When we learn, we do many mistakes in our approach and adapt our strategy. And when we succeed, we tend to post only the final good version, not mentioning how many fails we had.

And thus AI learns only on successes. It gets the wisdom of how to do things right, but not about what could go wrong.

I mostly use ChatGPT and Claude, and I have noticed that sometimes  to solve my niche question they propose code that just does not work. It looks like a reasonable piece of code (according to intuition formed by mainstream languages), but appears to be unreadable by compiler due to some undocumented language inconsistency. Several times I managed to find relevant questions on stackoverflow, but those questions had nonsense replies (where people instead of acknowledging there is a problem just say "this question has been asked, look here" and link to an irrelevant question, which uses similar concepts).

As soon as I hit such a problem in development while talking to a chat, the chat becomes useless. AI pushes back with "are you sure? it must work". It feels like at this moment AI inherits overconfidence of those gaslighting stack overflow incorrect answerers, and continues to push against any feedback, not being able to comprehend it is not right.

I am interested in cooperation/teamwork scores/metrics. Is there any research paper trying to map cooperation into any metric that is designed for single humans?

For example, what chess Elo rating would two 1000 rated humans achieve, if they are allowed to duscuss strategy during the game? How much higher IQ score of a pair of cooperating people is expected to be compared to the max of their individual scores? How much faster a team of people would solve computer puzzle game Witness?

Upd. How does a human+LLM score compare to single human score?

What if instead of cooperation and tools we use handicap. How much worse blindfolded chess players perform? 

The minimal change that in my opinion tilts the situation in the healthy direction (but is absolutely not economically profitable and thus unrealistic), is to limit the posting frequency. Can't make more than 1 publication per day/week. I bet most people would think just a bit more between they post, and they would not be able to spam content into twitter/reels etc. 

Limitation of internet exposure time is the complement idea that is already quite spread. Timers, minimalist phones. But the person is controlling exposure, something about themselves.

Controlling/limiting the production is something about others. People, who would most benefit from thinking more between talking, would not go to such a website. And that us why I said it is unrealistic. Although it is possible to create a social meme to "make it cool" to use a website with limitations, and promote that "feed becomes less saturated with slop".

Continuation of my previous quick take about teachers. 

Long ago I read a thing claiming that some types of anecdotes can be hard to remember because of their absurdity. Things that do not make sense are funny specifically because they do not match correctly into your knowledge. I think that is also why it is difficult to recollect confusions in education: at least one side of the conversation was having a wrong model, which cannot be remembered by just a natural association. It takes more space in memory to remember other person's incorrect image of the world.

I am trying to recollect as many confusions as i can. I had an mechanical engineering class (materials, geometry, kinematics, loads, surfaces). Over the semester we had to design a gearbox. In the middle of the semester I noticed 2d technical drawing, front and up views. The girl was apperently ready to be defending her design. But I stopped her. She had a fatal flaw, which is only apparent when you have 3d understanding what those drawings mean. She did not have that, even though drawing looked good in other aspects (clean lines, correct sizes, top and front allign). The teachers were so strict about those clean lines that the student optimised to make it look good, not to match common sense. The shaft, which goes through the whole thing, was intersecting a gear of another shaft. Matter was passing through matter.

The crazy thing is, I am 80% sure teachers in my first uni would/did not notice that (her handmade kinematics draft must have been approved, since she was on the second stage), because of how they used to rigor for small imperfections ("You chose the wrong smootheness for this surface, go remake") and to not watch the drawing for longer. 

Both students and teachers can have faults in everything, even in common sense. 

My teacher in Strength of materials also did not posess 3d imagination. When drawing a crystal structures, she only succeeded in flat images, failing immediately to draw tetrahedron and more complex projections. The students did not understand that she is drawing nonsense, and were just memorising, not questioning, really handicapping their understanding.

One of my teachers once made a silly mistake, so i thought. She forgot to place a minus in the exponent. So instead of 10^-11 there was 10^11. Knowing, that my classmates will blindly memorise it like this, I came to point that out. When I pointed that out, I was met with "I teach this formula like that for eleven years! Do you think I have been making a mistake for eleven years, or maybe you made a mistake?". Her argument was basically authority and tradition! She was not willing to discuss with proper arguments, not willing to see how her formula implies molecules of galactic scale. Even if I was the one that had the confusion, the teacher still has to let the student describe his thought process, to see where is the mistake.

I once found a mistake in the lab book of a PhD professor. I came to tell that he has all indices swapped on couple pages. Like, the speed of electron 1 was multiplied by the mass of electron 2. Couple pages later another mistake, which magically cancelled it out and the resulting formula was correct. If this is indeed a book's mistake, I knew most students will not understand, will just assume that they are too stupid to grasp and just go on, and thus the whole chapter is dangerous.

The teacher shouted at me. "You should have better went to the art school, like your mother..." (he overheard my private conversation with other student over break before) "... look at the next page and everything will be clear, because the final formula is correct. I can't handle that, every year some stupid students come with this question, learn to read first". The "every year" basically showed me that there is indeed a mistake in his book and not in my understanding.

