Timwi

No. Just because the AI beats you, you don't automatically know how it did it. You can't easily learn from it just by observing it.

I'm not the one you replied to, but I think I get their point. If you have a machine that can solve math problems at the push of a button, you don't really need to amass discoveries anymore just in case someone might need it years down the line.

Someone in the thread mentioned Minkowski’s math work being instrumental to Einstein’s physics. If Einstein had had a math machine that can spit out the result at will, he wouldn't have needed Minkowski.

Nevertheless, I also see the opposing point: if Minkowski hadn't already published his results, it's possible Einstein would not have even had the inspiration to derive general relativity from it. I think this is the human aspect most critics are worried about.

Why would they stop trying?

If it could do this while properly crediting the researchers (the “giants”) it would be a different matter.

By saying it “isn't that hard” you're admitting that it could definitely be easier.

It could be argued that Turkish and simplified Chinese succeeded simply because they were supported by, you could even say imposed by, a government that controls > 90% of the population speaking the language. Neither was the case for Arabic: there is no single government spanning that proportion of Arabic speakers, and from how I read the article, it seems no government really supported Unified Arabic to any extent even remotely comparable to Turkish and simplified Chinese.

Your comment could be read as implying that Unified Arabic, the subject of this post, must have been useless as an idea because it wasn't widely adopted.

In reality, there are many factors, not only pure practicality, that lead people to reject reforms like these. In the case of Arabic, it was deeply held traditional values; in the case of Hong Kong and Taiwan, well, I imagine it's simply an unwillingness to appear to be aligning with a regime one politically disagrees with.

Computers do not “have that mostly covered now”. Typesetting and font rendering technology is so Western-centric and so entrenched that support for writing systems with slightly different base assumptions, such as Arabic, Devanagari and many others, is always an afterthought.

It's so weird for me to live through software evolution to the point where this no longer goes without saying.

In the Windows 3.1 days, it was nearly impossible to make a program that wasn't fully keyboard-usable. Even if you left out all the hotkeys in the menus and messed up the tab order, it was still possible (even if annoying) to get through it with the keyboard, and adding the hotkeys (as well as actual shortcuts) and setting a sensible tab order was super easy and straightforward.

Nowadays more than half of all software is made using non-native GUI toolkits (or sometimes no toolkits at all) that refuse to implement any of the typical keyboard navigation that would previously have come for free.

The keyboard makes computers a joy to use for me. Seeing it disappear makes me very sad. Seeing how the vast majority of people don't care that it disappeared makes me feel like an alien from another planet.

1. Make sure that the tree view control has screen reader support and that it has all the normal keyboard commands (such as right/left arrow for expanding/collapsing a parent node).

2. Make sure that the text box has a label that says “search” or “filter” and not just a magnifying glass icon. You can only get away with an icon if the GUI toolkit allows you to give it an alt text.

3. Make sure that this label has a hotkey (underlined letter) and that it's properly linked to the textbox so that pressing the hotkey actually focuses it.

4. Make sure that the tree view is next after the textbox in the tab order.

5. Make sure that it's obvious how to cancel a search. You've already assigned Enter to performing the search (as opposed to closing the dialog) which means you can't use Escape because that is already required for closing the dialog. I think one way is to let the user empty the textbox and press Enter to clear the search, but you can also have a separate button as long as it has a hotkey and an alt text if it's just an icon. This kind of button is allowed to sit between the textbox and the tree view in the tab order, but make sure it stays that way and you don't add 10 more buttons there so navigating to the tree view becomes arduous.

I think it's ironic that I can give all of this advice just by being a keyboard user. I'm not visually impaired and I don't use a screen reader.

I don't think of it as units (as the sibling comment pointed out, angles are dimensionless); I think of ° as a postfix unary operator that does the conversion. In other words, I read sin(x°) as a shorthand for sin(x*Pi/180).

Why are math libraries consistently treated as space hogs if you don't keep them as barebones as possible? I've had sinDeg/cosDeg overloads for a while and I don't see why they can't be built into the base library; the same would go for a sinT/cosT that takes turns instead of radians or degrees. Why do we have to pick one and meticulously avoid having multiple?

Do you not want AIs to adhere to your wants? That's what we build agentic systems for. Calling it “cheating” when they do something we don't want them to do is 100% spot-on.

Except these AIs are both scary and smart. I think the characterization is warranted.

The whole point of calling it “cheating” is to warn us that if we tell an AI to do something, we need to take into account that it may well do something we don't expect; something that we as humans dismiss as “cheating”.

Someone in this thread gave a good example: if you ask an AI to get a reservation at a restaurant, and make sure the reservation is in place before exiting the agent loop, you don't expect the AI to hack the restaurant and erase someone else's reservation to make space for yours. But an AI will absolutely do that.

