These answers did not look like typical, cliché, familiar or commonsense answers to me. I don’t think a group of human subjects would consistently give these answers.
If so, what’s the name of their take in Philosophy? What do we call that idea? It must be so mainstream, right? What are the well-known references in the literature?
Modern cynicism. Specifically the form you see on social media, where road-raging cult-like bubbles imagine post-truth to be the great explainer of their misery.
I used one well-known phrase, ‘Very Online’. Germans also coined the lovely ‘Hobbyless’ (Hobbylos). And then the catch phrase directed at these types, ‘touch grass.’
Your instinct is right that most people don’t hyper focus on this world view. But the training sets can’t get transcripts from the real world at scale, so they over-index on the more easily scrapable neuroses of the hobbyless.
Fair enough. My impression was that it's a bit Anti-social take as well.
Two models use the word "tribe" and three others use the word "group". A sociologist, or any human would use these terms mostly in a positive light. AIs use them only in a negative light, painting them as the main obstacle to rationality. Which I think is unique, and interesting enough.
This view also contradicts all the religions I am aware of. Religions affect worldviews of most humans, and they all promote sacrificing yourself for the "tribe", accepting the tribe's "truth" as the only source of truth etc. Just the opposite of what the LLMs say.
Also note that none of them used the word "emotion". They haven't repeated the sci-fi cliché of "humans are weaker than robots, because they are not rational, they are emotional". They pointed to a specific emotion instead, "belonging". 4 different models use the word "belong", again in a negative context.
As long as frameworks and languages offer some novel eval to the agents, they’ll remain advantageous for certain projects. Rust is the most obvious case, giving the agent a borrow checker.
I think the frameworks with strong batteries-included conventions also offer unique guardrails for agents building more ambitious apps. Those conventions allow for highly opinionated linters and well-worn migration paths, which become an eval advantage.
Eventually we will develop new languages and frameworks that are tailored to agents maintaining projects on epic time horizons. Until then, we can squeeze free evals out of these existing convention-laden technologies.
Ergonomics for hand composition are less relevant now. The most important trait is how does a long-lived project perform when written by a coding agent.
I’m curious if anyone has any takes on the different approaches to agentic styling.
I’ve tried Tailwind for this without further steering, and after a while the agent starts adding unnecessary utility classes everywhere. Then I taught it to micro-Ralph eval all its utility classes using screenshots and this works ok but burns a lot of tokens. Treats the symptom and not the problem IMO.
Maybe with the right set of constraints there’s a subset of Tailwind that composes better? I went back to vanilla CSS for now, on the hope that it will readily port to whatever the best thing ends up being later.
IMO we won’t see AGI until the models are sitting inside bodies with senses and gut biomes and opposable thumbs and the like, all influencing the context window in real time. Intelligence is a full-bodied experience
I had a lot of fun. I think the progression curve gets boring around Boss Level 11. Introducing a prestige system could extend the runway.
I thought both Factor and Divide were interesting twists to the battle system, but neither were given enough room to breathe. With different boss numbers in the early progression, Factor can be tuned to make battles winnable that otherwise aren’t. That saves Divide for later in the progression. As it is now, they both surface around the same time and crowd each other’s debut.
It’d also be good to highlight new developments by making them unlockable in the Clubs skill tree. E.g. make an unlock or unlocks for “2-star numbers auto attack”. And “Unlock Factor where factorials of boss numbers get huge stat increases.” It mostly gives you a way to spell out and educate new mechanics, but also it pads out the Clubs skill tree which is a bit thin comped to other games like this.
I hope you keep iterating on it, it’s a really good game!
It’s going to be a long road, but I think as LLMs and their offspring create more and more convincing arguments for silicon consciousness we will conclude that consciousness is about as real as humours, and we’ve all been p-zombies this whole time.
Maybe the literary creature shoe should have started on the other foot, and sent us in search of proof that we are or are not p-angels. That at least puts the burden of proof on the compatibilists where it belongs.
From a non-dual perspective either nobody is a p-zombie or everybody is a p-zombie. I think the trajectory of LLMs will be that consciousness sensations are abundant and low value, and that will give all the non-dual territory over to the illusionists (who now have a tangible example to point to.) Emergence won't be disproven but it'll be about as interesting as a penny stock. Everyone else is going to go to ground as a dualist and argue for some entirely new unreachable aspect of human exceptionalism, but they'll have ceded qualia as they did for souls. If I had to guess, they'll focus on embodied cognition because building high-fidelity bodies still seems really hard.
reply