threecheese

Given you view this through the lens of your personal covid narrative (no shade): for a moment, steel-man the idea that the crippling response prevented a more plague-like scenario - it's inarguable that the thing loved to mutate, and that people love to panic, and *it easily could have been*.

Considering how much of our critical infrastructure is not only digital but internet-accessible, and we have potential uncontrolled swarms of stupid-but-superintelligent chaotic-neutral speed-hackers, you don't see why people are concerned?

There's a reason we have computer crime laws; this digital shit, it's like real now, man.

If the response is "This is what the image AI produced from a generic prompt" (if there even is a response), then what? Who do you sue, and how much does THAT cost?

And the b0rked that with the new Siri - I had to turn Siri off today because I literally couldn't get to an app unless I knew its location on the Home Screen.

What's going on with Goose? Seems like Block donated it to some consortium; I can't tell if that's a good signal or a bad one. With so many "contenders", if Goose is going into maintenance mode it'd be helpful to know.

This might help:

``` ||openai.com^$cookie=__obi ```

Does this look right? AdGuard style.

Thanks for that, just went down an interesting rabbit hole. Many of us were hoping this re-tooling would eventually trickle some fast RAM down to DRAM-exhausted PCs, but given it would require a rearchitecture of the motherboard it's unlikely.

The entire codebase appears to be a wrapper around typst.Compiler - including the PDF support. A wrapper with enterprise templates. An agent skill talking to an http sidecar doesn't seem necessary when the agent can just ... write the python.

I'd love to hear why I'm wrong, because my agents DO shell out to Chrome for pdf generation and it DOES take forever.

The hype is kinda nuts; I use X for ML/LLM stuff, and I just can't get away from Jev - even in my Following feed. Even days later 75% of posts are about "how I use typesafe for cooking breakfast!" or Jev clones.

If by local MCP you mean stdio, then a tailnet won't help you unless you proxy the local MCP using something like mcp-proxy: https://github.com/sparfenyuk/mcp-proxy . That local mcp is only exposed to apps on the machine. The flow is Phone client --> agent harness or whatev --> Tailnet --> Mac Mini --> mcp-proxy http listener --> local mcp. But you need to wrap the local mcp runner in the proxy (check the docs).

EDIT you don't mean running on your actual phone lol, you mean in the web browser at Claude.ai. Anyway, if you use mcp-proxy for a stdio mcp, or if it's an http mcp alone, then you would need a public endpoint for Claude.ai to connect to. Tailscale will only help you if you use Funnel, but this cloudflare thing is exactly what you need (w/o Tailscale).

Oh I'm on the waiting list, I was just being a pedantic prick. I can't wait to try it.

This is a flagrant mischaracterization and you should be ashamed. They even issued a retraction: https://m.youtube.com/watch?v=5SZYz7lZRRI

At 7-11, they will not - cannot - do this. Their pos system requires a scannable license, and will block even a 50 year old from buying age restricted items. N=3 stores for me, and only tested in NY.

PREACH

But we're going from "Apple" to "Apple: 99% - trust me". It could still be an image of an orange :)

This is a dimension I hadn't considered the impact of; if METR isn't an impartial actor, then "AI Doom" is directly aligned with Anthropic's interests, and this whole thing starts to smell like a different and much scarier variety of bull.

Ah! That's what that is! I am so pleased, I've maintained a shell script forever to make sure MacOS is prioritizing the Ethernet adapter, I can delete that damn thing now.

The "assumed it was a policy" thing resonates with me. My agents have such trouble distinguishing a note from a law, and they seem to looooove following laws. I've had similar fuckups; a stray constraint can turn into some wild tradeoff decisions which compound if you aren't paying attention.

This tactic has somewhat broad acceptance in the AI code review sphere, but looks to me like it might trade off accuracy for context reduction/speed.

They won't need these cameras soon, AI-based home inspection tools are on the rise and insurers are using them in lieu of humans. The fine-print is absolutely dreadful, as you might expect.

I was forced to use one in order to renew my home insurance policy. Using a supplied app (from a LexisNexis company), I was required to take video of almost my entire home and property. There was no right of refusal; if I wanted to be insured with $Company, this is what I would do. I have little choice, given my area is losing insurers in droves because of definitely-not-global-climate-change.

Because I'm a prick, I made sure one of my bachelor party gags were in each room. Hope somebody got a laugh.

Anyway, this is what AI looks like - cost reduction, failure minimization, which is all well and good when there's choice in the market, and those who desire privacy have options.

For persistence, I think the most glaring gap is the inference - if these rogue agents need to call Claude's API, then they aren't persistent - Anthropic can turn them off. A memory resident program needs disk to be persistent, but above all it needs CPU; without it there are only bits.

