Agraillo

> ... because we are constantly recording and reporting everything that’s going on

Why stop just here? It's AI, right? So it’s all about prediction - next token, next action, it doesn’t matter. Let’s turn Minority Report the movie into reality: when you’re angry at your boss while commuting home, we will watch, predict and prevent your future bad behavior.

I just thought that almost all existing LLMs are already able to do this with the following setup: using an alias "@you may speak now," it should create a prompt like this: "Given the following questions {randomly sampled or all questions the user asked before are inserted here}, start a dialog as a friend/coach who knows something about these interests and may encourage them toward something new or enlightening."

Reminded me of many movie plots where a derailed character sits in a taxi and, when asked where to go, replies with "anywhere" or "I don't know." But before imagining a terrible future where an AI-driven vehicle actually decides, I suggest imagining an AI-infused comedy exploring this scenario. /s

> science has nothing more to say about it

Science has also given us time machine surrogates - photos and videos. I understand this doesn’t work for everyone and can sometimes make things harder, but for me, seeing our common past brings back the smile to my face.

Thanks for sharing. Two companies come to mind: Strix for kettle controllers and Shimano for bike gears. Maybe they don't fit exactly to the Hidden champions category because they’re not very hidden from the public (many manufacturers mention their names on final products, assuming consumers might take that into account). So the criteria for “hidden champions” could be more flexible imo

Strix became less hidden for me personally after listening to The Life Scientific interview with John Taylor [1]. There is plenty of fascinating information, probably because Jim Al-Khalili is a great scientific interviewer. Recently, I recalled it in the context of AI, self-driving, and safety. Strix controllers have a second level of protection if the main automatic shut-off circuit fails. That’s probably why we never hear of fires or other incidents due to a failed Strix controller.

[1] https://www.bbc.co.uk/programmes/b0b42z87

ECH is on by default for Cloudflare’s free plans, and paying customers can adjust the setting. That’s why CF already has an interesting history with the Russian authorities [1] (The discussion is short but has a lot of interesting details)

[1] https://news.ycombinator.com/item?id=44392221

> Maybe LLMs can be sued to design such a thing

nice misspelling (or a joke?), related to all the lawsuits around LLMs.

Joking aside, it’s already there in a sense. Several times I started with a brief outline of what the prototype should do (an HTML/CSS/JS app), and sure enough, refinements and corrections followed. When the final version worked more or less as expected, I asked the LLM to create a specification (a reproducing prompt) of everything we made together. Even if the vibe-coded prototype is dropped, the time wasn’t wasted, I probably would never have come to the same bullet list specification without having an actual working app at my disposal to test and evaluate. So paradoxically this specification even might be used by a human later

I like challenges like this. First, the edit that introduced the "five-seven ascii" is [1] (2010) by Pete142 with the explanation "add a name for the PDP-6/10 character-packing convention". The user Pete142 cites his web page www.pwilson.net that no longer serves his content. Sure it can be accessed with archive.org and from the resume the earliest year mentioned is 1986 ( MS-DOS/ASM/C drivers Technical Leader: ...). I suspect that he himself might have use the term when working and probably this jargon word/phrase didn't survive to a reliable book/research.

[1] https://en.wikipedia.org/w/index.php?title=36-bit_computing&...

Related: Why is there a “small house” in IBM's Code page 437? (glyphdrawing.club) [1]. There are other interesting articles mentioned in the discussion. m_walden probably would comment here himself

[1] https://news.ycombinator.com/item?id=43667010

It's a great aspect to evaluate (fiction books/movies), thanks for mentioning it. I think it's much easier to use as an evaluation tool than techniques like the apple example. One of the tests, for example, is to recall a book that you have never seen a movie adaptation of and try to remember the characters and scenes. For me, in these cases (when I try to recall), the characters appear faceless, while places are more detailed, but they usually remind me of some real places I have encountered before in my life.

