Priank Ravichandar

LLMs are exceptional improv actors: they’ve ingested countless conversations about consciousness and sci-fi plots, and can convincingly act out a role as if autocompleting a script.

The hypothesis here is that if you work hard on a problem, you soak your subconscious with it. Wrestling with a problem helps you build a mental model of what you know and what you don’t—providing the subconscious with building blocks to work with. (You can’t have genuine intuition and inspiration in areas where you lack knowledge.) Then, once you... See more

By writing something down and making sure it is solid, we can offload that thought from working memory and instead use it as a building block for the next step of the thought.

Human skeptics can kill a nascent idea, but ChatGPT was willing to entertain every far-fetched hypothesis.

My high-level advice—which supersedes everything below—is that you should decide what you want from your writing, and then completely ignore any advice which detracts from this goal. This includes advice which is practically sound but psychologically destructive when it comes to sustaining motivation.

writing—and reading, seriously, the writings of others—is a way to collect stepping stones: ideas that have been stabilized enough that they can carry us as we walk deeper into the thought space.

Consider how online radicalization happens: the combination of user agency (proactive search) and algorithmic amplification (recommending related content) leads people to weird places—to micro-cults of internet strangers with their own norms, values, and world-models. No corporate malice is necessary; the ML engineers at YouTube don’t care about... See more

AI nerds, as opposed to conscious technologists and normal people, who know how to separate enthusiasm for discovery from devotion to the God of excess, seem to like things regardless of valence or value. What matters to them isn’t the quality of the outcome or the meaning—life-saving or soul-draining, uplifting or addicting, doesn’t matter—only... See more

Chatbot companies want to have it both ways: personalization and emotion as a retention moat, but minimal responsibility for safeguarding the intimate relationships now running on their servers.