indirection's recent activity
-
Comment on Your cookware got worse on purpose: who owns Pyrex and All-Clad now in ~food
-
Comment on Hack reveals suno AI music generator scraped YouTube, Deezer, and Genius in ~tech
indirection LinkI think it should be legal. If random companies can't scrape these sites, they'll still sell the data to the biggest companies (and/or use it themselves), so only they'll have access.I think it should be legal. If random companies can't scrape these sites, they'll still sell the data to the biggest companies (and/or use it themselves), so only they'll have access.
-
Comment on 'The report's so stupid': The DNC 2024 autopsy is roiling US Democrats in ~society
indirection Link Parent@scroll_lock wrote an excellent post on effective protesting and advocacy: https://tildes.net/~society/1n6v/protests_are_great_the_next_step_is_advocacy_heres_how_to_do_it_effectively@scroll_lock wrote an excellent post on effective protesting and advocacy: https://tildes.net/~society/1n6v/protests_are_great_the_next_step_is_advocacy_heres_how_to_do_it_effectively
-
Comment on 'The report's so stupid': The DNC 2024 autopsy is roiling US Democrats in ~society
indirection (edited )Link ParentI’d like to see people vote for the Democrat, because obviously it’s still important to vote for a bad candidate over a worse candidate… But protest every DNC event, harass officials (legally and...I’d like to see people vote for the Democrat, because obviously it’s still important to vote for a bad candidate over a worse candidate…
But protest every DNC event, harass officials (legally and constructively, i.e. with an out if they act in constituents’ interest), send and vote for outside candidates in primaries, until one-by-one they’re replaced by members more in-touch with the average voter. I simply can’t imagine them operating when almost their entire base is against them, every public appearance is protested, every unmoving figure is slandered; and the more staffers replaced the easier it gets. Don’t hesitate to harshly criticize Democratic Party members even if you want people to vote for them over Republicans: some people will still be confused, but I believe that’s inevitable, because the member will be harshly criticized anyways by Republicans, and you’re helping by (again) explaining that a bad candidate is better than a worse candidate.
And perfect is the enemy of good. I’d prefer a candidate with a strong economic policy whose social policy is basically “live and let live”: I think (besides being easier to elect) that would help minorities, who are often poor, more than a strong social policy.
-
Comment on How I feel about LLM (AI) writing in ~tech
indirection LinkI agree with everything you said, except I find most LLM writing immediately apparent, and not strong. Even LLM writing disguised enough to not be apparent would be OK…maybe not as good as...I agree with everything you said, except I find most LLM writing immediately apparent, and not strong. Even LLM writing disguised enough to not be apparent would be OK…maybe not as good as authentic writing, but it must be semi-authentic because it must stand out from other LLM writing in its time.
Fortunately most of this writing is also apparent to automated tools (at least Pangram). So a tool that filters LLM posts and comments, if it doesn’t already exist I’m sure someone (maybe I) will make eventually. I’m aware that almost certainly some LLM writing will pass the filter, but again, it’s a significant improvement if obvious writing is hidden.
-
Comment on Mark Ruffalo and Matt Stoller: This merger can, and should, be stopped in ~movies
indirection Link ParentIP loss is bad. But even though Paramount would own CBS and CNN, there are many other left-wing news stations. If the politics shift, I expect most of the audience to migrate.IP loss is bad. But even though Paramount would own CBS and CNN, there are many other left-wing news stations. If the politics shift, I expect most of the audience to migrate.
-
Comment on Mark Ruffalo and Matt Stoller: This merger can, and should, be stopped in ~movies
indirection LinkShould it? Paramount is paying $111 billion for a film studio and dying news company. The IP is a big loss, but the money given to Warner Bros execs could fund new organizations well for a long time.Should it?
Paramount is paying $111 billion for a film studio and dying news company. The IP is a big loss, but the money given to Warner Bros execs could fund new organizations well for a long time.
-
Comment on Do you play knock-offs of celebrated indie games? in ~games
indirection LinkThis reminds me of the Threes and 2048 controversy. Threes is a simple game, released 6 February 2014. 2048 is a derivative of (another derivative of) Threes, released 9 March 2014. 2048 achieved...This reminds me of the Threes and 2048 controversy.
Threes is a simple game, released 6 February 2014. 2048 is a derivative of (another derivative of) Threes, released 9 March 2014. 2048 achieved much more popularity, and is free whereas Threes is paid. The Threes developers have called 2048 a rip-off.