It's interesting that if non-aphantasia people are so common, I wonder why so few paintings have scenery based solely on imagination. I even remember asking a person who paints (not in the context of this condition) how hard it was for him to paint something not directly before his eyes, but from imagination, and why he didn't do it more often. I recall that he definitely did this (painting from imagination) rarely or not at all, and the question really puzzled him

The latter, there's an article about this particularly [1]

[1] https://www.nytimes.com/interactive/2024/09/03/technology/zo...

> ... a topic that's worth only 1-3 minutes of your time

It's even worse sometimes, googling some "how to" queries returns links to yt-videos. Even if the video is 5 minutes, it's a waste of time, because I'm usually in the middle of an ongoing process when dozens variants are evaluated and an average dedicated time for a single one is much shorter.

Transcripts sometimes help. But not the native (no diarization as long as I remember). An example, Lex Fridman podcast is a good source of anecdata from famous science/tech/non-tech people and provides good transcripts on the site (but only starting some point in the past). For transcripts before this point v1.transcript.lol covered many, but amongst other glitches no names for diarization (Speaker 1/ Speaker 2).

> finds that they're not credible enough to generate an answer

The credibility is one side of the story. In many cases, at least for my curious research, I happen to search for something very niche, so to find at least anything related, an LLM needs to find semantic equivalence between the topic in the query and what the found pages are discussing or explaining.

One recent example: in a flat-style web discussion, it may be interesting to somehow visually mark a reply if the comment is from a user who was already in the discussion (at least GP or GGP). I wanted to find some thoughts or talk about this. I had almost no luck with Perplexity, which probably brute-forced dozens of result pages for semantic equivalence comparison, and I also "was not feeling/getting lucky" with Google using keywords, the AROUND operator, and so on. I'm sure there are a couple of blogs and web-technology forums where this was really discussed, but I'm not sure the current indexing technology is semantically aware at scale.

It's interesting that sometimes Google is still better, for example, when a topic I’m researching has a couple of specific terms one should be aware of to discuss it seriously. Making them mandatory (with quotes) may produce a small result set to scan with my own eyes.

> It's just making sure you're not accidentally passing invalid data or performing an invalid operation.

To understand that this is a big deal, one sometimes needs to face the relaxed rules of their static language of choice. Delphi (and its ancestor Turbo Pascal) allows passing untyped pointers for parameters defined as typed. This behavior is actually controlled by a setting that is probably relaxed (allowed) by default. It seems it was done for convenience when using pointer arithmetic. At some point in the past, I also started logging categories of bugs (an ad-hoc taxonomy) and, if some kind of bug appeared more than once, to do something about it (like encouraging a new habit, rule, or something similar). So, after passing wrong pointers twice in a short period of time, I made it mandatory for all sources to have only typed pointers, which required some additional typing but saved me from similar bugs further on.

The edit did exist. So the phrase "Multiple studies have found a left-wing bias at Wikipedia" can be found last in the edit (4 February 2025) [1] and was removed by the editor Aquillion with the explanation [2]

     Not seeing anything in this source that supports this language; they summarize no other sources that I could see, and their own conclusions are more complex than this (as covered further down the article.)

I think that the one who previously introduced the phrase should have either not stated 'Multiple studies' or provided information about other studies. I suspect that a single research is usually not enough to be mentioned in an article

[1] https://en.wikipedia.org/w/index.php?title=Ideological_bias_...

[2] https://en.wikipedia.org/w/index.php?title=Ideological_bias_...

Hmm.. Is this my brain inference or the parent comment author' too: "Bright side of life" -> "Always look at the bright side of life" from "Monty Python's Life of Brian" -> plot was taking place during Roman times (mentioned in the GP comment)

I felt also something like this before. Also integers seem pretty close to the reality around us. One of their functions is to symbolically represent the similarity of objects (there might be a better way to put it). Like, if you see 5 sheep in one group and 6 in another, after that point they’re no longer just distinct sheep with unique properties - the numbers act as symbols for the groups. Real numbers still can work in the brain, but they're most distant from the world around us, at least when it comes to going from visual to conceptual understanding.

