Could this be a trend?

[Janjohn]

Ford rehires experienced engineers after AI misses the mark

https://www.foxbusiness.com/technology/ford-rehires-experienced-engineers-after-ai-misses-mark

[allinon72]

There's no bigger AI cheerleader than boomer executives, who think they can use it and fire masses of people. Until they learn the models are trained by Reddit posts.

[Car-Addicted]

I use AI to write web copy, help trace website intrusions and as a starting point on correcting internal errors but in all cases the AI is a tool and not a sole solution. It still takes personnel.
Had a conversation with a senior tech at a medical facility and he is using AI but only to review his code. They are using Anthropic's Claude to write C# code but the issue is still one of defining problems and solutions.

[dcstep]

Originally Posted by Car-Addicted

I use AI to write web copy, help trace website intrusions and as a starting point on correcting internal errors but in all cases the AI is a tool and not a sole solution. It still takes personnel.
Had a conversation with a senior tech at a medical facility and he is using AI but only to review his code. They are using Anthropic's Claude to write C# code but the issue is still one of defining problems and solutions.

Exactly.

Still, I can now use Claude to research positions and help me write position papers to The Federal Reserve, OCC, Financial Accounting Standards board. Claude and I can put together a meaningful, thoughtful response in three or four hours, where it used to take a week or so, with a couple of staff researching and doing initial documentation for my review.

I find errors by Claude in almost every case. He'll often grab just the easiest data to find. When I say, "I know that there must be such and such", he'll apologize, go get it and evaluate it, BUT you can't rely on Claude's work blindly. Someone in the process has to have deep knowledge.

How much better will it be in another year or two? I'd be surprised if it weren't much more reliable.

BTW, I'm a 78 YO boomer. My youngest daughter is a very successful techy millennial. We have lots of fun with the generational battle.

[KeFoZ4]

This is 1 step back after 2 steps forward. The capabilities will continue to increase. Ford may lay off only half of the engineers they just rehired in the next phase.

[JABCAT]

Originally Posted by dcstep

I find errors by Claude in almost every case. He'll often grab just the easiest data to find. When I say, "I know that there must be such and such", he'll apologize, go get it and evaluate it, BUT... Someone in the process has to have deep knowledge.

My university just vetted & launched versions of Chatgpt & Claude for all faculty & staff to use. I've been messing around with it for a little over a week. It's wrong more than it's right on so many things. Then the "I'm sorry, you're right" or the best is when it argues with you, even though it's wrong. It take more time verifying the info it gives you than if you just researched it yourself.

Then you have students using it to write basic papers, which is blantently obvious. And I laughed yesterday when I received an AI-written email from a student (who happens to be a student that has submitted 2 AI-written papers already this Summer term). If it weren't a 6-week class, I'd have referred the student to Dean's Office for academic dishonesty already. I have a feeling that student won't make it very far.

Or, maybe Claude IS a student in my class...

[dcstep]

Originally Posted by JABCAT

My university just vetted & launched versions of Chatgpt & Claude for all faculty & staff to use. I've been messing around with it for a little over a week. It's wrong more than it's right on so many things. Then the "I'm sorry, you're right" or the best is when it argues with you, even though it's wrong. It take more time verifying the info it gives you than if you just researched it yourself.

Then you have students using it to write basic papers, which is blantently obvious. And I laughed yesterday when I received an AI-written email from a student (who happens to be a student that has submitted 2 AI-written papers already this Summer term). If it weren't a 6-week class, I'd have referred the student to Dean's Office for academic dishonesty already. I have a feeling that student won't make it very far.

Or, maybe Claude IS a student in my class...

It's a real paradox. If you have deep knowledge already, then it's like a team of young analysts doing some basic research for you. I remember that they made mistakes and half of my job was educating them and rerouting them. Claude is the same, EXCEPT that those young users are not getting any guidance and can't tell a bad answer from a good one.

