When I give talks about AI at work I use the phrase "AI is like an army of junior employees". That’s what it is. They work fast, but make a lot of mistakes. They're still valuable
This is congruent to my impressions, and what some experts imply. Is AI still improving quickly? Or is it stymied at simplistic LLM and waiting for a major breakthrough?
It depends what your expectations are. There is a pretty solid theoretical ceiling to how good AI can get. AI is basically a statistical engine that works on probability. That means that any knowledge that is either cutting edge or extremely specialised AI can't help you. AI is awesome on any knowledge that is mid range. But as soon as you get to any of the extremes in specialisation AI goes retarded.
Yes, AI will keep improving. But there's zero risk we'll all be replaced by AI. Or to put it another way, if you get replaced by AI then you suck.
And when it comes to following simple reproducable rules, like grammar, AI will outperform humans. Right now is a bad time to get a degree in translation. That's going to be fully AI in the future. Lawyers is another one where the market is set to become crowded.
What amuses me the most with AI is the art. What AI proves is that you can make great art with mediocre skill. It's simply a question of practicing. AI will never be able to produce unique master works. Of course. But for stuff people pay €1000 for to hang on their walls at home, AI can bang that out i seconds and it'll be superior. There's so many dirt poor artists who continuously whine about how hard it is to survive as an artist. When the reason they're poor is because they suck so much, and should get another job. AI has demonstrated this. I think it's an awesome reality check.
I read reports that some companies are disappointed in their AI efforts, but these reports may be "fake news" as OP suggests.
I have two replies to that.
1) Based on my own company, disappointments comes from people who think that AI is a "magical problem fixing machine". It's just not. Like any tool AI is great at certain things and bad at other things. If you try to use it for something it's bad at you'll think it sucks. Prompting is also a skill. Learning how to use AI is a learning curve. People are still getting used to it. I remember when Internet came. I thought it was the best thing since sliced bread. Normal people made fun of me. This is no different. Once normal people catch on it will transform the world into a better place. Just like the Internet did.
2) Journalists love hating on AI. So they are mining responses for the negative. So yes, fake news.
Dr Z: Without infringing on confidentiality can you outline your experience with AI? How much does it cost? Is it replacing workers or just increasing their productivity? What specific chores is it best at? How rapidly is its applicability expanding?
I can't actually say what we use or what we paid. That would violate my nda. But I can say that most corporations use a private Microsoft Copilot, ie does not send data outside the company to train the AI. It's baked into what they pay for Microsoft Office anyway. So very little extra cost. The big cost comes from training staff in AI. Based on this statement you can probably work out what we are using.
We fully embraced AI in November 2023. Which was early. In May 2025 we fired almost the entire HR department. AI answered faster and gave better answers. Stuff like this will keep happening. Novo Nordisk, Denmarks biggest and most profitable company (yes, the people who make Ozempic) are in the process of cutting 11% of its workforce. It's HR roles, admin and support staff. All roles AI does better. I don't know this for a fact, but I think it's AI. They're a high tech company, so will be fast at adopting new high tech solutions. They're certainly not doing it because they're struggling financially. They're one of the strongest companies on the planet.