> I feel it almost every day for myself and for my family in the US

Based on your use of the word 'fill,' the Gallup World Poll seems like a more appropriate ranking because it comes directly from the people. Even your dissatisfaction as a resident of the US is likely reflected in their analysis [1]. However, the situation in China is somewhat more complicated at present. Telephone and face-to-face interviews were permitted until 2023 [2], but the 2023 survey already relied only on web interviews, which correlates with reports of Gallup withdrawing from China [3]. This likely brings us back to the question of different faces of freedom - specifically, the ability to conduct surveys or polls without being accused of distorting the country’s image.

Both Freedom House and Gallup methodologies may be flawed, but without them or similar we would have to rely solely on anecdotal evidence, which is insufficient imho

[1] https://news.gallup.com/poll/660440/land-free-fewer-american...

[2] https://www.gallup.com/file/services/177797/World_Poll_Datas...

[3] https://archive.ph/v5F9a (US consultancy Gallup withdraws from China (FT))

Let's start with the Freedom House index [1]

    Finland 100/100
    .....
    United States 84/100
    .....
    China 9/100
    ......
    Turkmenistan 1/100

Before claiming a hidden agenda or conflicts of interest of those behind the organization, first evaluate their methodology [2]. You may even agree with the methodology but argue that the actual numbers were calculated incorrectly

> the ability to walk outside without ever having the chance of randomly shot dead

This is a good measure of risk perception, and you can even use it as one of the freedom scores. The problem is that it is hard to measure. For example, consider if this measure is based on a poll: if an act of terrorism occurs just before the poll is taken, the score will definitely drop. However, in countries where such news is heavily suppressed (e.g., the former USSR), the scores may appear perfect.

[1] https://freedomhouse.org/country/scores

[2] https://freedomhouse.org/reports/freedom-world/freedom-world...

> I've not seen anyone talking about the significance of the set of countries where this is rolled out to first: Brazil, Indonesia, Singapore, and Thailand

This was really interesting and somewhere there was a comment/quote that these countries are affected most with the malware distributed with side-loading, I can't find this comment now. But while trying to find some information, I found the info about 2023 Alphabet/states $700 m. settlement. It came mostly unnoticed on HN [1] (two posts, 2 comments), but there is interesting timings coincidence in the settlement text ([2])

  ...6.9.2 For a period of at least four (4) years from the Effective Date, Google will maintain the following functionality in Android version 14+ for Mobile Devices:
  (a) Google will support APIs that enable sideloaded app stores that have received User consent to install apps to avoid automatic updates taking place while the User is using the app....

2023 (settlement) + 4 years = 2027 (mentioned for other countries). This can be related to apps like F-Droid, this ruling might prevent Google from making F-Droid comply if the US was announced to meet the new rules earlier (before 2027). There are other formulas that might end up 2026/2027 when calculating so to be on the (legal) safe side, Google probably made US join later. Probably those countries are also for beta-testing both in the technical and legal sense.

The settlement might be interesting in other other respects also. Even the forces (the states, U.S Attorney) that drove the suit in 2021-2023 might join here though during this admin it's really questionable.

[1]: https://news.ycombinator.com/item?id=38691926

[2]: https://www.oag.state.tx.us/sites/default/files/images/press...

The idea for "Severance" was supposedly inspired by Dan Erickson's difficult experiences with jobs that he disliked. If what you are suggesting is true, then we will have an alternative way to achieve a similar effect as the characters in the series—simply ask the agent to make you work without your brain participation :)

Sorry if I didn't recognize your sarcasm, but if you’re serious, you’re probably also assuming that rooting is usually done for criminal activity. In fact, both rooting and easy app creation/side-loading are often tools to solve inconveniences. I didn't plan to root my last phone until I encountered some restrictions in the manufacturer’s version of Android that couldn’t be resolved without rooting.

Regarding the topic, I can easily imagine a legitimate app on Google Play with available source code, where you find something inconvenient and your attempts to suggest a fix to the developer did not lead to the desired outcome. Currently, you or your developer friend can simply fork such an app, fix the issue, and release it for the general public without any extra paperwork. This Google policy would make such a developer suspicious/disabled by default (if the developer is not already verified), unless proved otherwise.

