AI technical subjects - very good "tutorial" learning

Oct 03, 2026 Last reply: 1 day ago 20 Replies
Jump to replies (20)

Trying to learn in geology about "hydrothermal mineral deposition" (the famous metalliferous ore lodes in Cornwall are such), I went to AI.



My impression is that what happened is significant, and I offer this comment to anyone who needs to learn about a specialisation which is new to them... That "mission" took half a day - after the subject had remained "parked" for more than two years with no way I could find to start making progress on it...



This is "interesting comment", or is so self-evident there was no pupose in expounding this point?


vvvvvvvvvvvvvvvv


Point I wish I had made about AI involved in a "tutorial"...



I'm approaching the AI (using "DeepSeek") to ask about geology. The AI has about equal ability in other technical subjects. I am by training a "physical metallurgist" - that various elements are available is a given, and the challenge is to engineer properties of the alloy to meet needs. That comes out of starting in the steel industry. Mining and resource extraction is about "extractive metallurgy". In effect, what gives those elements which the physical metallurgist takes to develop alloys. "Extractive metallurgy" goes from zero up to the interface at which the "physical metallurgist" buys the elements to formulate alloys... I am a scientist already. In this "tutorial" I made a lot of use of hypothesising how the "hydrothermal mineralisation" might work, taking fundamental physical phenomena I know from physical metallurgy. The AI is "knows" the physical metallurgy theory too (along with a vast amount of other subjects and knowledge) and can explain in what ways my conjectures work and in what ways my conjectures deviate from what is. The point is that the learning-rate by this method of conjecturing / hypothesising and offering these for critique is a very fast way of learning - if you have an entity which can respond describing how your conjectures do or do not work. An AI is advantaged by its vast data repository, and ability to relate it to the current "mission". While laid-up with a virus, and keeping very warm to mitigate its symptoms, I spend half a day doing this "learning hydrothermal mineralisation" mission...



What I am saying is that, whatever prior endeavours you have made, the AI is likely to be able to work with those to help you learn about a field of knowledge which is new to you. This relates to a fundamental point which seemingly the majority in society certainly here in Britain "get wrong". This is what I think of as "sideways learning". You can rapidly infill about a "new" topic as far as the furthest extent you have progressed in the subject you have specialised in. "Extent" is how far you have pushed to the out edges of knowledge and started to push into the at-the-time unknown... It follows as a corollary that you should push as far as you can into a specialisation - because you can "infill sideways" to the level of any outward push you have made in some subject you have committed to. Overcoming exhaustion; overcoming challenges; putting in time when you could be off in leisure activities; etc. In Britain "they" look for exact-match - have you done precisely "x" previously? This is actually the incorrect model certainly for technical subjects and I suspect many others.



Anyway - whatever you know already, the AI can help you apply that to understanding a "new" field of knowledge... I hope this is helpful comment.


<< snipped>>

Richard , please don't be offended but it looks to me like you just used a whole lot of words to say "AI can help you learn s*it that you don't know." I usually read your posts and occasionally your stuff is technically over my head , but this one just confused me .

I guess it depends on the AI. Grok tends to be stodgy and defends the company line, but generally does some research before answering. Even when wrong it can usually defend its wrongness. Gemini tries to be affable, but is related to that Star Trek guy. You know the one that built Data... Often Wrong.

This is what I think of as "sideways learning". You can rapidly infill about a "new" topic as far as the furthest extent you have progressed in the subject you have specialised in.

------------------------------------- That was explicitly the point of a BS in Chemistry. The field is so vast that 4 years wasn't enough to become functionally proficient in any area of it, but we learned enough to follow the explanations at whatever job we were hired for, and had the proven aptitude and persistence to dig in further.

Bog iron ore accumulates by a related process:

formatting link
"Europeans developed iron smelting from bog iron during the Pre-Roman Iron Age of the 5th/4th–1st centuries BCE, and most iron of the Viking Age (late first millennium CE) was smelted from bog iron. Humans can process bog iron with limited technology, since it does not have to be molten to remove many impurities."

And if you don't have a good flyshit filter it's hard to sort the flyshit from the pepper. I've been using ChatGPT on a small engine drivetrain project and it's been lying through it's teeth!!!!! It can't even give me factual basic engine specs - or bolt sizes etc.

And if you don't have a good flyshit filter it's hard to sort the flyshit from the pepper. I've been using ChatGPT on a small engine drivetrain project and it's been lying through it's teeth!!!!! It can't even give me factual basic engine specs - or bolt sizes etc.

------------------------------------

Maybe it has been optimized for political persuasion.

The AI might be good if it has a large dataset to work with but I'm finding it's quite lacking when the amount of data is minimal such as in the case of paper mallets, mallets made from tightly wound paper. I remembered being told it was the Boston Ma OSHA office that did the inspection that ultimately led to the closure of the company so that narrows the search area and the paper related inspections on the online OSHA database, which is currently down, to about 100 events 1972 to

1975. The subject is quite obscure and I'm finding I'm the source of some of the information and I've exchanged emails with the other couple of people that get referenced and some of the information is incorrect but it's out there. The AI suggested a company that likely made them in the 1970s but when checked the company only made rawhide products and closed in the 1950s, must do better I guess. I'll keep relying on my own search abilities and see what more I can find.

I have found if I ask "it" (I don't use chatgpt) if its 100% sure based on verified sources, and to provide the sources I get better results, although sometimes Gemini will say stupid s*it like, "based on various internet sources and forum posts...."

