Job Security

Larry Weingarten

Hi, I just ran across this visual article about the impact AI could have on the workforce. I'm glad to be one of those who carries tools. https://www.visualcapitalist.com/cp/which-jobs-artificial-intelligence-gpt-impact/

Yours, Larry

hot_rod

I wonder how they came up with the lists?
Dredge operators?

PC7060

@Larry Weingarten I read a similar article about how AI could benefit the middle class earners. The article explained that the introduction of the personal computer had the impact of empowering highly skilled groups at the cost of lower skilled groups such as secretaries and typists. The article went on to describe how AI will enable lower skilled groups to perform higher level background analysis tasks opening more opportunities thereby leveling the earning potential. 

My thought was the use of “leveling” was naive; suppressing pay across a wider range is more accurate. 

Larry Weingarten

Part of what motivated me to post this is the thought that if people who are looking for their career path thought about how AI might impact them, they might be more likely to consider the trades, which could be a very good thing. @hot_rod I don't know about dredge operators ... might be a rewarding job! Other people may muck things up, but you get to un-muck things

Yours, Larry

dko

While AI won't take over physical labor any time soon, it will dictate how you will do that labor.
Hydronic design, radiant design, piping design, etc. All current computing will be outdated, architects and engineers that implement AI will take over. Job description would probably just be AI operator at that point and later AI supervisor.

SlamDunk

With my luck, it will also decide how much I should be paid.

FStephenMasek

Ever heard of "garbage in, garbage out?" It is just a computer program with a database, nothing new.

JakeCK

Don't be so confident in the inability of AI to do manual labor.

EdTheHeaterMan

I know a fellow named Emmanuel who does Labor an my home every two weeks, when they mow the lawn. Is that the same thing?

SlamDunk

JakeCK said:

Don't be so confident in the inability of AI to do manual labor.

There are some jobs that I do that I wouldn't mind a robot doing.

delta T

Well, I just played around with ChatGPT asking it about boilers and combustion settings. Steadily got more technical with it to see how well it could answer questions and eventually got this gem:

"If the COAF level is significantly lower than the target or approaches zero, you may need to increase the air supply slightly to maintain a safe level of combustion."

Confidently incorrect. I'm not too worried. Yet.

delta T

Okay so just to add to my comment above. I got back to some simpler questions that *should* be easy for an AI to answer. The best one so far was: "How many ft/sec is 5 gpm through a 3/4" pipe?"

Boy did it screw up hahaha, here is its response:

"To determine the velocity of 5 gallons per minute (gpm) flowing through a 3/4" pipe, you can use the following formula:

Velocity (ft/sec) = (Flow rate in gpm) / [(Pipe diameter in inches)^2] * 0.0408

Let's calculate it for your case:

Flow rate = 5 gpm
Pipe diameter = 3/4" = 0.75 inches

Velocity = (5) / [(0.75)^2] * 0.0408
Velocity ≈ 6.38 ft/sec

Therefore, the velocity of 5 gpm flowing through a 3/4" pipe is approximately 6.38 ft/sec."

Now, on the face of it it looks great, looks like it got the right formula and did the math for you. But.....the formula is incorrect, kind of close but not quite right, and even then, if you do the math that it displays above, it doesn't work out to the answer it gives.

(5/(.75^2))*0.0408 = 0.363
and
5/((.75^2)*0.0408) = 0.023

The top equation should be correct according to its own notation though it could still be ambiguous, but I did both possible orders of operations to show that it is not a simple matter of arithmetical error. The answer it gives does not even match the math it used to explain it. The math is kind of correct, but its off by a factor 10. It uses 0.0408 instead of 0.408. Why? got me! Why does it not even do the math correctly? confidently gives an answer that doesn't even match its own incorrect logic!

And btw, engineering toolbox gives 3.63 ft/sec., which is actually correct.

I think a lot of people are really jumping the gun on how competent AI is in its present state. We tend to trust computers to, if nothing else, do math correctly. This just goes to show that what we seem to have really done is to simply add the ability for computers to confidently lie to us....

Larry Weingarten

Hi @delta T , This sounds like that Russian proverb "Don't trust, but verify." Thanks!

Yours, Larry

PC7060

delta T said:

I think a lot of people are really jumping the gun on how competent AI is in its present state. We tend to trust computers to, if nothing else, do math correctly. This just goes to show that what we seem to have really done is to simply add the ability for computers to confidently lie to us....

Just like a lot of the humans we deal with. They don’t  call it machine learning for nothing! 

fentonc

Large Language Models (LLMs) like ChatGPT and similar are qualitatively different than something that searches for keywords in a database. These models are trained on huge volumes of text (think 'the whole internet' and then some), and they statistically generate a probable sequence of words/characters - that's where you get the 'confidently wrong part' from. The model doesn't really know that it's doing math, it's just generating math-ish text in response. Imagine typing the phrase into google, reading the first 100 responses and then trying to generate 'something like that', without having any understanding of arithmetic. This works astonishingly well for what it's doing, and on tasks that really are "Generate some paragraphs about this topic, I don't really care if they're correct" it can do an amazing job. I would imagine the general level of spam/SEO garbage on the internet is going to skyrocket.

Even if it was correct on heating/plumbing topics (and it might get there), I suspect people will manage to ignore heating related advice from an LLM just as well as they ignore installation manuals, books and existing design software.

JakeCK

You should see it's responses for programming questions. 

Dave Carpentier

Yay.. more video podcasters giving life's little instructions based on their use of GPT-type research.
How could that ever go off the rails.

PC7060

JakeCK said:

You should see it's responses for programming questions. 

Sure would nice to have a code analyser to deal with the billions of lines of legacy code!