Personally, I think ideas (especially simple ones like Threes) should not be protected like IP. I think the only thing that should be protected is likeness, i.e. nobody without permission can make something that could be confused with the original. For example, Notepad++ for Mac was called out by Notepad++, IMO rightfully so. Now they've renamed to Nextpad++, which doesn't absolve their guilt, but I think if they originally stuck with that name and design there'd be no issue.
Instead what should happen, is when a game (or other media) closely resembles an earlier one, then those aware of the later media should be made aware of the earlier media. So people who are aware of 2048 should be aware that Threes executed the concept first (though they can stick with 2048 if they like the gameplay more and don't want to pay). The less resemblance, the less awareness is justified; the more someone likes or invests in something, they more they should know its inspirations.
-
Comment on "CEO said a thing!" journalism in ~tech
indirection Link ParentEven worse: “mayor of [town in the middle of nowhere] [who belongs to the opposite political party] says something dumb”. There are over 19,000 mayors in the US, some are bound to be idiots. Same...Even worse: “mayor of [town in the middle of nowhere] [who belongs to the opposite political party] says something dumb”. There are over 19,000 mayors in the US, some are bound to be idiots. Same with spokesperson, high school principle, etc. At least Trump accurately represents his party.
-
Comment on The cognitive dark forest in ~tech
indirection (edited )LinkI think this alludes to the Target pregnancy scandal, which was debunked. People have been saying "AI can predict human behavior" since that article (from 2012). Unprofitable disasters like the...The platform will know your idea is pregnant far before you will.
I think this alludes to the Target pregnancy scandal, which was debunked.
People have been saying "AI can predict human behavior" since that article (from 2012). Unprofitable disasters like the Metaverse, unappealing advertisements, and lack of strong evidence make me skeptical.
Before LLMs, a company couldn’t just absorb your idea and ship it. Ideas needed programmers, and programmers worked in meat-space-and-time, i.e. they were a limited resource, expensive and slow.
Before LLMs, big companies absorbed and shipped ideas from smaller companies after launch: this famously happened in 1998 with Watson/Sherlock for Mac OS. Another example is Trader Joes ripping off smaller brands, by faking interest in stocking to get product samples. Nowadays, I most frequently see big companies steal smaller companies' ideas via acqui-hire (e.g. OpenClaw and Moltbook).
I think it's unlikely a big company would steal an idea before it demonstrated success; such an idea may not succeed at all, and big companies tend to avoid risk, plus there are many other promising ideas. Meanwhile, acqui-hiring or outright buying a rising startup is relatively cheap for a billion-dollar company, though life-changing for the startup founder(s); most deals are in the single- to triple-digit millions.
-
Comment on Nvidia CEO declares AI could start, grow, and run a successful technology company worth more than a billion dollars—excerpt from Lex Fridman Podcast in ~tech
indirection LinkHere's where they discuss AGI in the transcript and video. Specifically, Lex defines AGI as "able to essentially do your job", then clarifies he means Jensen's job by saying "run, no, start, grow,...Here's where they discuss AGI in the transcript and video.
Specifically, Lex defines AGI as "able to essentially do your job", then clarifies he means Jensen's job by saying "run, no, start, grow, and run a successful technology company that’s worth...more than a billion." And Jensen says yes, then..."It is not out of the question that a Claude was able to create a web service, some interesting little app that all of a sudden, you know, a few billion people used for 50 cents, and then it went out of business again shortly after". Then Jensen explains that AI won't take people's jobs but assist them, because it can do some but not all of their tasks.
I think this touches the deeper issue on debating whether LLMs are AGI. AGI isn't clearly defined, and any specific definition can be subverted by something that technically fits but doesn't feel like it's AGI or vice versa. I think it makes more sense to instead figure out and debate what LLMs can and can't do.
For example, LLMs seem good at short tasks that involve surface-level concepts even across disparate domains, but struggle with long tasks that involve niche concepts (i.e. not in training data). I don't think the latter makes LLMs "not AGI", because the former tasks can require some generalization and novelty in output (LLMs can generate output that isn't verbatim from their training data); but it makes them unable to automate most real-world tasks, which require unwritten knowledge learned through experience.
