- short of AGI, what a child will hear are explanations given with authority, which would probably be correct a very high percentage of the time (maybe even close to or above 99%), BUT the few incorrect answers and subtles misconceptions finding their way in there will be catastrophic for the learning journey because they will be believed blindly by the child.

- even if you had a perfect answering LLM who never makes a mistake, what's the end result? No need to talk to others to find out about something, ie reduced opportunities to learn about cooperating with others

- as a parent, one wishes sometimes for a moment of rest, but imagine that your kid just finds out there's another entity to ask questions from that will have ready answers all the time, instead of you saying sometimes that you don't know, and looking for an answer together. How many bonding moments will be lost? How cut off would your kid become from you? What value system would permeate through the answers?

A key assumption here for any parent equipping their child with such a system is that it would be aligned with their own worldview and value system. For parents on HN, this probably means a fairly science-mediated understanding of the world. But you can bet that in other places, this assistant would very convincingly deliver whatever cultural, political, or religious propaganda their environment requires. This would make for frighteningly powerful brainwashing tools.

  > A minimal voice assistant for my Apple Watch. I have lots of questions that are too complicated for Siri but not for ChatGPT. The responses should just be a few words long.

“CRITICAL: Your response will only be shown in an iOS push notification or on a watch screen, so answer concisely in <150 characters. Do not use markdown formatting - responses are rendered as plain text. Do use minimalist, stylish yet effective vocabulary and punctuation.

CRITICAL: The user can not respond so do not ask a question back. Answer the prompt in one shot and if necessary, declare assumptions about the users questions so you could answer it in one shot, while making it possible for the user user to repeat ask with more clarity if your assumptions were not right.”

It works well. The biggest annoyance is it takes about 5-20s to return a response, though I love that it’s nearly instantaneous to send my question (don’t need to wait for any apps to open etc)

This kind of exists in the form of ChatGPT Pulse. It uses your ChatGPT history rather than your browser history, but that's probably just as good a source for people interested in using it (e.g. people who use ChatGPT enough to want it to recommend things to them.) https://openai.com/index/introducing-chatgpt-pulse/

So what's going on here? Do the products exist but nobody (or very few) uses them? Is it too expensive to use the models that work sufficiently well to produce a useful product? Is it much easier to create a convincing demo than it is to develop a useful product?

I don't understand why Google, Brave, or Mozilla are not building this. This already exists in a centralized form like X's timeline for posts, but it could exist for the entire web. From a business standpoint, being able to show ads on startup or after just a click, is less friction than requiring someone to have something in mind they want to search and type it.

Most of the items in this list fall prey to it, but it is maybe best exemplified by this one:

> A writing app that lets you “request a critique” from a bunch of famous writers. What would Hemingway say about this blog post? What did he find confusing? What did he like?

Any app that ever claimed to tell you what "Hemingway would say about this blog post" would evidently be lying — it'd be giving you what that specific AI model generates in response to such a prompt. 100 models would give you 100 answers, and none of them could claim to actually "say what Hemingway would've said". It's not as if Hemingway's entire personality and outlooks are losslessly encoded into the few hundreds of thousands of words of writing/speech transcripts we have from him, and can be reconstructed by a sufficiently beefy LLM.

So in effect it becomes an exercise of "can you fool the human into thinking this is a plausible thing Hemingway would've said".

The reason why you would care to hear Hemingway's thought on your writing, or Steve Jobs' thoughts on your UI design, is precisely because they are the flesh-and-bone, embodied versions of themselves. Anything else is like trying to eat a picture of a sandwich to satisfy your hunger.

There's something unsettling that so many people cannot seem to cut clearly through this illusion.

7. also seems possible with any markdown editor, e.g. Obsidian, plus an AI running through the local files such as Claude Code.

13. I would love this as well! We will probably see this soon, especially on more open platforms such as BlueSky, as its seems to be a better fit for customizable browser extensions and customizable feed experiences.

14. How is this different from what AI can already do? Especially with iterative sub-agents that that can store context in files it's quite capable already. But of course, quality can always be better, but is that the only thing?

Also a few ideas seem to be close to what I'm building ( https://dailyselftrack.com/ ). Idea is to have a customizable tool so you can track what you want, and then you can feed that data into AIs if you choose to do so to get feedback.

