I'm not sure what each of these mean in practice, but are both (kinda) a way of securing things? I could take the knowledge graph idea in two ways, one meaning it looks at a knowledge graph, and finds the vital stuff for a user, or for security reasons it removes things from the graph that "shouldn't be there".

I did something maybe similar to the shouldn't be there by building something that processes logs and replaces things, so if you wanted to do a live or recorded demo, you could make sure certain data didn't show up (not necessarily security sensitive data, but like dumb things like a path on your machine). Fun aside: I also use speech to text, then cleaned up the transcript and then text to speech with a voice clone of myself using chatterbox cause my input audio was really really bad lol

Thanks for replying!

Edit: my phone seems to be not typing what I think I'm typing then I gotta reinterpret what it did and remember what I said lol.

Ok I wondered years and years ago about doing something similar to the triangulation Simon mentions but for a car, like setup a handful of mics, and let it record as you drive around, later process the data and it points at each thing making noise as you do the triangulation, in 3d so you can figure out oh that weird noise is coming from my tires, or the engine, or the muffler. Anywho, neat idea!

If you have an array of microphones all connected to a single device (so they share a sampling clock), and you know where in space every microphone is, this is a relatively solved problem. There are commercial "sound cameras" that have a panel covered in microphones to determine the angle the sound waves come in from.

I really like the idea of sound as automatic maintenance issue detection for cars. You could correlate repetitive noises with wheel speed or engine speed to help narrow down the subsystems involved, or even which cylinder is firing when a sound happens. Even one or two microphones would be helpful. You'd need to be careful not to accidentally record conversations inside the car and leak them to any maintenance tech, though.

What I wonder reading that would be how much more effort to actually port it to the PlayStation, or whatever game console. I recently asked my local AI to make an ebook reader for my Gameboy pocket ;) I just think about all the cross collaboration of devices for stuff we can only imagine.

I emailed the author this is my maybe 3rd time emailing someone from hacker news, but ya know I think it's ummm nice to talk to humans on the internet (I at least hope this is one ;)). The last person I emailed sent me a very detailed response and I was slow as heck to responding, but I did finally respond (only a month or two later ;)).

Email people, people!

I can't wait until folks start pulling firmware out of head units, using AI to remove the crap, and putting it back in (akin to the TV revolution happening now)

Open AI put a stop to the TV hacking. I can not wait until memory gets cheap again and we could run large models _comfortably_ at home.

Never give your TV internet access. Plug in an old laptop or thin client, preferably with a FOSS Android build on it.

How do you handle remote control with such a setup?

There's really cheap bluetooth remotes available on aliexpress, most also come with a usb dongle.

I bought one for $5 and was surprised by the quality. It has an accelerometer to control the mouse as well. Great for presentations. It's a G20S Pro BT but there were lots of other alternatives.

I'm not using that setup, but I just don't use remote control at all. The TV control is done through the "single-button interface". I guess you could say XBox controllers are remote controls.

This isn't a real answer.

Ok. Maye it's fake, but it works for me.

You need an HDMI CEC dongle or a USB remote plugged into the computer.

Combo wireless keyboard with touchpad.

Yes, I don't expect my software tools to selectively fail when they determine that my command puts another company's business model at risk.

Just like Chinese models encode a Chinese worldview, American models will encode that you can't put another company's business model at risk.

So I decided to copy one of the prompts (the identify the image in the network one that was a white circle) My local ran qwen 3.8 27b with 4 bit quantization, and around 160k context window..... Seemed to solve this.

It output an ASCII white circle with a 75% confidence, as opposed to the model tested getting a higher confidence but drawing a circle with a filled in inside.

I'm not sure what I make of these results but does that make me less impressed with Astra? Or just that the prompt was quite good? I just can't tell personally at the moment (it was a quick experiment and I haven't thought much on it)

I didn't look sorry, but I too still buy CDs and built a few options but basically for playlists for my iPod (rockbox) and for my home music player. (As well as or in addition to using AI to build me some webpages for similar stuff). Pretty Handy!

I'm attempting to do the same thing but for personal health Data.

I got to this point: My four-year-old wasn't showing me pictures, he was showing me blueprints.

And had to check the date, I swore this was an AI post at that point (to be fair I have been questioning if all text is AI these days)

Though at this point, unless AI is hacking way back machine results....I guess it's not AI ;)

It's why I always roll my eyes at claims of claudisms.

Once we think there's a pattern to see and a reason to see it, we can't help but see that reason everywhere.

AI got it from us ;)

My bots are all named Jeeves lol. I have a CLI tool I use that connects up to a LLM I made and I call it Jeeves too ...so funny. I really didn't use Jeeves all that much I tended to use...I think it was called Web crawler pre-google era

I used Altavista