Folks know that I’m pretty “embrace the new” – I’m on record as being pretty pro-AI (GitHub Copilot with AL) at times, including having spoken with Dmitry Katson about it (Directions EMEA 2023) and contributing my books to CentralQ.ai for inclusion to make that tool stronger.
We can see how much pressure there is to talk AI in our ecosystem when you look at the fact that there are over 50 sessions at Directions EMEA 2024 on Copilot or AI, not counting how much Copilot Fever will be present in the Keynotes.
And look, Large Language Models are a fascinating breakthrough. Generative models are interesting, though I’m continuing to personally hire good graphic artists and human transcriptionists for quality results I can trust. But, all new technology has uses, misuses – hype and value.
The “Blender” analogy
There’s a lovely analogy that I respect, while I also take it with a grain of salt.
A metaphor I’ve been using lately is that being a tech-interested person and watching the AI hype is like if you had followed the development of blenders for years. You watched them go from prototypes that were basically just a spinning open blade all the way to a design that has the potential to be a consumer Vitamix! It’s really cool! Blenders have come such a long way, and they’re ready for prime time!
– Penrosesun
And then you turn on the news and see otherwise rational, intelligent people saying “gosh, imagine, soon we’ll replace all of our chefs, and our surgeons, and our high school teachers with blenders!” and your friends and family all nod and agree and say things like “wow, blenders can basically do everything now!” And when you ask people “are you HIGH?!” they show you the new blender they bought, and how well it makes a smoothie, and then they act like that’s evidence for a statement like “blenders will replace 90% of the workforce” not being utterly nonsensical and deranged.
“Scientists just have to fix the hallucination problem!” they say. When you ask what the hallucination problem is, they say “I’ll show you” and then they put their unfinished math homework into the blender and hit pulse. “You see, I wanted it to solve those math problems, but it just shredded the paper. It must have hallucinated a world where the answer to 2+2 was puree.” When you point out that it did exactly what it was designed to do, because a blender cannot do math and it will never do math and expecting it to be able to do math just because it can make both smoothies and soup is ludicrous and bizarre, they tell you that they’re sure that blenders will be able to do math any day now, just you wait, “I mean, look how far they’ve come! A year ago I would have said that blenders could never be strong enough to blend ice into sorbet, but now they can. So who are you to say they’ll never do math???”
I think LLMs, search powerups, and generative AI have great potential. But people who are a lot sharper than I am have be speculating that this is just the latest bubble after crypto, NFTs, web 2.0 or 3.0, etc. Things that have uses, but were overinflated.
How does this affect our little Business Central world?
People will (usually) admit when they don’t know. LLMs are happy to spit out nonsense that are answer shaped.
The below quote came up in a support ticket in October. One of our new-to-BC customers wanted to do something interesting, restricting values based on Dimensions. Sounds cool, and you can restrict Dimension Values based on which master data is involved, so the reverse sounds plausible.
I heard the sentence I knew I would hear and was afraid of the day:
I checked with Copilot how to do this, but I can’t find the settings it said to use.
– BC Support Ticket
Rather than saying the correct answer – “you cannot do that, I’m sorry” – Copilot very happily spat up an answer that sounded real.
So, we had to explain to the customer not only was the feature requested not possible, nor an intended use for Dimensions, but we also had to explain to this new BC Customer “Copilot gave you wrong information”.
I don’t look forward to trying to teach that customer how Copilot can help generate his item descriptions, match his bank recs, or interact with his customers. No amount of flagging a dialog with “Preview” is going to give that customer confidence.
So What Next?
Be proactive.
Talk to your internal teams about it, talk to your customers about it. But it’s ok and possibly even the responsible choice to say that the new technology is just that – new!
Try it, but verify. Explore the possibilities, but keep your expectations grounded.
Meanwhile, when it comes to artists, musicians, writers, actors, and all manner of other creatives – please continue to support humans who depend on it for life.
- Sometimes Cutting Edge is Bleeding Edge - 3. November 2024