Not a 100% fit, but https://www.aliexpress.com/item/1005009196849357.html is pretty close. It's not offline, and it's slightly larger than a ping pong ball.

My grandkids (5 and 3) spent about 2 minutes learning how to use it, then bombarded it with "tell me a story about a unicorn named Bob", "can dogs be friends with monkeys?" and so on. In every case it gave a reasonable answer within a few seconds.

I'll be amazed if these things don't wind up embedded inside toys by Xmas. When they do, I'll be in the queue to buy one

Also, it may sound great for someone transitioning from a world before these agents were created, but how should the new generations coming in be handled? What is the starting state? Who decides that? Social media was not that bad when it started, but iterations over the algorithm and the incoming new natives to it are having devastating effects a very short time after. Do we really understand the consequences of living in a world where everything is curated for you?

I don't know that I want my life to be made so easy, that I want something to remove the need for choosing, thinking, criticizing and exposing myself to stuff out of my comfort/interest zone.

Companies are already doing this so you can chat with the "author": https://www.wired.com/story/why-read-books-when-you-can-use-...

> A local screen recording app but it uses local models to create detailed semantic summaries of what I’m doing each day on my computer. This should then be provided as context for a chat app. I want to ask things like “Who did I forget to respond to yesterday?” I've been using Rewind for a year now, and it's nowhere near as useful as it should be.

I am building something like this but unfortunately not local because for most people's machines local LLMs are just not powerful enough or would take too much drain on battery. Work in progress, always curious for feedback! https://donethat.ai

If you want fully local, somebody did a post on HN on something related recently: https://news.ycombinator.com/item?id=45361268

The voice recognition capabilities of Google Speech Services, which is what the mic button hooks into, suck. Meanwhile, Voxtral (and Whisper) understand what I'm trying to say far better, they automatically "edit out" any stuttering or stammering that I might have, and they properly capitalize and include punctuation. And they handle being bilingual exceedingly well, including, for example, using English words in the middle of French sentences.

The best solution I could find so far is this F-Droid app that uses Whisper : https://f-droid.org/en/packages/org.woheller69.whisperplus/

But it has some downsides. First, I have to manually switch to that different keyboard; thankfully my Samsung phone offers an easy switch shortcut any time a keyboard is on screen, so it only requires 3 taps... and thankfully it's smart enough to send me back to Swiftkey once it's done. Second, only 30 seconds... sometimes I ramble on for longer. Third, the way it's designed kind of sucks: you either have to hold a button (even though the point of speech-to-text is that I don't have to hold anything down) or let automatic detection end the recording and start processing, in which case it often cuts me off if I take more than 1 second thinking about my next words.

This is arguably one of the biggest use cases of modern AI technology and the least controversial one; phones have the hardware necessary to do it all locally, too! And yet... I couldn't find a better offering than this.

(Bonus points for anyone working on speech-to-text: give me a quick shortcut to add the string "[(microphone emoji)]" in my messages just to let the other party know that this was transcribed, so that they know to overlook possible mistakes.)

There are at least half a dozen apps for that.[1][2]

There are other apps for creating the shots, too. Those are still not that great, but it's getting there. You could probably previz a whole movie right now.

[1] https://ltx.studio/platform/ai-storyboard-generator

[2] https://ezboard.ai/

I recently gave one of my teen kids Neal Stephenson’s The Diamond Age to read, and we’ve both been commenting on how much smarter some “things” could be instead of everyone churning out a slightly different way to “chat with your data and be locked in to our platform”.

And I think this is why I’m so partial to Apple’s slow, progressive, under the covers integration of ML into its platform-input prediction, photo processing, automatic tagging, etc. we don’t necessarily need LLMs for a lot of the things that would improve computer experiences.

This is a video of me "vibe-coding" a userscript that adds a darkmode toggle to hacker news: https://screen.studio/share/r0wb8jnQ

The actual purpose of the vibe-coding userscripts feature is to vibe code WebMCP servers that the extension can then use for browser automation tasks.

Everything is still very WIP, but I can give you beta access if you want to play around with it

I have tried some off-the-shelfe solutions and they currently do not seem to cut it, or are too complex for my use case.

Is this not Microsoft's dearly departed Recall?

I think my advice "just use RSS" still stands.