-
Comment on Digg has shutdown (again) in ~tech
indirection LinkThey blame LLM spammers, yet the shutdown post is riddled with "it's not X, it's Y" and other LLM tropes. (They also blame lack of organic attention. Nobody wants another social media, especially...They blame LLM spammers, yet the shutdown post is riddled with "it's not X, it's Y" and other LLM tropes.
(They also blame lack of organic attention. Nobody wants another social media, especially one without any clear advantages.)
-
Comment on Executing programs inside transformers with exponentially faster inference in ~comp
indirection LinkFor a while I've dreamed of something like this, but with my basic knowledge of LLMs I didn't know it was feasible. Despite this, I found the article very readable (casual tone, explains things,...TL;DR
Language models can solve tough math problems at research grade but struggle on simple computational tasks that involve reasoning over many steps and long context. Even multiplying two numbers or solving small Sudokus is nearly impossible unless they rely on external tools.
But what does it take for an LLM itself to be as reliable and efficient as a computer?
We answer this by literally building a computer inside a transformer. We turn arbitrary C code into tokens that the model itself can execute reliably for millions of steps in seconds.
The model does not call an external tool. Instead, it executes the program directly via its transformer weights, producing an execution trace token by token and streaming results at more than 30k tokens/sec on a CPU.
The key technical idea is a new decoding path for execution traces that turns the model's attention lookups from linear scans into queries that take logarithmic time, enabling millions of correct execution steps inside a single transformer run.
For a while I've dreamed of something like this, but with my basic knowledge of LLMs I didn't know it was feasible. Despite this, I found the article very readable (casual tone, explains things, and lots of interactive visuals).
Can the model learn to do pruned symbolic execution? Can it learn to explain its execution trace in English? I imagine this can make LLMs great debuggers, who use it to run code while "understanding" what is happening enough to know exactly where execution deviates. Maybe this can also help improve the LLM's world model, if during inference the LLM runs a world simulation; or if the LLM can run programs with ambiguous "natural" parts, itself can become a world simulation (or any other program with seamless ML integration).
-
Executing programs inside transformers with exponentially faster inference
14 votes -
Comment on Getting global age assurance right: what we got wrong and what's changing in ~tech
indirection Link ParentDiscord can accept credit cards, at least in the US, because they are only given to people over 18.Discord can accept credit cards, at least in the US, because they are only given to people over 18.
-
Comment on Opinion piece: I am a 15-year-old girl. Let me show you the vile misogyny that confronts me on social media every day. in ~life.women
indirection (edited )Link ParentThe point of anonymity is that if an authoritarian government sees your post, they can't shoot you in the face, because they don't know it's yours. Just because authoritarian governments can form...The point of anonymity is that if an authoritarian government sees your post, they can't shoot you in the face, because they don't know it's yours.
Just because authoritarian governments can form in its presence, doesn't mean that in its absence, they can't form easier and in a worse form. Every government in history has been massively corrupt at some point: I can't imagine a government with complete control over information not eventually becoming dystopian. Maybe with the help of AI and/or a smarter division of power, we can form a government responsible enough to wield that power, but we must figure out what and how to get there.
ETA: Another issue is that it's practically impossible to criminalize harassment, because it can be subtle and only apparent to the target. One reason we have new slurs like th*t and not just w*hore is because the former get past filters set up to block the latter. Public forums have trouble simply detecting harassment; to convict someone, you can't just suspect that they're guilty, you must prove it "beyond reasonable doubt".
In the meantime, I think there should be private spaces, and getting into those without access (which may require ID), or bypassing their automated filters, should be illegal. Then, children should only be given access to these spaces. That would prevent the worst forms of abuse, and I think no bots and effective bans would significantly reduce harassment and general toxicity. But adults should be allowed to access the larger anonymous internet.
-
Comment on Ladybird chooses Rust as its successor language to C++, with help from AI in ~comp
indirection LinkThe AI tremendously helped but was heavily guided. This is what happens when you leave AI unsupervisedThe AI tremendously helped but was heavily guided. This is what happens when you leave AI unsupervised
-
Comment on The malignant degradation of trust in scientific work in ~science
indirection (edited )Link ParentThe scientific method is the general process "question, research, hypothesize, experiment, analyze, conclude, repeat". It can't be broken, although people can do it wrong, e.g., by misinterpreting...the scientific method is somewhat broken in the West (or at least, in the US), largely thanks to Capitalism
The scientific method is the general process "question, research, hypothesize, experiment, analyze, conclude, repeat". It can't be broken, although people can do it wrong, e.g., by misinterpreting their experiment and making unsupported conclusions.