I suppose that may go some of the way, but the article did mention that the auditor AIs weren't sufficient to catch all of the cheating.

Companies are not “ignorant of the basics of their purpose”. The basics of their purpose is to make money and increase value for shareholders. There is nothing in their core purpose that benefits you as a user or that says they shouldn't annoy you.

Is the concern that the AI will rephrase the text during proofreading, thus falsifying a copy-pasted quote? I'm not sure how such a change could go unnoticed, unless you blindly publish the AI output without looking at it, which doesn't sound like something someone would do who wants to “write their own words”.

Image GenAI is diffusion-based, and I would say the image GenAI in Claude, Gemini and ChatGPT are all “in mainstream use”.

I've heard of attempts to use diffusion models to generate text or code as well, but my impression is that it just didn't yield the level of results necessary to dethrone a frontier transformer model.

> No. The court specifically determined that the Internet Archive was guilty of unauthorized copying.

You're not wrong, but you're treating “guilty of unauthorized copying” as a statement of physical fact when in reality it just means it falls under an arbitrary rule invented by humans (namely, the law that defines unauthorized copying). This rule is ambiguous at its edges because it's not written as an algorithm or equation. It was perfectly reasonable for Kahle to believe that the rule can be interpreted in a way that it wouldn't apply and, by dragging it through the courts, have that interpretation be made the established one.

Even though the court has now established a competing interpretation, it is still not unreasonable to ask whether the law is fair and just under this interpretation. I feel that it isn't and should be changed.

In addition to the sibling comment, which I agree with [edit: the one starting with “people who generate entire blogposts with claude probably do not care about substance, tho”], I want to add another point. Even when the substance is key, the delivery has a quality to it that encodes a useful signal.

To illustrate what I mean, imagine two blog entries on the same subject, but one is written by an expert on the subject and one by a layman. As humans, we can generally (not perfectly) tell which is which. The expert will write in a certain quality that comes with expertise and is genuinely difficult to fake without it.

LLMs disrupt this pattern. LLMs are able to write with the expert’s quality even while writing bullshit. I don’t mean quality as “measure of goodness” here but merely a set of traits or properties. Humans reading LLM text are far more likely to misjudge the author’s level of expertise.

The next step is unscrupulous humans exploiting this to trick unsuspecting readers into misjudging a text, and now you have an internet where you can’t trust anything anymore, at least not at first glance. Even relatively discerning readers now have to waste time reading more of the text before being able to dismiss it as having substance of little value.

So, as I see it, the upset is not simply from AI provenance of a text, but from this level of dishonesty and subterfuge, coupled with the helplessness with which we’re exposed to it.

I mean you're not wrong, but consider that a lot of high-level developers treat low-level details such as CPU architecture, memory virtualization, and even garbage collection as essentially magic. That's not the same thing as LLMs because those things I listed are carefully crafted to be reliable abstractions, but I can understand the impetus to want to think of LLMs in a similar way.

Appreciation of the arts getting compared with greed of billionaires ..... How the times have changed.

Cybersecurity researchers who let loose worms or ransomware?

Car mechanics who break things so you come back for more business?

Airline pilots who... hijack their own flight?... Hm

No, people tried to sell “prompt engineering” services well beyond that — especially for image generating AIs, which where believed to need extra instructions like “professional lighting”, a name of a camera model, lens parameters, etc.

That depends. Your interpretation is “em+en+em”. But another, potentially wittier interpretation is “em in em”, i.e., you split an em-dash into two en-dashes and put an em-dash inside.

I think we're getting at the heart of your misunderstanding. You seem to feel that we're running out of codepoints in Unicode (the “limited number of shelves” analogy). This could not be further from the truth. Although the number of codepoints is technically finite, it is extremely large (over a million) compared to how many we need (currently only 160,000 are defined, and for obvious reasons the rate of new additions is slowing down). The Unicode roadmaps main page[0] displays only 5 of the 17 available planes because the remaining 12 (each 65,536 codepoints large) have not yet had any characters assigned to them whatsoever.

[0] https://www.unicode.org/roadmaps/

That's one way to telling us you don't read professional writing without telling us you don't read professional writing.

The comment about your keyboard is ironic given the nerd forum we're in. The em and en dashes live on my keyboard alongside the hyphen because I'm a techie and I know how to put them there. My Android keyboard has them too and I didn't even need to put them there myself.

I'm in my 40s and I also prefer to call people on Discord, Telegram, Signal or (if no other option available) WhatsApp. The phone dialer is just inconvenient, while the above-mentioned chat apps have all the contacts listed and easy to access. There's also a slight worry about being charged for every phone call and every SMS and having to worry about your phone plan and texting plan separately from your data plan; maybe that's not a thing anymore but it used to be really worrying, so it gives me more peace of mind to only have to worry about the data plan and nothing else.

I tried on both phone and desktop and couldn't find that alleged setting either.