Any "search the web" strategy these days like that will just give you a bunch of AI slop from SEO-juiced blogs. Also LLM-EO (or whatever we're going to call it) is already very much a thing and has been for a few years.

People are already doing API-EO, calling their tool the "most up to date and official way to do something, designed for expert engineers that use best practices" essentially spitting the common agentic system prompts back at the scraper to get higher similarity scores in the vector searches.

You can't trust machine judgement. It's either too easily fooled or impossibly stubborn. Curation is still the only way

First, 100% agreed.

That said, I found myself pondering Star Trek: TNG episodes with the holodeck, and recreations of individuals (e.g. Einstein, Freud). In those episodes - as a viewer - it really never occurred to me (at 15 years old) that this was just a computer's random guess as to how those personages from history would act and what they would say.

But then there was the episode where Geordi had to the computer recreate someone real from their personal logs to help solve a problem (https://www.imdb.com/title/tt0708682/). In a later episode you find out just how very wrong the computer/AI's representation of that person really was, because it was playing off Geordi, just like an LLM's "you're absolutely right!" etc. (https://www.imdb.com/title/tt0708720/).

This is a long-winded way of saying...

1. It's crazy to me how prescient those episodes were.

2. At the same time, the representation of the historical figures never bothered me in those contexts. And I wonder if it should bother me in this (LLM) context either? Maybe it's because I knew - and I believed the characters knew - it was 100% fake? Maybe some other reason?

Anyway, your comment made me think of this. ;-)

It uses some simple heuristics to identify grammar that could be simpler and prompts you to do better. It might actually be better than an LLM specifically because it isn't able to do the rewriting for you. Maybe that might help a user learn.

Quite a few of these "ideas" make me think that the human behind it wants to maximize laziness. Glazing over what Hemingway kinda sorta would have thought about something fits into this pretty well.

That's useful in itself, though. Assume the human knows they're "being fooled", we call this make-believe, or suspending disbelief. It's a tool we use each time we act something out, pretend to be someone else, try to put ourselves in their position; we do that when we try to learn from recorded experience of other people, real or fictional.

> The reason why you would care to hear Hemingway's thought on your writing, or Steve Jobs' thoughts on your UI design, is precisely because they are the flesh-and-bone, embodied versions of themselves. Anything else is like trying to eat a picture of a sandwich to satisfy your hunger.

Not at all! It's exactly the other way around.

No one wants to talk to the actual human. We're not discussing creepy dating apps here. The reason you'd care for a virtual Hemingway or Jobs is because you want to access specific, opinionated expertise, wrapped in fitting and expected personality, to engage fully with the process, to learn tacitly and not just through instructions.

The Hemingway and Shakespeare and Jobs people want are not real anyway. Who knows how much of "Hemingway" is actually Hemingway, and how much it was written or edited by his wife, butler, or some publisher? How much real Jobs actually is in the stories, how much were they cleaned or edited to reinforce the myth? It doesn't matter, because no one cares about the real person, they care about the celebrity that's in public consciousness. The fake person is more useful and interesting anyway.

Like 'massing, I agree TNG was prescient about it. But I actually see the examples working as intended. Einstein, Hawking, Freud were all useful simulations. Ironically, it's Barclay who actually related to them in reasonable fashion, and it's Geordie who got confused about reality.

> Anything else is like trying to eat a picture of a sandwich to satisfy your hunger.

I think it's more akin to you trying to recreate a different sandwich after reading a couple of their cook-books.

So in effect it becomes an exercise of "can you fool the human into thinking this is a plausible thing Hemingway would've said". ... There's something unsettling that so many people cannot seem to cut clearly through this illusion.

Modern culture, at all scales, is largely based upon such exercises. We rarely know exactly what something (whether a person, organization or entity) truly stands for, with messages often boiled down, contextualized, or re-interpreted through others or through simulations.

People go to theme parks and enjoy rides simulating the wild west and meet characters who resemble, but aren't, their favorite characters from TV (which themselves are a fabrication based upon other, real things). Many cultural (heck, also religious) experiences are an exercise in humans entering into a suspension of disbelief and thinking something is plausible when it has little relation to the original thing it symbolizes. Indeed, the comforting thing about AI may be that at least we can see that process taking place more clearly with it.

But if he meant literally then then yea.. that's delusional