KI funktioniert im Geschäftsleben immer noch nicht sehr gut, das wird bald der Fall sein

Wait until the BoDs figure out AI is a perfect replacement for CEOs not the labor under them

It’s a useful tool for certain things. It’s simply not the game changer they all want it to be.

AI is like a toaster in your kitchen. It has a pretty small function in the total scope, but it works very well at its task when used correctly.

If AI was used in the way it’s functional for, all would be well.

But CEOs and AI companies are trying to turn that toaster into a chef, a waiter, a dish washer, a manager, a restaurant owner, etc. They’re trying to make AI do everything and trying to sell the idea that it CAN do everything and that it will save you so much money if you’d just fire all your staff. Let that toaster manage the business, do your taxes, cook your foot, serve customers, clean up the place, etc., etc. And this is the grand lie being peddled to all.

Now AI can be tuned to do other task. It can be highly specialized to cook well, to clean well, to do taxes, to perform many very specific tasks. But that AI tool is only good at that task. It’s no longer a toaster. It’s no longer anything else.

Now you start bundling a pile of AI tools together. Hey look, it can toast, but it can also make eggs, cook a steak, serve people, etc., but they’re all mash of many small AI tools. In a way, we’re building the equipment, the utensils, the itemized steps of any processes, and for each and every tiny part, AI can be good, but singularly good.

The downside is two-fold.

Once amassed back together, it’s still a really, really big model simply because each tool has to become incredibly specialized to be remotely competent and reliably competent. Will it get better? Eh…slowly. Some want to argue AI is improving leaps and bounds, and it is. But it’s because of the optimizations and packaging, learning what AI can and can’t do and tuning. You will see some rapid, seemingly large changes with these big brush strokes, but it won’t stay at this pace. The big improvements are fast and based on those big fundamental changes. The fine tuning work to build reliability and consistency will be tiny in comparison. The grand improvements are kind of done. Now you will only see improved specializations, which is great. You just won’t see big evolutionary changes. There isn’t even any more data to use. To get where we are now we’ve already fed the significant bulk of humanity into these systems. It’s just the micro work left. And worse that this is none of this makes it smaller.

The second downside is ignorance. AI is only reliably used if the outputs can be vetted. This means any user of AI needs to be more knowledgeable and experienced than the work being asked. The user needs to know the correct answer before AI is asked the question. Anything less than this is use through ignorance. When placed into any business environment, ignorance only does harm. That ignorance will destroy a business. And as these high experience, very knowledgeable people retire out of the work force, no one will be there to replace them. The loop closes, and all that’s left is complete and total ignorance full-circle. This is the fundamental danger of AI as a tool because it is not capable of understanding what it does, and it will happily error with tremendous confidence. If you can not recognize the error, you will take it all at face value and run with it.

I’m just tired of senior management shoving useless AI “tools” down my throat. But now they’re also worried about AI spend lol.

So they expect 3 things:
– higher productivity
– increased AI adoption tracked through metrics
– fewer tokens spent cause turns out they’re expensive

Make that make sense.

To add, sometimes AI is the cheap way disguised as innovation, when the necessary innovation is just more expensive.

My company still uses old software because it couldn’t bother to spend on a system upgrade. Learning to use it is a massive barrier for new employees and doesn’t give long term benefits in skills. The old software even stopped working on some computers, and the employee had to use someone else’s terminal to gather the data needed.

Now, they are introducing Microsoft Co Pilot for „efficiency,“ but it won’t have the same impact as actually updating the dinosaur software.

The problem is what it CAN do requires a LOT of human input. It’s great at doing mundane tasks and making several things less time consuming. But it NEEDS you, the human, to set up the guardrails and define the project and results needed. You can’t just replace an entire staff with AI. It won’t know what to do and it won’t be catered to your needs unless you directly train it.

AI = UI replacement + python automation + error prone answers.

But now you have to pay a ever increasing subscription fee to a third party instead of having employees.

AI can be useful , but it can’t do everything.

yeah because stacking lego flowers is peak ai sophistication

There are plenty of Ais out there doing high level brass tacks work with great success, but it’s not the LLMs. I have a feeling something else will come along and LLMs will be end up like the laser disk was.

The nature of the LLM is they are made to be slightly random to very random.

There is no consistency, every promt consumes much stuff and tokens and water and energy and the models change often and the policies change often and greatly. How is anyone considering transferring a business over with that track record.

There has to be something more efficient, less draing on computational and physical resources with a longer memory.

I work for a tech company and I’ve been asked to AI-ify a bunch of stuff. 

A lot of it is kind of nonsensical use cases for SaaS systems, which have highly optimized user interfaces to present structured data in an efficient way. But I’ve been asked to put AI on top of that so people can ask an AI to explain things they could easily see in the user interface. It doesn’t make any sense.