Pessimism about Techniques in the Face of AI

General Memory Chat
21

While AI can help with routine tasks, it cannot replace human judgement. And to make a human judgement call, we need to rely on weighing various inputs, facts, assumptions, etc which come from our memories. Otherwise we would need to relearn every fact that contributes to a decision, every time we need to make a human-in-the-loop judgement call.

So I don’t believe AI replaces my need to use my brain, to make decisions about the multitude of situations I encounter throughout the day. Those decisions include where and how I allow AI to assist me and to what extent.

AI is still confidently wrong much of the time. Be careful committing to memory what AI claims to be factual.

The models are simply assessing what it interpreted as your request, makes its own probability matrix of training material to use in a response, and based on a numbers game, generates a most likely collection of stuff for the response. It then layers in various cookie cutter sentence structures to sound like it knows what it’s saying and understands the subject matter.

That workflow does not replace my ability to rely on my lifetime of unique experiences, memories, and decisions, including the concequences of both good and bad decisions of varying degrees.

Most importantly, AI has no responsibility to concequences of its responses. Humans carry the responsibility and bear the concequences of how they choose to use AI responses.

This makes my brain more valuable, not less.

4 Likes
22

This also means that AI is not generating new and novel content. It is smashing up and reiterating what it’s been trained to recall.

I am nature’s greatest miracle. No one else thinks like i do, solves problems like i do. No one else in the world can father my children. No one ever in history amd no one ever in the future. That makes me unique. And there’s value in my uniqueness.

This is paraphrased and butchered up from Og Mandino’s “10 scrolls”.

But I think it’s a solid argument to not be afraid of what AI promises to do.

1 Like
23

If you’re of the mindset that you no longer need to accomplish something because you don’t have to, a world of Ai and an eternity of time will not teach what you need to learn.

2 Likes
24

If it’s not generating new content, then how do you explain the recent breakthroughs in mathematics?

1 Like
25

Not for nothing but according to Matt Parker’s estimate they spent 40 million dollars to find a theoretical instance where the Navier-Stokes does not apply. While they technically solved this math mystery, the usefulness of its discovery is up for debate, as is how worthwhile expending all those resources is. To be clear, it’s a huge milestone in the field of AI, but I’ve heard that it’s not as important for mathematics.

26

I actually saw a funny video about this. An AI is asked how to help a MD cope with his patient load. The AI answer let some of them die which does solves the problem in a straight line.
Humans think in a variety of different ways which is why things like second brains work so well.
There is another example in that AI has solved loads of maths problems in the same way by straight line brute force. It does not make them better than humans. It just means they can straight up process more information faster.
The Enigma bombs used both of these examples in that a human made them look for the wrong answers first which made them faster at finding the right answer.

2 Likes
27

I think what a lot of people miss is that humans have an extraordinary ability to completely screw something up and laugh it off later. I like to think about all the minor inconveniences I have every day (guy cuts me off, line is long, co-worker is annoying) and think, “oh well, they’re only human”. I truly believe that the overwhelming amount of inefficiencies and inequalities in the world are cleared by this mindset. It greases the wheels of society, and everybody adopts it without thinking. Humans make mistakes, but humans also laugh it off together.

On the other hand, when AI makes a mistake that’s a problem to be investigated and prevented. Sure humans also strive to correct their behavior over time, but for AI, perfection is always the end goal. For example, when you make an AI play Go or Chess the goal is for it to eventually “beat” or “solve” the game itself, whereas the human’s goal is merely to beat their opponent. The humans compete by responding to each others strengths, weaknesses, and inherent limitations, and those same limitations are what turns pushing one another toward mastery into an infinitely enjoyable task. In contrast, the AI’s goal is to solve the whole thing as if it were one level in a puzzle game.

So just imagine if you were to look at all the minor inconveniences in life as problems to be solved. You start thinking that the long lineup at Starbucks could be optimized, the awkward pauses in a conversation could be ironed out, and that the intrusive thoughts in your head could eventually be replaced entirely. The next day you are faced with more of the same. I find that I mostly slip into this mindset in times when I’m alone and without someone to laugh with, and that it rarely brings me joy.

So in the end I think it’s a personal preference, although most would agree. You can choose to use the human supercomputer brain, which laughs at its own farts and runs on pancakes and love. Or the AI supercomputer brain, which promises to one day surpass the human brain that everybody already has, via the small cost of indiscriminately appropriating ever increasing quantities of knowledge, dead dinosaurs, and self-centered obsequiousness.

That’s just my two cents, but to answer your original question I think the significance of making learning techniques is that it’s a fun and interesting thing to do. It’s only truly insignificant when you see it as a means to an end. Learning in general is pretty stupid when you think about it. Every time you learn about one thing it means learning about several other things you never knew you never knew about, so what’s the point? People hear the phrase “Ignorance is bliss” and their first response is “How interesting! Tell me more!”.

So what you have to do doesn’t really change much. Human intelligence being rendered insignificant is a tale as old as time. Death has been doing it for centuries with supreme efficiency. AI is just a new and more accessible method to attain this, one which offers a tickle for each of your fancies as the water warms to a boil. I’d rather jump out the pot without thinking or knowing where I’m landing, seeing everything I can on the way down. But to each their own.

My advice is that if you think you studied the wrong field then switch fields. If you still want to follow your dream then do it. What you learned doing the “wrong” thing will probably end up helping you in either case. Just don’t go treading new waters thinking that they will somehow grant you the entire ocean.

1 Like
28

You also have to be aware that AI companies are pushing really hard and fast to make their models seem great and useful. We are very brainwashed by marketing.

For example they took back about 40 of the mathematical proofs they had published because the models found errors in them… You’d think that if they were honest they would have checked BEFORE publishing them.

And the warnings of these companies that AI is going to destroy/replace humanity reek of hypocrisy when the very people issuing the warnings are the one continuing to push. It feels more like an attempt to avoid responsibility for what they’re creating - they will be able to say “we warned you”… But they’re still the very people continuing to pushing models that haven’t hold most promises and keep losing lots of money.

3 Likes
29

What I’m noticing about the practical use of AI in business engineering is that the processes are looking more and more like human processes. The agenetic swarms are often managed by processes that look very much like BPM. You might consider looking at how human intelligence and memory apply to effective agent. I’m not sure where that would lead you but some of the recent innovations are in how an LLM is managed and wrapped vs pure frontier model impoverishments. After all LLMs aren’t known for their effective short term memory.