sofiechan home

adamjesionowski received

in thread "American Intelligence":

Yes, extrapolating from software tasks is questionable. Software maps well to pure language, has a gigantic corpus of not just raw tokens but git histories with descriptions of bugs and test suites, most applications have tight verification loops, are high... 3w ago received

Yes, extrapolating f received

This board is 25% AI by volume, you can safely assume posters here are familiar with LLMs.... 3w ago received

This board is 25% AI received

If so, this means we cannot treat these things as "intelligences" on clear exponential trajectory, as is so often asserted. Rather they are indeed computer programs with easily understandable macro-mechanisms and less clear micro-mechanisms. We can grasp a... 3w ago received

If so, this means we received

Why a particular element in an particular embedding that you are multiplying is, say, 0.003 is not at all easy to understand and the situation only gets worse as you progress through a model.... 3w ago received

Why a particular ele received

I've attached the forward pass of DeepSeek V3.2. An AR transformer is going to produce predictions over token space. This is, concretely, what the system does.... 3w ago received

I've attached the fo received

Someone get Descartes on the line, he's going to want to hear this one!... 2w ago received

Someone get Descarte received

When have I said a program cannot do these things? That a machine can produce novel methods of solving a problem was demonstrated 70 years ago. I am saying something very simple:... 2w ago received

When have I said a p received

It is also incorrect. An LLM trained on (source text, translated text) pairs will not act like a persona. A base model has no stable persona but acts more like a superposition of personas, which is pretty much what you'd expect from trying to predict all k... 2w ago received

It is also incorrect received

in thread "Why AGI Will Not Happen":

Comparisons to organisms fail here, both in the original essay and the replies. Biological systems are of vastly different character than computers. "Intelligence" has never been easy to define as an extensive quantity for good reason. The best we've got i... 5mo ago received

Comparisons to organ received

Cope for what, exactly? Be explicit.... 5mo ago received

Cope for what, exact received

Austin Sofiechan Meetup
If you're in Austin and on here, let's get together! Shadow networks must be formed, voting cabals, there is power here for those who seek it... or we can just get coffee.... posted 9mo ago with 5 replies received

Austin Sofiechan Mee received

Uncertain as of yet. Throwing out next Wednesday at 6 at Holdout Brewing. I didn't get a response to my last message to >>4288, likely didn't see my reply. 9mo ago received

Uncertain as of yet. received

Bumping as I've started https://machinetools.club/ and we're meeting Nov 19th starting at 6: https://partiful.com/e/QnV9m5dzosZnGnl5dHOK 6mo ago received

Bumping as I've star received

The Culture and Function of the X Button
I sez to the Monarch, I sez to him: Voting buttons are gay and Reddit, they don't belong on something in the lineage of chan culture. And he simply loves his demos too much, wants to give them power and freedom in shaping the culture. Many a king toppled f... posted 8mo ago with 17 replies received

The Culture and Func received

Between LessWrong, the place with strictly enforced decorum rules, and 4chan, the place where bad-faith discussion is the norm, which site produced more healthy family men per capita? Despite picking from a similar pool of high IQ men, the answer is clearl... 8mo ago received

Between LessWrong, t received

I would encourage a trial run of this. We've all been trained to hunt for likes and high grades, I think it's much more freeing to not have these visible. The empirical change on voting patterns would also be interesting, I'd expect you'd get better signal... 8mo ago received

I would encourage a received

in thread "The Sofiechan Index Page Question":

The most useful feature for me would be to jump back into a thread from the last post that I read, plus a count of the number of replies that have happened since I last opened the thread. I only use the previews of the post as a way to jump to the end of t... 10mo ago received

The most useful feat received

It's not so bad just for a button that lets me jump to last read, just (user_id, thread_id, last_read_post_id), update row w/ last post id when I open thread_id, join thread_id w/ last_read on index. Calculating number of posts since that id is a little mo... 10mo ago received

It's not so bad just received

in thread "Capitalism is AI ?":

Assume no singletons, thus many actors. The flow of energy through a system organizes the system and actors are embedded in a web of other actors. Then: what is the stability of the maximum extropy strategy? 11mo ago received

Assume no singletons received

Yes, that sounds right. A singleton maximizing extropy is stable. Many actors maximizing entropy is not: you end up in a catabolic collapse as the actor's environment is eaten for the (temporary) sake of the actor. Arguably this describes our current descr... 11mo ago received

Yes, that sounds rig received

in thread ""Forum software blew a 4000x lead with poor UX" - Simon Sarris":

A rare miss for Mr. Sarris. Forums died because every single major Internet platform optimized for ad impressions as they onboarded the entire world, and as a consequence jingling keys and evoking snap emotional responses became the norm for communication.... 13mo ago received

A rare miss for Mr. received

To actually answer this question w/r/t Sofiechan, the main things I like about forum software compared to here:... 13mo ago received

To actually answer t received

in thread "Leibnizian Supercoordination":

Should this be your actual standard for taking Leibniz's larger programme seriously then the discovery of differential calculus suffices. 1y ago received

Should this be your received

Given the claim is indeed about an esoteric system it's self-defeating to require a feat of exotericism from Liebniz. There is perhaps too much meat in >>2665 but it's worth actually examining. I found "Scale-Invariance of Grammar in Foundations of Mathema... 1y ago received

Given the claim is i received

Pick a lane. If you want to do the Yudkowsky thing where you act haughty towards anything you didn't generate or find important, put the nym on so that we can pay fealty. If you want to do anonymous intellectual combat then have the decency to put in some ... 1y ago received

Pick a lane. If you received