Why digital long-form discussions belong on computers, not phones

[Josh_Ketry]

@DannyM

And then something beautiful started happening: these small communities began federating . FidoNet, UseNet, they created ways for BBSes to share messages between communities while maintaining their local identity . You could have a conversation that spanned multiple communities, but it was still structured, still threaded, still comprehensible . This was the dream, right? Scale without sacrificing the human elements that made communication work

This sounds an awful lot like a guild of guilds, no? I had the dunbar number in mind with the idea when it started. Small circles connected to big ones.

[Josh_Ketry]

@DannyM

And here’s the thing that’ll make you laugh-cry: LLMs are literally just probability machines. They don’t understand anything. They’re playing an incredibly sophisticated game of “what word typically comes next?” based on patterns in their training data

100%
LLMs are just math. I would counter by saying that they are useful for one thing: haboring and organizing what David Deutsch calls “exiting knowledge.”

So for example if you had a jiu jitsu problem - say you were stuck in a choke - you could use an LLM to find existing answers and knowledge that humanity already has. BUT if you wanted to come up with a new, creative answer, AI could never do this. AI can only regurgitate or recombinate using math. As you said, it is just predicting the next word based on the data it has. Where did that data come from? It came from people. We create that knowledge.

So let’s say that Einstein’s theory of relativity was never invented yet. AI could never come up with it on its own. This kind of knowledge David Deutch calls “new knowledge” and it can only come from “people” (of which he includes humans). Creating new knowledge is not probability. It is instead created by conjecture, trial and error, and more conjecture. And people are very very good at imagining, creating, and using conjecture.

There are a LOT of great videos by David on the internet on knowledge, knowledge creation, AI, AGI, etc. He was the father of quantumn computation, the main proponant of the many worlds theory of quantum mechanics (which btw is the most reliable theory we have in all of science next to entropy), but in my opinion his best work is in epistemology and knowledge creation. This was 2019 before the AI boom when he said this @DannyM :

Why has AGI not been created yet?

[Josh_Ketry]

@gino @DannyM

My notes on your article:

1. The memory problem you outlined by “chunking” is something I have studied as well. There are two ways to remember things bigger than the typical 6-10 characters people can memorize. The first is chunking and the second is with a story. Here is a fun thing I use when I describe memory:

’If I told you to remember 50 letters in a row, most peope would forget after 7 or 8. For example if I said to repeat these back to me:

LSKENFJKALDJENAMPQQWSNRQFSKDLKKWMORJSFMWKKLOAJI

That is 50 letters, and you would not get far past the first K.

But if I said to you: THE LITTLE BOY AND LITTLE GIRL WENT TO THE STORE TO GET A GALLON OF MILK

That is more than 50 letters and you could easily remember it - because of both chunking and the story painting a picture in your mind. There is an amazing TED talk on this here, where journalist Joshua Foer is covering the memory world championships, and then learns the techniques and comes back the next year and WINS first place in it:

Feats of memory anyone can do | Joshua Foer

2. The Dunbar number is really about trust. That can be scaled in my opinion the same way Amazon and Ebay do it - decentralized. But I do think we need small circles of trust connected to big ones, with a base code of conduct we have for both. I also think - and both Danny and Gino know I expressed this - of having a way to mix IRL and URL (in real life with digital URL life). This was a topic at a Network State convention Gino and I attended once. By mixing them we can lock AI and bots out of our groups and systems

3. @DannyM you criticize text based communications pretty hard here, but your entire thesis (which I mostly agree with) is text based - but long form. I also think the criticism of AI is too sharp, as I believe that any powerful too can be used for both good and bad. Nuclear, electricity, flight, etc. Dr. Rosenberg has been using a form of conversational collective intelligence enhanced by AI and I think the results are showing that we can use both, together, as long as the systems are trustworthy. I posted this in another thread but if you missed it his methodology here is eye opening.
   https://www.youtube.com/watch?v=FrY0Gr5p144

[DannyM]

Yep, I think that eventually I want to have an even better version of it, but I’m ok with publishing it

@Josh_Ketry

So for example if you had a jiu jitsu problem - say you were stuck in a choke - you could use an LLM to find existing answers and knowledge that humanity already has. BUT if you wanted to come up with a new, creative answer, AI could never do this. AI can only regurgitate or recombinate using math. As you said, it is just predicting the next word based on the data it has. Where did that data come from? It came from people. We create that knowledge.