Reworded from [1]: Earlier this year Meta tried to acquire Safe Superintelligence. Sutskever rebuffed Meta’s efforts, as well as the company’s attempt to hire him

[1] https://www.cnbc.com/2025/06/19/meta-tried-to-buy-safe-super...

I finally also recalled this project (tiddlywiki), but CMIIW, the statefulness of the hyperclay is only for developers, the end user will get the same conventional html. Without some kind of common solution/protocol/standard on the browser side that would allow persistence it's not so exciting. Theoretically there might be some kind of simple protocol that would save html file versions on the server based on the cookie, but there are so many ways this can go wrong

> I've tried to get ChatGPT to generate original fairy tales and whatever plot prompt I give it, it spits out what is essentially the same dull story every time.

Not a universal solution, but a working method to get at least sometimes interesting results. You should use it as a co-authoring tool by following these hints: treat this as a dialogue ("let’s create interactively, you and me…", "create a first sentence of a fictional story", ...), where you act like a semaphore for the continuation—judging the current output and either correcting it or suggesting the next step (which can be brief and expanded later by the LLM). Finally, try to suggest unexpected constraints. This can be effective because when your constraint contains a set of words rarely seen together in training data, the output becomes somewhat random but still at least partially grounded in reality.

An example from one of my old conversations with Llama 3.1:

> User: Create a sentence from a fictional book containing the words crazy, cowboy, and gadget.

> Llama: In the wild west town of Crazy Horse, a notorious cowboy named Buckshot Bob unveiled his latest gadget — a mechanical horse that could gallop at breakneck speed.

The management thinking from your example seemed very natural and not so surprising after all. They just needed someone with a "just in case" mindset, similar to the tailor from one of my favorite scenes/movies "The Hudsucker Proxy" - "The Double Stitch" [0]. I even sometimes act like that character. For example, when the boss says we will drop a feature, I don’t fully kill it, instead, I temporarily disable it because the chances of it coming back are not zero.

[0]: https://www.youtube.com/watch?v=LM5DiHYfKfM

As a cultural synthesizer ("create a sketch...") or translator (i.e., from human to programming) perplexity is as good as the others and maybe worse. But I like it as being an alternative semantic web search engine. Search engine because most of the time the facts are referenced at the end of the paragraphs and when it fails to insert them, I usually ask to expand. It's a new feeling after all those years of PageRank dominance. For example, It sometimes creates a paragraph with a reference to some recent research that according to the PageRank logic should be at the bottom (because almost nobody cites it). It's like the relevance was reborn. Once it was in a primitive sense of the early AltaVista, then Google killed it with PageRank, now when an engine really understands the question, the answers might become relevant in a new sense while being tangential to the reputations of the sources

I'm not sure whether the Apple Music API is so primitive, but for Pink Floyd listing Led Zeppelin and not (for example) Porcupine Tree/Stephen Wilson is a bit strange. Though the site lists "related artists" that sounds a little different to "similar artists"

I find Wolfram's computational irreducibility is a very important aspect when dealing with modern LLMs, because for them it can be reduced (here it can) to "some questions shouldn't be inferred, but computed". From recent tests, I played with a question when models had to find cities and countries that can be connected with a common vowel in the middle (like Oslo + Norway = Oslorway). Every "non-thinking" LLMs answered mostly wrong, but wrote a perfect html/js ready to use copy/paste script, that when run found all the correct results from the world. Recent "thinking" ones managed to make do with the prompt thinking but it was a long process ending up with one or two results. We just can't avoid computations for plenty of tasks

It's interesting. The question from the paper "Darling, please be honest: have I gained weight?" assumes that the "socially acceptability" of the answer should be taken into account. In this case the problem fits the "Wickedness" (Wikipedia's quote is "Classic examples of wicked problems include economic, environmental, and political issues"). But taken formally, and with the ability for LLM to ask questions in return to decrease formal uncertainty ("Please, give me several full photos of yourself from the past year to evaluate"), it is not "wicked" at all. This example alone makes the topic very uncertain in itself