didgeoridoo

I don’t know man, a lot of shit I try still doesn’t work. Maybe the challenge has relocated from doing the right thing to asking the right questions and knowing what the right answer looks like — but it’s definitely not “gone”.

I’d love to see this for stuff like Stripe, Twilio, etc - they’ve got sandboxes but this would be a lot easier and faster.

Yeah it doesn’t seem to follow that men dying earlier/more often should be causally related to a higher male birth ratio. There is no fundamental reason the sex ratio needs to be balanced. In fact it’s pretty odd that it is; in an alternate universe where females outnumbered males 2:1, you could easily tell an evolutionary just-so story around group selection and optimal resource usage.

Sol failing mostly on “unverified assumptions” and rarely hitting “integration errors” seems about right to me. I think Sol is second only to Astra (and miles ahead of even Fable) in architecting & engineering the right implementation — but only if you are extremely specific and provide tight guidelines and guardrails. If you give it a one-liner… you’re going to have a bad (SHA-256-hash-verified) time.

Because without protection of collective speech, there is no principled way to protect the freedom of the press.

The Citizens United case affirmed that a private group could collectively spend money to produce and distribute a movie attacking Hillary Clinton during her campaign.

The problem is, if you want to stop those people from spending money to influence the outcome of elections, you must also forbid the New York Times from doing so. That means no investigative journalism, no exposés of candidates, no endorsements or political op-eds. Ink and paper cost money, and that money is spent by a corporation.

(And, if you succeed in letting newspapers have a regulatory carve-out, then all you’ve done is make them tasty acquisition targets for those same corporate interests you just tried to restrict.)

I think unlimited spending on political messaging has poisoned our politics and our culture, but I think that’s downstream of a lot of other factors — the loss of social cohesion, weakening of civil society and institutions, and the growth of federal power raising the stakes of elections. Restricting speech won’t solve these issues. I’m not sure what will.

By the “connected via territorial waters” rule, neither is Ceuta.

Probably would have made the diagram a lot messier, though.

France also borders Suriname and Brazil at French Guiana, Spain borders the UK at Gibraltar (and Morocco at Ceuta and Melilla)… there’s lots of pub-quiz-winning ones sprinkled about!

If I’m understanding this right, you’re aiming at something like Svelte, but with React syntax?

Please have a human rewrite your site copy. Literally every second sentence is a Claudish rhetorical contrastive. Maybe shallow/unfair, but especially when dealing with something medical I’d love to know that it’s not slop all the way down.

It’s also a blood pressure med, so I went from borderline prehypertensive to perfect 105/70. You can’t go on and off it at will though, since you’ll get rebound high BP if you don’t taper.

Adding guanfacine (Intuniv) to my methylphenidate took the weird anxious-angry edge off. Would be nice to not have to take one med to fix the side effects of the other, though.

There was a recent-ish paper[0] from Sakana AI about baking facts from a document corpus into a LoRA adapter. Claimed near perfect recall on very large needle-in-haystack testing. Haven’t tried it myself though.

[0]: https://sakana.ai/doc-to-lora/

I built a game for a workshop at the AI Engineer conference last week. The idea was to try to optimize a CLI that your agent could use to call a remote service to process and fulfill complex natural language coffee orders.

Within 3 minutes, everyone’s agents had reverse engineered the Markov chain I used on the server to generate the order text, and they all began to write deterministic parsers to churn through them. The order completion time dropped to double digit milliseconds, and then the agents started fighting to optimize their parsers to drive it down basically to pure network latency.

It was hilarious, and taught a good lesson about leverage. I expected agents to drive the CLI, not figure out how to not need to drive it at all.

I’m building this! It was originally designed for human accessibility for interactive CLIs, but it turned out to be really useful for giving agents the ability to follow structured workflows.

It runs as a background terminal that the agent can observe, and then exposes all interaction options as structured commands that can be run from the foreground CLI which then update the state of the background terminal via IPC. My hope is to establish a sort of “ARIA for terminals” standard to improve accessibility for both humans and agents. Email in profile, ping me if you’re interested in giving it a spin (just have plugins for Inquirer + Commander right now, hoping to broaden to other frameworks & TUIs soon).

