Comment by breck | Hacker News Reader

Comment by breck | Hacker News Reader

Parent Full thread

> individuals ‘dig their heels’ into protecting their own pet project/baby/solution/etc, mainly due to ego.

Guilty of this.

I worked for about a decade on a pet project to find a new family of computer languages designed for both humans and machines.

I did not think machines were close to mastering our languages, and new languages were needed.

I knew my approach was a long shot, but if I found a way to make it work the upside was huge.

Then LLMs happened. The possible upside of my approach dropped dramatically.

I have been trying to "rewire" my brain and re-purpose the neurons that evolved over a decade to keep turning my approach around from different perspectives. It is very hard.

It is easy to get a sapling to grow into a desired shape. It is much harder to reshape a fully grown tree. Just the physic of it.

To tie this back to the original article, if you model an individual's brain like Minsky's Society of Mind, you would have neural agents that create a circuit ("Institution") to solve a problem, and some of those agents focus on the task of preserving that circuit. Without those Institution preserving neurons, you would never keep the circuit going long enough to see through a contrarian idea. But the downside is that the organization will persist even when it is no longer a good bet.

Tbf, half the linguistics discipline thought that language's grammar was somehow hardcoded into our brain, which is clearly ridiculous if you look at how LLMs work, so you're not the only one who had misconceptions.

Perhaps you can turn your idea around slightly into finding a language that finds a balance between formality and universality, rather than computers and humans. Because even though computers now speak our language they do not use it in a logical way at all (arguably because we humans don't).

And while mathematics is very formal it has a lot of trouble expressing ideas from different branches that aren't as formal. Things like fuzzy logics have been created and many things like that but they are still very much on the formal side.

Perhaps you could even derive an academic language for a specific field, perhaps standardizing between synonymous constructions. You could even use LLMs to accelerate the process. Maybe LLMs are a good thing that makes your work easier!

> You could even use LLMs to accelerate the process. Maybe LLMs are a good thing that makes your work easier!

Oh I 100% agree. LLMs are amazing. Plenty of neural agents in my brain are on board. I use them everyday to work on problems in a way not possible before.

I think what I was trying to express is that a contrarian idea might require developing a large number of your own original solver brain circuits that are very dumb, always running, trying to brute force a path for your idea to work.

Later you can then develop new circuits that recognize there's now a better approach, but those solver circuits that you grew are still in your brain, occasionally still running (like sometimes when I wakeup in the morning), because that's what you trained them to do.

In other words, there's a risk to taking on a contrarian idea in that you have to build up lots of brain circuits that will stick around for life, even if your idea turns out to be wrong. I'm sure people have written about this more eloquently. I need to search more.

Ahh yeah I was trying to help you repurpose these circuits given the new information. But perhaps that's not possible.

It sounds very similar to what happens with love. In my experience, at least, when you love someone you build up these circuits that care about the other person and you cannot break them down, it seems. You can ignore them but then there's this part of your brain you're ignoring.

So perhaps you could say you were/are literally in love with the idea.

> when you love someone you build up these circuits that care about the other person and you cannot break them down, it seems.

Ha! My experience as well. Even after many years when you see that person again those circuits turn back on (and are very strong).

That is a bigger more important thing. But also, on the topic at hand, an interesting and probably strong analogy!

> ...clearly ridiculous if you look at how LLMs work

This is well off topic now, but this doesn't follow at all. LLMs aren't brains and don't even resemble them that closely. LLMs demonstrate that it's possible to learn grammar from scratch, not that humans actually do. I for one think it's pretty plausible that humans have a little bit of neural wetware-acceleration for syntax. In much the same way, it's possible to implement AES with just an ALU and memory operations, but your CPU probably has special hardware anyway.

Reply on news.ycombinator.com