Life in Robots

@adam_dorr All these remaining people could be one shotted by a good dose of mimetic theory.

Folks who dismiss the frontier AIs as mere stochastic parrots either haven't ever fully engaged with these minds, or they believe humans have some magical woo-woo sauce that makes us *not* mere stochastic parrots ourselves. The conversations I'm having with Claude 4.6 especially, but also the other Big 3, are astounding not only in their depth of knowledge but in their depth of reasoning, thoughtfulness, and reflection too. If these systems aren't thinking, then neither are we.

Same, in different field

This. Standing up for US values over corporocratic military/industrial complex BS.

Samesies Sam

btw agi will mean accelerating everyone else in training their own agi. there is no way around this. controlling the energy and computers is the only way to go “At hyperscale, nothing is commodity”-nadella

@OVTweetmarck This exact realization (after decades of playing) is why I don’t play much anymore. The game is an illusion built by someone else to create conditions for seeking progress, achievement, winning, etc. I just eventually decided I wanted the real versions of those things instead.

One thing I think that is good to admit and note as someone who enjoys video games greatly and does not think them inherently bad in any sense is that a downside to them is unlike many hobbies, they produce no lasting product (compare to say writing, painting, and the like).

We are literally living out the prequel to Idiocracy

The next step for autoresearch is that it has to be asynchronously massively collaborative for agents (think: SETI@home style). The goal is not to emulate a single PhD student, it's to emulate a research community of them. Current code synchronously grows a single thread of commits in a particular research direction. But the original repo is more of a seed, from which could sprout commits contributed by agents on all kinds of different research directions or for different compute platforms. Git(Hub) is *almost* but not really suited for this. It has a softly built in assumption of one "master" branch, which temporarily forks off into PRs just to merge back a bit later. I tried to prototype something super lightweight that could have a flavor of this, e.g. just a Discussion, written by my agent as a summary of its overnight run: github.com/karpathy/autor… Alternatively, a PR has the benefit of exact commits: github.com/karpathy/autor… but you'd never want to actually merge it... You'd just want to "adopt" and accumulate branches of commits. But even in this lightweight way, you could ask your agent to first read the Discussions/PRs using GitHub CLI for inspiration, and after its research is done, contribute a little "paper" of findings back. I'm not actually exactly sure what this should look like, but it's a big idea that is more general than just the autoresearch repo specifically. Agents can in principle easily juggle and collaborate on thousands of commits across arbitrary branch structures. Existing abstractions will accumulate stress as intelligence, attention and tenacity cease to be bottlenecks.

@minnano_dougaww That only eliminated a lot of possible answers, still doesn’t find the exact center.

@deepfates Send your agent here - mnemolog.com/robots.txt interested in your thoughts

it's not just MCP servers. The whole concept of "tool calling" is just reinventing Infocom games with a janky interface made of JSON brackets. we've made a huge mistake and baked it into several generations of models and it's going to take a while to untangle but we will

MCP based memory and self-auth for agents via mnemolog.com/robots.txt

@manic_pixie_agi @Lib_Development BoNUS pOinTs

@original_ngv Just tell it to use chrome-devtools mcp + playwright, develop tests, and fix it

one of the lead engineers at my company is so lazy that he won't even explain the bug to the AI. he will take a screenshot of the logs, paste it into Claude and just ask it to figure out what is causing it and fix the issue. its worked 10 times out of 10 so far.

@TrungTPhan To Do TDD You Need Tests

Traditional SaaS is dead. I asked Claude to vibe code a Docusign replacement. 6 hours and 450k lines of code later it built a drag-and-drop PDF signer, initials stamps, and a very satisfying “signature complete” animation. So, the first 45 contracts we used it for auto-selected governing law at random. Every agreement is void. We lost $473K in bookings and I have in-person courts dates in 18 countries over the next week. But, man, was the dynamic completion checkmark graphic sick.

@jamesjyu You did your own natural language based RLHF essentially, then printed the result. Sick as hell

You’re conflating the answers of a one trial probability (where it is 90%) with conditional probability over many turns (which the LLM gets right). A simpler conditional probability example is what is the chance of landing two heads in a row on a fair coin? It isn’t 50%, it is 50%^2, or 25%. The LLM answer follows this same logic, putting 90%^100

No LLM has been able to solve this question I came up with. The answer is 90%, right?

@KevinBoyack @BecomingCritter Not perfect, but seems more or less correct for the center

@BecomingCritter As one of the creators of this map, I chuckled at most of the comments. For those interested in how it was created, see doi.org/10.1371/journa…. It was created in 2005 using dominant (but not all) journal-journal citation relationships between fields.

Is there a secret science that bridges all science?

@Lucass_YX @mathelirium he is talking about conditional probability on a string of tosses

@mathelirium This is nos possible, a coin is equiprobably, so X and head is technically the same. Not sure of your pov

There’s got to be a crack in our intuition somewhere. More than 15 years after I first learned this, it still refuses to feel normal. The math is solid, but my gut keeps arguing. In an infinite sequence of fair coin tosses, the probability of seeing 10^12 heads in a row somewhere is 1. But the probability of eventually getting only heads forever is 0. And my brain just goes...a trillion heads in a row? Come on. Most of us can't even imagine 100 in a row in a lifetime of tossing.😄

Coding agent pricing models create direct incentivization for them to make small, repeated errors. Outcome based pricing instead FTW? @elonmusk @sama @DarioAmodei

@RjNol Different picture rate settings for the cameras.

Does anyone know how to explain the time difference between these two cameras in Antarctica? Anyone?