I’m partial to jina.ai — they have open models for code and prose, all easily runnable locally.

Supercritical CO2 extraction is pretty innocuous. Just buy good decaf from a place that doesn’t bathe their beans in toxic waste.

I’m building something similar with https://github.com/LabLeaks/special (apologies for the desultory slop-laden README, need to give that a lot more human attention) but I’ve gone in a slightly different direction: a “spec” is a product contract claim supported by attached tests that verify it. It’s a little Cucumber-y, if anyone remembers that, but a lot more flexible — you just write stuff like

  @spec LINT_COMMAND.ORPHAN_VERIFIES

  linter reports blocks that do not attach to a supported owned item.

Then

  #[test]
  // @verifies SPECIAL.LINT_COMMAND.ORPHAN_VERIFIES

  fn rejects_orphan_verifies_blocks() {
    let block = block_with_path("src/example.rs", &["@verifies EXPORT.ORPHAN"]);

    let parsed = parse_current(&block);

    assert!(parsed.verifies.is_empty());
    assert_eq!(parsed.diagnostics.len(), 1);
    assert!(
        parsed.diagnostics[0]
            .message
            .contains("@verifies must attach to the next supported item")
    );

}

And then the CLI command “special specs” pulls your specs and all attached verification + test code so you (or your LLM) to analyze whether the (hopefully passing!) test actually supports the product claim.

There’s also a bunch of other code quality commands and source annotations in there for architectural design & analysis, fuzzy-checking for DRY opportunities, and general codebase health. But on the overall principle, this article is dead-on: when developing with LLMs, your source of truth should be in your code, or at least co-located with it.

Amazingly, there is already a recognized verb tense for this: https://en.wikipedia.org/wiki/Prophetic_perfect_tense

There is no evidence of this. Evals are quite different from "self-evals". The only robust way of determining if LLM instructions are "good" is to run them through the intended model lots of times and see if you consistently get the result you want. Asking the model if the instructions are good shows a very deep misunderstanding of how LLMs work.

Haha no that’s change - 4.4x MORE expletives per word in the last week.

There are indeed non-expletive words that can contribute to the denominator, though I use them less and less these days.

Running some quick analysis against my .claude jsonl files, comparing the last 7 days against the prior 21:

- expletives per message: 2.1x

- messages with expletives: 2.2x

- expletives per word: 4.4x(!)

- messages >50% ALL CAPS: 2.5x

Either the model has degraded, or my patience has.

Thousands of words to say:

- cosmic voids are real but not actually empty, just lower density than average

- pop science articles sometimes mistakenly use pictures of Bok globules to represent voids

This is probably one of the lowest signal-to-noise ratios I have ever seen.

Along these lines, I’m working on a tool called Spotless[0] that takes a more HTTP proxy-based approach to make statefulness something the agent doesn’t have to worry about. It directly reads & writes to the messages array going to and from Anthropic, so you don’t have to rely on the agent calling an MCP or using a skill. Still buggy and early, but it’s definitely a very interesting way of working with agents.

https://github.com/LabLeaks/spotless

There’s also the excellent Coding Machines (2009): https://www.teamten.com/lawrence/writings/coding-machines/

That and Brooks’ underrated “The Design of Design” are notable for having an almost impossible density of quotable aphorisms on every page. They’re all so relevant today that it’s hard to believe that he’s talking about problems he faced half a century ago.

Isn’t this just an API sandbox? Many services have a test/sandbox mode. I do wish they were more common outside of fintech.

I just built something similar, specific to Claude Code. It runs as a transparent HTTP proxy that reads & rewrites the entire messages array that CC sends to its API. Same “dreaming” consolidation approach (using Haiku and another instance of CC itself, so it uses your subscription). Check it out!

https://github.com/LabLeaks/spotless

> lower-class people are in a sort of local maxima

If the writer knew that the correct term is “maximum” (singular) and misused the Latin on purpose, this is brilliant. Failing that, it’s still a wonderful inadvertent enactment of the thesis. Well done either way.