zahlman
zahlman on Hacker News
This is not "self-editing". We're comparing Astra to a model with not even 1B parameters.
I'm reminded of how humanity recently finished mapping out the fruit fly connectome; this does not put us meaningfully closer to, say, doing surgery on humans to cure schizophrenia.
Sure, but "not looking at Claude usage" is a bizarre strategy when it's being incorporated into an argument against Anthropic's IPO.
I'm pretty sure the other person didn't mean "probability mass function", so Google is not really helpful to me here.
One might wonder how we ever managed to do it before the Internet existed.
I have trouble making sense of this argument. The underlying premises seem to be:
1. "isolationist parties" are inherently aligned with Russia and that it's good for Russia, in a zero-sum way, for countries that are not Russia to implement those policies.
2. Attacks like these are definitely done by Russia and we don't need to wait for evidence to connect them.
But there are problems with that:
1. "Isolationist parties" are nationalist, so if they knew that Russia was causing a problem for their own country, why wouldn't they oppose Russia? More generally, the claim that "small remainders like these is enough to increase that fear" is basically arguing that when a country engages in terrorism that somehow makes everyone else less inclined to take a stand against them. If that were true, I'd expect to see far more terrorism going on in the world than actually occurs.
2. The main "isolationist" policy is restricting legal immigration and/or enforcing immigration law. But if it's good for Russia that other countries do these things, it should also be good for Russia for them to do the opposite (because again, the idea that the way other countries govern themselves should impact on Russia, requires zero-sum thinking). If the powers that be in Russia believe this, we should expect them to pursue expanded legal immigration for themselves aggressively. But in fact their rate of legal immigration (percentage of population per year) is roughly on par with the USA, and it mostly comes from former USSR republics.
3. The truth of this narrative would be far too politically convenient for the people who have been trying to tie Trump specifically to Russia for over a decade despite many of their stories completely falling apart.
> H1B abuse and offshoring have both existed for decades
Yes, and they've had negative effects for decades and shouldn't be shielded from criticism because of whatever new thing.
> If you’re blaming immigrants for this, you are either doing so in bad faith
The bad faith thing is to round off the criticism to "blaming immigrants" because it creates a more convenient moral landscape for the argument.
The entire reason they are put on the Internet is to allow for remote control. What problem would it solve to allow remote monitoring of refrigerator data without control?
> The problem is that these refrigerator cases are not as simple as your refrigerator at home. As a simple example, many systems allow you have control over the deadband: the exact low temperature at which the compressor turns off and the high temperature at which the compressor turns back on. You also have control over defrost cycles, compressor pressure and many other things (I worked with this a long while ago, and I don't remember all the details). … This is not something that a non-expert supermarket employee would be able to handle well.
I still don't see why the expert has to be remote, or why an integrated computer can't do the optimization. Or why it needs to be fiddled with at any significant frequency.
>>If I was an Anthropic employee expected to dog-food the product…
>I wonder how much dog shit has crept into their codebase by this point.
You'd think they'd know not to shit where they eat… ?
> I haven’t owned a TV in years
… But you feel the desire to go back to it? Why?
…Why would it be surprising for the USA not to have such a requirement?
The issue here is that your statements do not reflect "skepticism", but confident disagreement that is well refuted by the available evidence.
Interesting (pun intended) to me that apparently everyone wants to talk about the rate speculation, rather than about the finding that jobs are picking up in spite of AI.
This seems like straining the definition of "compression". We have a delta-coding filter and then palettization? That's it?
> they're off.
Help me out here — to where?
> People get far too much of it
I find it hard to imagine how. Carrots don't seem to be especially popular last I checked; and to my understanding, produce generally has become more nutrient poor in recent times due to soil quality issues.
> Pulse is a simple monitoring tool that does not require you to sign in, or setup anything. You just download it with a single curl command, and run it on your device.
This should be the first paragraph, and it should explain what the tool monitors.
Everything about the page is very obvious SEO'd AI slop; the "scripts" pages don't show the important content without JavaScript (presumably because they're actually web apps — perhaps the backend is written in Python but there's really no reason that would matter to the user, nor a clear way to evidence it); the "downloads" ask people to unpack archives and run Windows-specific .exe files for things that we are told are "built with Python" (no sensible HN user would do this), and which in at least some cases are hosted on "catbox.moe" which doesn't exactly seem like serious, permanent storage for anything.
>nasty stuff like vitamin A
Pardon?
I write mine, and will continue to do so, by hand. I would never consider letting an LLM touch my writing.
Posting frequency is another matter…
As much as I wouldn't want to write that much assembly, you make it less daunting the same way that you would with any other programming language. There are simple idioms for loops that you can macro-ize; similarly for function setup and teardown; more importantly, the concept of a function still exists at that level.
(2015)
I don't know why I'd "(probably) think" they're anything like symmetrical. After all, I have first-hand experience of reading code and writing code; clearly, very different processes.
> They're bundled in with appliance sets sold to first time home buyers and preinstalled in every apartment.
Do you object to stoves and refrigerators in the same way?
> The rest of the world doesn't have them.
This is not at all true. Source: my non-American household, which has had one for decades. (Also, web search.)
> even TC himself admitted that the dishwasher is sort of a solution in search of a problem
I don't recall him saying anything of the sort. I don't suppose you could provide a timestamp?
> Calling you all "naive" is really an understatement once the echo chamber begins reverberating. Jeez! Yes, you're part of it. I'm simply pointing it out. Yes, "citation needed" is a pseudointellectual crutch. No, you don't have to trust an influencer just because everyone else does.
It's not a matter of trust. TC presents a complete, reasoned, independently verifiable argument; and it turns out that his results replicate quite well. As per the skeptics.SE link you brushed off, and as per the personal experience of myself and many others.
At any rate, you are using inappropriate rhetoric and then disingenously trying to brush it off as "simply pointing out" claims that don't actually stand up to scrutiny.
How is it different, though?
It would only be half a eurocent to update Firefox, though.
I don't see the appeal of these modern games.
Oh, I don't doubt that. But most of the time I can pick out useful information from the claudeslop. This piece seems almost deliberately obscurantist.
Something about the writing style here makes it hard for me to follow what actually happened. The section about JSONP for example seems to be explaining how JSONP works and how it's designed as an attempt to work around these attacks, but that explanation is then passed off as an explanation of how they bypassed the protection. And the whole premise seems to be about how manipulating the Preview content let them work through multiple defensive layers, but I don't see an explanation of how they directly controlled that content in the first place. I would have thought that the only control you have over the content of "an ordinary public ChatGPT share" is the prompt text, and that any migration of that text into "a small working webpage using its Preview feature" is up to the model deciding to generate that "HTML block" in response.