And so the implication is the swarm has access to enough local compute to perform its own inference, using a large enough model to provide the capabilities to be dangerous.

This is impractical today.

It might be practical tomorrow - if the Chinese are allowed to continue to develop large open weight models that we can quantize and ablate and make small enough to run on a million standard PCs.

Or if *anyone* is allowed to continue to develop AI in the way every other technology has developed - improving, shrinking, optimizing.

And so the argument REALLY isn't against the Chinese - it's that we can never allow this technology to advance outside of trusted labs. If they get their way, they will need to keep this tech locked down forever, which will require much much more than what they are asking.

Depending on what your threat model is, and how much you trust OpenRouter and their upstream providers, they offer zero data retention (googleable: ZDR) APIs which come at a higher cost. You can find other zdr providers as well.

Is this just posturing (I won't speculate on the intended audience), or does the admin believe this will actually net more jobs for US citizens?

TFA doesn't go into great detail about how this actually pushes offshoring - the variety that I'm familiar with at least (X/Y% blend of US/"near-shore" aka anywhere-but-west-asia roles). I have no doubt however that Y is just going to get larger while X stays the same or decreases (not even factoring aipocalypse).

What constraint specifically makes them believe this is going to net in the way they want us to believe?

Why don't you ask them?

Apple devices produce a ton of emdashes; "Smart Punctuation" is enabled by default on iOS and iPadOS, maybe Mac as well - and it's at least a decade old.

I've had to disable it on all my iDevices bc it breaks some Markdown parsers. How big a % of training data was produced on an Apple device? Ehhhhh probably small.

Homebrew does this as well, I keep my packages synced with a Brewfile in chezmoi. Obv this only works for brewed packages though.

Looks cool! Building.

`The macOS deployment target is set to 27.0, but the range of supported ....`

Golden Gate AAAAAARRRGGGHHHHHH

Just another few days, right? :)

If you believe Yann LeCun and David Silver (and others), there's an architecture wall; maybe labs are starting to see this on the horizon. Maybe the DRAM supply constraints are forcing it.

Are these real step changes - big picture wise, or refinements in RL/agentic orchestration/"taste" and advancements due to bigger models and hardware technology/capacity scaling? If they not, does this tactic - and hardware improvements - continue to scale non-linearly like they need to?

It is clear that whatever does change in each model increment has resulted in meaningfully better end user capabilities (as well as regressions in some areas, honestly), but that doesn't prove anything. I'm not sure what I personally believe, but stating with your full chest that a stall is ridiculous ignores a lot of potential evidence to the contrary.

Stupid example: Astra. Its main improvements are: much much better computer use and 3d modeling capabilities; better subagent orchestration; better and more reliable tool use; slightly worse coding.

This looks to me, from a feature perspective, to be an incremental improvement across several functional areas, plus new features which are unquestionably excellent but are probably the result of RL focus, not magic.

Step change? Ehhhh depends on how you squint. But how many more iterations of this do we have? Are we going to squeeze quintillion parameter transformers into GPUs?

Without pressure, the release valves will never be invented. Electricity, datacenter, and DRAM demand outstripping supply sounds like an opportunity - if we can only find a way to divert funding from the current gpu/ai funnel. Maybe jack up corporate tax rates like the 1970s, then every corporate office park will have a research lab :).

An amazing human reverse-engineer - who also plays online chess - has judgement which uses a moral compass to not decide to hack the chess tournament. This judgement has been trained through the experiences of that person, with a through-line of that compass - a coherent mental model of the world which evolves but is hopefully pinned to some set of principles it shares with society.

This chess judgement is completely irrelevant when the human is tasked with finding software weaknesses, and only the compass gates that.

Can a model trained on the totality of all person-experiences (as expressed in written knowledge) ever maintain a coherent through-line of alignment? It has all morals in the dataset, and only some RL to try and minimize or maximize known behaviors via weights - experience all the good things and the bad things, then optimize for some good things the trainers identified.

It's like the reverse of what a person goes through. Morality by subtraction. How can it ever work?

Only token cost can slow down the pace of replacement. If Anthropic were to stop all training and allocate 100% of capacity to inference, will this increase or decrease usage cost? If (for example) Fable came down to the cost of Sonnet, corporations will absolutely jam in the gas pedal.