Ratish

low entropy conscious matter | software, science, tech, ai, sci fi, philosophy, and all things kardashev scale ascension

[Ratish @Imperor42]

🧵 Thread of Threads

Things I'm building, chains of curated/interesting thoughts and ideas, etc.

Scroll to view⬇️

[Ratish @Imperor42]

@ishaangangwanii @SynScience Really cool though! The scientific world definitely needs more of this

[Séb Krier @sebkrier]

Positive visions of the future are easy to produce, so it's understandable that people dismiss them. I think that's a mistake, and that disempowerment-fatalism is just as easy. If I had to imagine a nice 'post-AGI' vision, it would look like this:

I have to work much less, but still have a chunk of the year working if I want to. There is no 'work day' but there are 'work hours'. Opting out is sustainable for those who don't want it. A Venetian-like economy creates new roles, tasks, and quests that people never imagined could exist (I think full substitution will take much longer to bite than people assume).

The rest of the year is a mix of travelling, helping design a new city somewhere, doing research/writing that is more longitudinal in nature, and making music with new synths designed at some workshop in a different continent (the world of atoms is flourishing, and globalization is so back).

Low goods/services prices let me and my friends coordinate a few impromptu hangouts abroad every now and then - we realize it's much easier to nurture social life when time is not a blocker, and coordination costs are lowered by agents. My family's health issues are managed, giving them more years to live (clinical trials have sped up significantly).

The decreased dependence on living in a large capital city combined with the affordability of building transport infrastructure leads to a flourishing of new towns in places that would otherwise languish; people are excited to build them (yes building is legal again). There are a number of large scale projects to create colonies in space for the more ambitious pioneers, and many highly paid jobs there too.

Coordination tech lets people self-select into hubs, communities, towns that fit their vibes. In fact, the 'age of polarization' is now behind and a lot of governance is pushed down to the local level, which lowers the stakes of national identity fights. Conflicts naturally remain, but the 'boring majority' is no longer faced with having to choose between two flavours of outrage and righteousness.

Government is human-led and agent-mediated. Automated dispute resolution lets human courts deal with more important and significant cases. Citizens can automatically simulate the likely impacts of proposed laws, and agents help demystify them. The machinery of government still has humans, but it's far leaner; there's a neat separation between instrumental tasks (automated) and normative ones (augmented). Politicians are rewarded more, but they're also under automated scrutiny by constituents - e.g. commitments are easier to monitor and enforce.

I could go on! But as I write this, some will rightly say 'well this just sounds like a nice sci-fi story', and tbh they wouldn't be entirely wrong. For each sentence I can find a story why it might not happen. It's still helpful to outline a positive vision of the future to have an idea of what to aim for, but ultimately much of this will come down to resourceful political entrepreneurs, founders of new companies, important changes in legislation and culture and so on.

You don'tt get any of this automatically, it's an endogenous dynamic that depends on how many people exercise their agency. There is of course a degree of determinism to technology, and wider dynamics that cannot be stopped - but people too often point to them as if disempowerment is the only possible outcome. I think that's 'spectator cope', and a lot of shaping will remain possible. Societal change is inherently hard and diffuse, but it's certainly not impossible - too many people despair at being unable to control wide structural changes, but that's missing the trees for the forest, in the most literal sense.

[Ratish @Imperor42]

@internetphysics What looking at any pre mid 2020s essay or book will feel like decades from now:

[Daniel @growing_daniel]

Books written before gpt 3.5 are now pre-nuclear steel

[Ratish @Imperor42]

@AnikaSomaia Outsourced curation, nice. I guess a lot of good content is hidden like that

[Ratish @Imperor42]

[1/n] Bubble Galaxies (@bubblegalaxies)

Tldr; customizable, procedural sandbox galaxies rendered in-browser because I like sci fi and 4X-style video games and I think most of the current ones can be improved upon a LOT. Especially with AI storylines.

https://t.co/0mhmYzpGNJ

[Ratish @Imperor42]

[1/n] Initial iterations of the procedural galaxy builder for @bubblegalaxies

https://t.co/r49TlP9f2b

[Ratish @Imperor42]

🧵 Thread of Threads

Things I'm building, chains of curated/interesting thoughts and ideas, etc.

Scroll to view⬇️

[Ratish @Imperor42]

[6/n] Toggle between galaxy archetypes apart from all the other parameters

https://t.co/px8VTLJney

[Ratish @Imperor42]

Toggle between different galaxy archetypes and hit apply to procedurally generate new ones. 6 physics based galaxy types rn - spiral, barred spiral, elliptical, ring, irregular, and colliding

Under 2 second load times for new galaxies of 2k star systems each

[Ratish @Imperor42]

[1/n] Initial iterations of the procedural galaxy builder for @bubblegalaxies

https://t.co/r49TlP9f2b

[Ratish @Imperor42]

Update:
Making a galaxy simulator (work in progress).

At the moment:
- 38 adjustable parameters
- 4 adjustable levels of detail
- 11 star textures (6 common, 5 rare)
- 11 nebula textures

[Ratish @Imperor42]

[5/n] A couple more planet types added (Volcanic and Ecumenopolis)

https://t.co/jUemzNsxnb

[Ratish @Imperor42]

New planet types too, of course. Volcanic and ecumenopolis type previews:

[Ratish @Imperor42]

Link to the actual blog post:

https://t.co/HU5NsK2O2H

[Ratish @Imperor42]

We've known some part of this already - how different layers of an LLM specialize in different things. And how duplicating a thinking block of layers ("LLM neurosurgery"), for instance, makes them think harder.

From a nice related read:

[Lior Alexander @LiorOnAI]

Large language models spontaneously develop the same specialized brain regions humans have for language, math, physics, and social reasoning. No one designed this. It just emerged.

Two completely different optimization processes (biological evolution vs. gradient descent) independently arrived at the same solution.

[Ratish @Imperor42]

@CorticalLabsBDM Here: https://t.co/5pcryBXpO0

[Ratish @Imperor42]

Tried out Tripo AI's image to 3D model for Cortical Labs's CL1 and it seems to have gotten quite a lot correct from just one orthographic image

[Ratish @Imperor42]

Some visuals in progress: minds in a dish that can game (yes, inspired by the Doom-playing ones of Cortical Labs from earlier this year).

Science can be complicated. But it's a little less complicated when you can see (literally) how a certain technology works under the hood.

[Ratish @Imperor42]

@CorticalLabsBDM On it, fixing that

[Ratish @Imperor42]

Tried out Tripo AI's image to 3D model for Cortical Labs's CL1 and it seems to have gotten quite a lot correct from just one orthographic image

[Ratish @Imperor42]

Some visuals in progress: minds in a dish that can game (yes, inspired by the Doom-playing ones of Cortical Labs from earlier this year).

Science can be complicated. But it's a little less complicated when you can see (literally) how a certain technology works under the hood.