Yep, true

@Josh_Ketry

@DannyM you criticize text based communications pretty hard here, but your entire thesis (which I mostly agree with) is text based - but long form.

I mostly criticize text based communication on devices with limited context. (such as phones)

@Josh_Ketry

I also think the criticism of AI is too sharp, as I believe that any powerful too can be used for both good and bad. Nuclear, electricity, flight, etc. Dr. Rosenberg has been using a form of conversational collective intelligence enhanced by AI and I think the results are showing that we can use both, together, as long as the systems are trustworthy. I posted this in another thread but if you missed it his methodology here is eye opening.

I tend to agree with you, however I haven’t found a case yet where AI beats out someone who actually knows what they’re doing in that field ^^

[Josh_Ketry]

You should see the tests with doctors diagnosing patients. Individual doctors, vs AI doctors, vs swarms. Swarms win, but AI doctors are second. Swarms beat individual doctors by 30%.

[DannyM]

I’ll be honest I’m skeptical of an article claiming that an “AI doctor” is better than a human one.

Something I would believe is that machine vision trained specifically on one problem (such as say cancer screening) can have a higher correct detection rate than a human doctor, but at the same time it will also have a higher false positive rate, and a more devastating false negative rate. That’s a far cry from a general AI doctor tho.

I believe the swarm tho. Of course multiple minds are better than one!

[Josh_Ketry]

It wasn’t just an article, it was a scientific paper that compared results. It did not do false positive rates and was really not the best structured experiment for comparing with AI. But it was interesting nonetheless.

[Josh_Ketry]

@Josh_Ketry

do false positive rates and was

Danny could you email me your long form article so we can publish it please? Complete with links.

jketry@pearlstreetgrill.com

[DannyM]

@Josh_Ketry

It wasn’t just an article, it was a scientific paper that compared results. It did not do false positive rates and was really not the best structured experiment for comparing with AI. But it was interesting nonetheless.

Well, if it doesn’t take into account false positives and false negatives, then the only reason for a group of scientists to publish that paper is to get further funding because ai sells well, haha

@Josh_Ketry

Danny could you email me your long form article so we can publish it please? Complete with links.

of course!

[Josh_Ketry]

Hey Danny

It publishes tomorrow.

The footnotes didn’t transfer too well but I kept them anyway. I also know you like you privacy so I used your name from on here - Danny the Cyber Guy. I can change it to whatever you want.

After it publishes I can edit out the footnotes if they are still too weird.

I chose your title:

**Why is Everyone so Angry Online Today?

subtitle:**
Danny The Cyber Guy’s manifesto on why digital long-form discussions belong on computers, not phones, and specifically on high trust networks like Eve.

The rest should be verbatim.

Set for 7am tomorrow

[Josh_Ketry]

@DannyM your post is live and getting good traction and reception. Thanks for letting the Society of Problem Solvers post it (you are part of the the society now, lol), and let me know if you want your real name on there and a contact. Thanks for being part of this and sharing that awesome piece! It has almost 1000 reads in one day.

Why is Everyone so Angry Online Today?

[DannyM]

My next long form article (which I’ll probably finish by beginning of february) is about why notifications as they’re implemented today are terrible, and how there’s a simple solution that everyone is missing

[Josh_Ketry]

Everything look ok? Want me to change anything?

[DannyM]

looks good to me, only thing I would want to change is that I wish the references were rendered better, as they are here, but whatever substack doesn’t support it, it’s alright