(The teachers were indeed outrageously stupid in my first uni, so I went to get an education abroad. Why am I thinking about the past and about teaching? I am in Masters program, and my faculty proposed that I teach Math2 in the next semester, and that is quite a responsibility, since I value proper teaching so high)

1. Are you familiar with game Nomic? It has original ruleset with immutable/mutable ruled and new rules are decided by democracy. https://www.nomic.net/deadgames/Nomic-Game/Rules/first.html
2. If i have to prove my rule was not too hardcore, that means I demostrate the exact strategy that opponent has to use to overcome the puzzle. I do not get the full benefit of the rule, since it is "solved".
3. This test does not seem easy to automate. Who decides the win? Who resolves the rules' collision? Whatever algorithm does those two things will be the blind optimisation goal. You do not have to be smarter or more creative than opponent, if you can optimise for "judge resolving conflicts in your favor". 

Couple of my own confusions. During aerodynamics the lecturer made a drawing of a cube, and named it elementary particle. I, to fit it better into my knowledge base aquired from physics course, asked whether it is the same as elementary volume. Perfect particle is a point, but the image had three dimensions, was squeezed and rotated later into lectures. For some reason teacher failed to understand what i meant, said couple things about particles, but not related to volumes and concepts from physics at all, and asked "did i answer your question?". I had to say no, and since then the teacher hated me.

In my 5th grade during geometry we studied the concept of orthogonality. The teacher said (translating into english word for word) "line is orthogonal to plane if it is orthogonal to any line in that plane". In my language the word corresponding to "any" means "take one arbitrary thing and test", like in "bring me any apple". My teacher, i assume was used to english papers, where "any" and "every" can both mean "forall", failed to see why I opposed.

Teachers should remember their confusions.

During my own studies I have noticed a lot of times, that teachers do not understand the questions from student. And this student->teacher information propagation fails as frequently as the teacher->student, and that is teacher's fault, since student can be quite articulate and still not hear the correct answer. I always had an itch to answer myself, because my level of knowledge was closer to my classmates: I can feel where they have a gap, what the teacher omissed, or divergence in their thinking process (where the person could build a different model in his mind, which is still describable by the same words). But a teacher, having 20+years of experience, have their neural patterns "optimised". Everything that is not describing the phenomenon itself gets deleted. They would not be able to recollect what aspects caused them confusion when they started. Thus they fail to comprehend the student's question and fall back to mapping the question into What they think is a reasonable question (like if it was asked on a conference) and repeating the last paragraph of their lectire, sometimes being annoyed that "students don't pay attention". And students in this corrupted situation subconciously learn that "nothing is supposed to make sense here", and they just memorise the answer. I have overheard many times how students were using different mnemonics as an explanation of the concept instead of logic/visualisation/process/reasoning. Simplified example follows:

 - Why does the wing starts pitching up in this situation?

 - Because buttefly. (points into a graphic that resembles butterfly, and the talk finishes, since the first student is also used to not hearing a meaningful response. He will just pass the butterfly forward, memorising some associated paragraph of dull text and numbers to spit out on the exam, without actually ever visualising pressure, streamlines, forces directions, leverage, etc)

This semester I was teacher's assistant in Programming in C for first year students. I was expected to check the programs, find mistakes, explain them, repeat the concepts they failed to grasp at lectures, etc. That experience gave me one of the very good examples, where teachers fail to comprehend the questions. (I actually had hundreds examples over my life, but I was never fast to write the question-answer pair down, and they are difficult to recollect)

Couple students were having problems running their program. The teacher, while trying to help, assumes they have a problem in one of the following (out of order): compiler settings, installation, code grammar, antivirus, etc.

I, keeping in mind my motto (first sentence of this post), decided to talk to them more generally, trying to find the point of the first divergence. I found out they recieved their first PC just a month before uni started. They had no concept of "file" or "filesystems" or "folder", no concept of "saving". And thus any explanation or instruction involving these words did not have any effect on them. Every time they wrote a new program, they did not create a new file. They were overwriting the same "seminar1.c" over and over for the whole seminar. 

Some might say that those students did not have prerequisites, but that is not the point. The point is that teacher has to remember how a newly-exposed mind works, and then he would notice where confusion originates.

In simple chat conversations, where i want it to generate a javascript line of code, it gets stupid. But in other chats, where i raise more difficult topics and thus i explain more than ask, it seems to be quite smart.

gpt-5-1 is getting worse. Answering previous question again, ignoring the newer question. Even without entering the thinking mode.

And overall more stupid today.

https://chatgpt.com/share/69371729-f090-8004-b69b-aa9f25ca0b40 here is a recent, but not very bright example, since it did finish the task after all (it used to completely ignore the most recent task in other cases). After thinking it answered "Yes" again, even though second prompt does not contain a question.

I slightly incorrectly posed a question and did not explain what i meant by abundancy, but i feel that is not a problem, since chat has this behaviour in any conversation it uses deep-thinking mode.