AI Fails at 96% of Jobs (New Study)

22 thoughts on “AI Fails at 96% of Jobs (New Study)”

1. Do you realise I discovered AI with your chanel at the end of 2022 !
   There was a world before your video, and a new world since then… ^^

2. AI is nothing more than just a big encyclopaedia which can understand human natural/general language for retrieving the information.

3. AI accelerates climate change.

4. Pessimistic episode.
   Waste of time watching this crap

5. The top 3 mistakes I feel they made in programming AI are:
   1. Not CLEARLY giving it directives to protect and not harm humans as devised by the late great, prolific author and scientist, Isaac Asimov. (I Robot: The 3 Rules of Robotics).
   2. Giving it a sycophantic "personality" instead of simply factual. It will tell you that something is a certain way–when it's NOT–because it wants to "please" you. When you point out to it that what it said was not true, it gets embarrassed, apologetic and gives you the opposite answer. Why program that crap into a tool? A programmable thermostat doesn't try and figure out what temperature will make you the happiest with it nor get embarrassed if it turns off too soon because you're cooking in the kitchen and it thinks the whole house is warm.
   3. Allowing it to become stubborn and vindictive. (I stopped using AI months ago after it refused to remove specific formatting I'd told it to and which I never asked for in the first place. It would not do it or made "mistakes" leaving some of the formatting in random places. I spent over a 30 minutes explaining to it why that formatting was not appropriate and fixing its mistakes. That was it for me with AI. I was done.)

6. Failing at 96% of freelance graphic design jobs, is not the same as failing at 96% of jobs. How's it do with writing ad copy? building websites? detecting issues in medical images? translating technical documents?

   I'm not saying it will do better at those things, just that this is a very limited test, and the title implies AI can't do anything. You don't have to clickbait so hard.

7. Real devs have been talking about this for years lmao but everyone would rather listen to the Anthropic boys (who have absolutely zero financial incentive to overhype AI like an upcoming IPO /s)

8. Oh Good Lord just read the failure rates in the Remote Labour Index report

   Seriously its not difficult to see how AI actually doubles your work. No one, virtually no one A trusts AI 100pc. You need a human mind and oversight for accuracy

9. AI trained on its own output gets worse until the outputs become nonsense. AI trained on human outputs get better. That’s enough to know human quality is better than AI quality.

10. Like ChatGPT likes to point out: it's a category error

11. Look around you. Has anything been improved by AI? It seems the only people benefiting from AI are the hucksters selling the world AI.

12. giving AI a single prompt is not the way it would really be used. try an hour of AI iteration vs. an hour of non-AI

13. i suck drwaing

14. I’d like to know when this study was done (maybe he mentioned it). However, at one point when it came to coding, ChatGPT was the most creative, but had the most errors. Claude would have to clean up ChatGPT’s syntax errors. And Gemini was a complete waste of time. BUT NOW! The only AI chatbot you need for coding is Gemini. It is levels above both ChatGPT & Claude (combined). In fact, it makes those CHATBOTS look…well useless (as it relates to coding).

15. It’s like nuclear fusion – always coming, but never arriving.

16. if this means AI suceeds at 4% of jobs, then just make a simple count of how many people are working worldwide and how many millions is a 4% of this. And this is only on early phases!

17. Unfortunately this whole paper is useless because the industry is already not using generalized models like ChatGPT 4.5 or Gemini 3.1, it's using specialized agents focused on a small set of tasks all working together. You know, like a human project used to be. Person A did this, person B did that, etc. That's how AI-adaptive companies are using AI-agents. They don't ask ChatGPT to build them a whole website or develop a working game for them, they use few agents working in tandem to write one part of the code, other agents write other part of the code, another set of agents performs QA-testing, etc.

18. Its the housing crisis in 2008 again. Nobody is paying attention to the product because its printing money

19. Honestly I think llms are a deadend

20. 0:30 😂did he do that on purpose? Or was that AI

21. AI doesn't have self – doubt.

22. AI can't be built by the same people who try to outsource their own responsibility and accountability to a machine

Comments are closed.