It's most useful when you interrogate, sometimes down several levels. The speed at which it can read through reams of articles and data is incredibly helpful. Also, as a solo practitioner, it's a wonderful reviewer, almost always making useful suggestions. OTOH, when it offers to write something for me and I let it, my rewrite takes more time than just doing it myself.

[JABCAT]

Originally Posted by dcstep

It's a real paradox. If you have deep knowledge already, then it's like a team of young analysts doing some basic research for you. I remember that they made mistakes and half of my job was educating them and rerouting them. Claude is the same, EXCEPT that those young users are not getting any guidance and can't tell a bad answer from a good one.

It's most useful when you interrogate, sometimes down several levels. The speed at which it can read through reams of articles and data is incredibly helpful. Also, as a solo practitioner, it's a wonderful reviewer, almost always making useful suggestions. OTOH, when it offers to write something for me and I let it, my rewrite takes more time than just doing it myself.

I do an exercise with my intro class where I ask them to find something like prevelance of type 2 diabetes & then I call on a hanful of student to tell us what they found. 9.9/10 they all have, not only the wrong answer, but different answers from each other.

I don't have to worry about as much with my upper-division students because they have in-class written exams & in-person practical evaluations. They also understand as future doctors, physical therapists, and occupational therapists, they need to know what they're doing & how to evaluate a patient, etc.

It's the ones coming right out of high school that have zero critical thinking skills, can't write, can't do math (even though the got an A in calculus ), and can't discern accurate information from inaccurate information.

[louielouie]

I work with a lot of Gen X'ers who believe roles below Director will be replaced by AI in 5 years. I have doubts due to all the resource limitations of data centers, and energy consumption on top of pc components.

Also at some point, Anthropic and all these other companies are going to want returns on their investments and start heavily increasing their pricing. Curious where things will go in the next couple years as I personally expect it's not where many execs think it will.

[allinon72]

There is an accelerated arms race to be the "preferred" model, but eventually the investment will have to be monetized. Consolidation will happen somewhere in between, but the rapid timeline is the concerning part. Countries are racing to be the one to house all this, not just for economic but national security concerns. We may be paving the way for our own accelerated destruction, all before they even allow the public to touch the "real" AI behind the scenes.

[dcstep]

Originally Posted by louielouie

I work with a lot of Gen X'ers who believe roles below Director will be replaced by AI in 5 years. I have doubts due to all the resource limitations of data centers, and energy consumption on top of pc components.

Also at some point, Anthropic and all these other companies are going to want returns on their investments and start heavily increasing their pricing. Curious where things will go in the next couple years as I personally expect it's not where many execs think it will.

There will be huge fallout, not unlike what followed fiber optic overbuilding. There'll be a few winners and lots of losers. There will soon be subscriptions at reasonable prices that turn into subscriptions at unreasonable prices. Today, you can research any subject with the same AI tool, but you'll need twenty in three to five years to cover a reasonable range of questions.

[louielouie]

Originally Posted by dcstep

There will be huge fallout, not unlike what followed fiber optic overbuilding. There'll be a few winners and lots of losers. There will soon be subscriptions at reasonable prices that turn into subscriptions at unreasonable prices. Today, you can research any subject with the same AI tool, but you'll need twenty in three to five years to cover a reasonable range of questions.

Claude was going x15 or x30 in price for their p code but rolled back the day after announcing. We're already getting a glimpse of what's in store...

[Ohio Enthusiast]

Originally Posted by dcstep

There will soon be subscriptions at reasonable prices that turn into subscriptions at unreasonable prices.

We're already there. GitHub Copilot used in my company moved to token based billing June 1st. Cost to the company went 10x, and people who exceeded the company set limit (some within less than a week) were given a very low daily budget.

For the original question, I'm guessing there will be an equilibrium eventually, similar to the work-from-home vs. return-to-office post COVID. Some companies will continue to use people with AI tools seen as a tool; other companies will go all in on AI and use it to replace people. Both categories will have companies that succeed and companies that fail. I doubt one can predict which companies will be which, though.