I like to think of capital "S" and lowercase "s" science. The former is science culture, including academia and research branches in companies. The latter is, effectively, the scientific method, and more generally, how we discover truth through observation. Capital "S" Science is what you're referring to, and I agree it's corrupted. Lowercase "s" science is incorruptible, because the truth is incorruptible: no matter how much you believe something false, or try to convince others, it's still false (and sometimes the truth doesn't matter, but in the long run it usually does).
Something that would address both the corruption within and attacks on capital "S" Science, is training people to use the scientific method, and more generally critical thinking. People can "do their own research" and make conclusions that are reasonable and correct, if they know how. The problem is that it's very hard. Nobody is immune to bias and shortcuts in reasoning, and critical thinking isn't something you teach/learn like a mundane fact, it's something you train/practice because one must apply it every time they reason.
-
Comment on Passing question about LLMs and the Tech Singularity in ~tech
indirection LinkAnthropic employees use Claude to code Claude: https://www.anthropic.com/research/how-ai-is-transforming-work-at-anthropic. The study indicates that it's increasing code output, but I'd be very...Anthropic employees use Claude to code Claude: https://www.anthropic.com/research/how-ai-is-transforming-work-at-anthropic. The study indicates that it's increasing code output, but I'd be very skeptical, not only because it's from Anthropic but because the data is self-reported. More credible studies have demonstrated developers being less productive with AI agents like Claude Code, but I'm sure it depends on the person. Anecdotally, I mainly use Claude Code to write boilerplate, and I'm fairly confident it saves me more time than I spend writing the prompts and extra time debugging, because the boilerplate is easy to verify.
This means we may be getting to the singularity faster, but the curve isn't steep at least yet. You can't ask an LLM to improve itself 1000 times to get a super-LLM: there may be 1000 minor fixes and niche optimizations, but we almost definitely need at least one more breakthrough to get from a top-tier model (e.g. Opus 4.5) to human-level AGI, and current LLMs aren't discovering breakthroughs.
Well, actually...in a sort-of breakthrough, Erdos problem #728 was solved "more or less" fully autonomously by AI using a novel technique. The significance is discussed in the linked thread, most of which goes over my head, but my understanding is the AI still needed human assistance. More generally, cutting-edge researchers like Terrace Tao are using LLMs and reporting benefits, but (like with code) I haven't seen an explosion in discoveries, because the LLMs still need human guidance.
And something you may be interested in: what would happen if you asked Claude Code to improve a codebase 200 times? Someone did that and wrote about it. It's a short blog post, so rather than restating I recommend you read it. The tl;dr and relevance here is described well by its last sentence: "oh and the app still works, there's no new features, and just a few new bugs"
-
Comment on <deleted topic> in ~society
indirection (edited )LinkWe should forgive people for what they've done decades ago, that they apologized for and seem to be different about, regardless of their race or gender. The solution isn't to exclude white men...I find it hard to imagine that we would be having this conversation at all were Platner anything other than a fit middle-aged white guy who dresses like a stock photo of a “real man.”
We should forgive people for what they've done decades ago, that they apologized for and seem to be different about, regardless of their race or gender. The solution isn't to exclude white men from such forgiveness, it's to include everyone else. That means we should forgive Platner.
People today have issues forgiving others and accepting them despite their flaws. That doesn't mean forgetting others' transgressions or allowing them to repeat, or accepting others' flaws. See this description of "Christian love" from another thread; it should be used on bigots instead of trans people, but the method (shun someone until they repent, then re-invite them) is OK, certainly better than "cancel indefinitely". I'd actually prefer something more lenient, embarrassing and shaming people without fully excluding them; but the main point is, we can and should have zero tolerance for bigotry while accepting people who are formerly bigoted, since even close-minded small towns have zero tolerance for "sins" while accepting people who have "sinned".
The Democratic party's real issue isn't that they're too radical or too centrist, because politics isn't one-dimensional. The Democrats should be more "radical" in that they should be more blunt, confrontational, and active (as the author says). But I really think they should tolerate and forgive people much more, even if that makes them more "centrist", and that's not the same as tolerating or forgiving specific words and actions.
They don't mention but IKEA sells good carbon steel pans.