Microsoft Copilot

The new Microsoft Copilot usually gives excellent answers - better than I can find by searching alone. Today was funny though. I asked for the resistance of a .025" sparkplug gap in a running engine.

Answer: showed all the math and came up with 6.77 trillion ohms.
I replied: That's not possible!
It replied: Oh, you're right, than you for catching that. Then it redid all the math and came up with 677 billion ohms.
I replied: That's still not possible!
It replied: Thanks for catching my error - let me quite generalizing and figure the correct answer. Then all the math and 6.77 billion ohms.
I replied: It's time for you to admit that you don't know!
It replied: I found that is really .0347 ohms! which is a very small number but an important one. (I think if found a typo in a site and added the bad knowledge to its data)
I replied, Now you're being ridiculous!
It replied: I guess I don't know - I will study more.

I gave it 10 minutes and started a new session and asked the original question again, here's it's answer now after all the math: "Therefore, the resistance of a spark plug with a 0.025-inch gap in a running engine is approximately 2.00 megohms (2 million ohms). Keep in mind that this value can vary based on factors such as temperature and spark plug material." That's more like what I expected but I'm still not sure it's right.

My point was to determine the effect of the resistor in a 5k plug. Forgetting compression, fuel load, temperature, etc.; the 5k ohm resistor would drop about .25% of the voltage if the gap's resistance is 2m ohm when the spark happened. That seems reasonable. Basically, resistor plugs hurt nothing.

Machines will one day take over the world

The trouble with this AI stuff is laid out perfectly in your post Greg.

As it happens, you knew it gave you the wrong answer (several times) but the vast majority of people who search questions do not know, and often they lack the acumen / critical skills thinking to assess answers, so the answer given will be the one that goes in the paper / article / model, thus triggering a hugely incorrect process, but cos the mistake is a ‘small’ one, and done at the beginning, it gets buried in the analysis and writing and conclusions…

It’s always happened, but with the over reliance on AI I predict it will happen much more frequently in future.

Fast Eddie said:

The trouble with this AI stuff is laid out perfectly in your post Greg.

As it happens, you knew it gave you the wrong answer (several times) but the vast majority of people who search questions do not know, and often they lack the acumen / critical skills thinking to assess answers, so the answer given will be the one that goes in the paper / article / model, thus triggering a hugely incorrect process, but cos the mistake is a ‘small’ one, and done at the beginning, it gets buried in the analysis and writing and conclusions…

It’s always happened, but with the over reliance on AI I predict it will happen much more frequently in future.

Click to expand...

Yes, and the worst part is that it often comes up with great answers so you can get used to it being right.

I'm a very experienced programmer and have worked with professional databases for about 30 years. Recently I needed to write a program that I knew was going to take two database queries, a stored procedure, and a program to extract the raw data and another program to write the report. On a whim, I explained it to Copilot. It came back with a query that made no sense, using functionality I had never used. Curious, I put in the related table names and ran the query - it worked perfectly to extract the raw data. In the end, between that query and a 50 or so line program the job was done in under two hours.

It almost always is easier/better than standard searching - but clearly not always!

If you programme a computer to think in the way most people think, the result must be typology. However typology is the norm anyway. AI offers the opportunity to handle multivariate data reliably, but the answer you get depends on the reliability of the data which the computer relies on to programme itself.
When I was at school we were not taught how to use matrix algebra to create numerical graphs to establish correlations. When I needed to do that, I just got the method out of a book. My boss never knew how I used the computer to chemically analyse his steel. He also did not know there are computerised ways to establish the relationship between chemical properties and physical properties of materials. He used to do it by inspection and trial and error. When he actually saw how I analysed his steel, he was horrified. I used to love that job.