2025-03-20
underlying
Scott Alexander on misophonia, noting that it appears to be generated by the concept of chewing rather than chewing or it’s noises itself. This is in line with what Chris Lakin and a lot of TPOT generally say about how these things seem to work.
Benjamin Riley interviews neuroscientist Paul Cisek about LLMs and consciousness. It makes me wonder why people don’t seem to be trying to empirically verify theories of consciousness using LLMs. For example, if the consciousness is resonance between different areas of the brain theory is true, why not take a take some different specialized LLMs, and force them to communicate through a sparse autoencoder. A very basic version could be something like an image classifier for ASCII, connected to a next token predictor. Then you can add different reward functions to the autoencoder output to test out various hypotheses, as well as gradually increasing the complexity by adding additional input “senses”, and potential outputs with things like robot arms. The goal would be to artificially evolve a brain, from its origins as just some retinal ganglia and a hindbrain, up until you have fully self-directed long-term planning.
Some interesting translations of Chinese thinkpieces on tech strategy, one by Kyle Chan on how to win the chip war (by focusing on mass production of non-leading edge semiconductors), and one in ChinaTalk about what’s necessary to produce AGI. Interestingly, essentially everything they are saying is more or less what policy experts also say that the US should be doing, which is strange given they have different circumstances and conditions.
Noah Smith reviews reviews of Abundance. It’s unclear to me what to make of this claim that the new battle in the US left will be between output-focused abundance types and ideological anti-corporate trust busters. If the former is focused on outputs rather than inputs or means, then the latter shouldn’t really be a true countermovement, but just another obstacle out of many that can be routed around. Also, antitrust appears to also be migrating towards the right along with degrowth. For example, here is Brian Albrecht, on how antitrust might soon be part of the UK conservative agenda.
Samo Burja on the potential for Indonesian development, as an Asian country with a large population, abundant resources, and a fertility rate above replacement. I’m also not sure if I buy the argument. Firstly, Indonesia is culturally distinct from East Asia, composed primarily of Malay Muslims, In which case, comparisons to groups like Malaysian Malays may be more apt for things like fertility or economic output. Secondly, insofar as their current fertility rate is high, this is probably due to the importance of Adat in resisting global cultural integration, and diversion of modernizing impulses towards Islamization because modernization in rural areas is associated with becoming more like Java (although I should note that I haven’t visited Indonesia, and I am basing this just on my reading of Indonesia, Etc). Both of these factors probably makes things like industrialization and financial integration more difficult. Related, Experimental History on nonconformity and breaking traditions.
Greg Miller on Harberger taxes. I feel like there’s a clearer and more general form of the argument he states, which is that the range of time-discounting across different people is actually very wide. People with positions far outside the mainstream consensus have to pay a premium for their preferences when transacting with others. The usual way to handle this is to choose not to play, but Harberger taxes force you to place an order.
Richard Ngo on coercive movements. It’s unclear to me if it’s true that bottom-up movements are more coercive than top-down ones (it seems to me that top-down mandates are definitionally coercion). My model of coordination is that movements succeed when there are more incentives towards spreading and enforcing some particular outcome than the counter-incentives towards actively resisting them. Coercion should actually be less common in bottom-up movements, because strategies of enforcement tends to make more opponents engage in active resistance, when what you generally want instead is for your opponents to ignore you until you’ve completely taken over.
Stephanie Murray fertility studies and links.
Works in Progress linkthread.
Andrew Cutler woo linkthread.