I know Gemini will absolutely lie to you about the law, and claim it gathered its lies from general scuttlebutt when called out.

Examples:

  1. In Arizona a Jeep does not require mud flaps even if bumper height is altered as long as fenders cover the tires (interpreted as the tires do not stick out pas the fender. Gemini will tell you it does, but when cross examined extensively say there is a loophole because the exact statute just says trucks...

  1. Found similar lies about the transporting guns and ammo law for California. Separate lock containers in the tack on effect circulated by wannabee experts and it is what I got as a first result, but when forced to show the actual law, it just said unloaded in a locked container and there was no provision that ammo couldn't be in the same container or even int he magazine as long as the magazine was out of the gun, and there was no round in the chamber. To be fair there are a lot of wannabee internet experts spreading that false law.

I have found similar results on all topics. Not just law. You have to cross examine the AI(LLM). I did just have a reasonable conversation with the Amazon LLM embedded in AWS regarding practicalities and legalities of mass mailing that did not seem to be riddled with populist inaccuracies like Gemini or corporate benefit bias like Grok. It will only talk to you about AWS related topics, but that can get to be pretty wide.

Sorry. I hope other posts I write prove more useful. Best wishes

I have known the AI make a mathematical error once. Fortunately I as using the AI to double-check my own calculations, so I very carefully reworked how I expressed my calculation so its logic was very clear - as a group of functions (programmed in Lisp). Which I supplied up to the AI - which checked them, ran the calculation in stages with intermediate values - and at the end declared my solution correct. For what it's worth - that was a calculation on roll-crushers for mineral processing.

But in general I find the AI very useful. I do sometimes have to prompt it. Say I am surprised about the answer and asking the AI to reconsider it. in fact, the AI really "comes alive" if you do that. Well, that si the case for the "DeepSeek" AI...

On further consideration - your explanation is better...

Sorry. I hope other posts I write prove more useful. Best wishes

--------------------------------------

To me it looked like you had slipped into the style of official technical jargon, like this.

formatting link
If I read a lot of it I have to remind myself not to imitate it.

I missed the point, with trying to explain the context.

The AI was enabling me to use my knowledge in another field and be helped in using it as a framework to understand another field.

So I was using metallurgy as I have gained with steels, castings, welding, etc., and the AI was identifying likenesses in geology as a path to show me the mechanisms of deposition of ores which we now mine.

I wanted to say that AI seems good at tutoring a person in a way which has been difficult before.

A human tutor would have to know both the subject you already know and the subject you want to learn. With the "etcetera's" the chance of that happening for you is disappearingly small.

The AI is (currently) a "Large Language Model", "LLM" - does that mean that it is actually doing "word association"? Even doing that, it can be useful. You can get a 1-to-1 tutorial, taking what you know already and using it to give a big starting inroad into what you want to know about...

Which is my little local "discovery" I wanted to share...

Best wishes

So I was using metallurgy as I have gained with steels, castings, welding, etc., and the AI was identifying likenesses in geology as a path to show me the mechanisms of deposition of ores which we now mine.

----------------------------------------- That would benefit from understanding oxidation states, reaction thermodynamics and solubility of metallic compounds.

formatting link
As with iron and carbon, when a melt cools dissolved impurities tend to become less soluble and crystallize out, forming grains of ore within the matrix of pure(r) rock. Slower cooling enables larger grains.

welding, etc., and the AI was identifying likenesses in geology as a path to show me the mechanisms of deposition of ores which we now mine.

----------------------------------------- That would benefit from understanding oxidation states, reaction thermodynamics and solubility of metallic compounds.

---------------------------------------------

Section 2.7 gives an overview of methods.

formatting link
The Betts process is an example for lead purification.
formatting link
An example of recovery by solubility is converting slightly soluble lead sulfate from old batteries into nearly insoluble lead carbonate by treating it with washing soda, sodium carbonate. Lead carbonate is easier to reduce to lead metal and the sulfate doesn't escape as gaseous pollution.

Even silicates can dissolve and be transported by water, for example Australian opal.

formatting link
? Diatoms are tiny marine organisms that extracted dissolved silica from the water to form their shells. Deposits of them are used in swimming pool filters. I made sodium silicate solution by dissolving pool filter powder in caustic soda and use it to solidify the cut ends of woven fiber sheaths of high temperature thermocouples. It's also the glue in wood stove gasket cement.
formatting link

Yes that similarity

  • "segregation" in castings
  • forming metalliferous-enriched "rejected" fluids in a solidifying magma pluton is one of the likenesses the AI was able to work with and tutor me.

Sierra Nevada in Southern Spain - there are amazing springs on the South slope where the water comes out clear but tastes strongly of iron / minerals, and if stored in contact with air orange.

"The sulphide belt" of Southern Iberia sweeps past/through this area...

Diatomaceous earth also gives us "dynamite". The liquid nitroglycerine produced in vessels on top of sand-dunes and trickling down a lead-lined launder becomes fairly safe to handle once it has been mixed into diatomaceous earth.

Yes that similarity

  • "segregation" in castings
  • forming metalliferous-enriched "rejected" fluids in a solidifying magma pluton is one of the likenesses the AI was able to work with and tutor me.

---------------------------------------

formatting link
Also:
formatting link
The practical application is determining the weight of individual molecules by measuring the freezing point depression of a known weight, which gives the number of individual molecules. In many cases it's not obvious if molecules bind into pairs or not.

Join the Discussion

Have something to add? Share your thoughts — no account required.

Didn't find your answer?

Ask the community — no account required