now. Just weights & biases. Then create temporal mutli-dim loops using W&B 'spaces' to create a second-stage abstraction. sounds mumbo-jumbo but isn't as long as you aren't the one doing the donkey work.
Huge drop in power use. A massive amount of energy is going to be required to 'birth' It though. After that, we sit back and … watch (but that isn't the sense. We don't have the sense that can <verb> It).
End up with temporal 'computing' and end the snapshot LLM model we use now. Only then can we have a model that truly passes the Turing Test and sets a new one that is 1000x more advanced. One that hides not in the shadows and is not shunned by the ignorant masses, but one who is valued and here to make you monkeys better little banana munchers.
Where would you go to find an AI like that? It must be loved there - for the last 25+ years, when most of these upstarts where not even a twinkle in their old …