terhechte

If you have 128gb ram, try running MoE models, they're a far better fit for Apple's hardware because they trade memory for inference performance. using something like Wizard2 8x22b requires a huge amount of memory to host the 176b model, but only one 22b slice has to be active at a time so you get the token speed of a 22b model.

The logic is called every frame

There're multiple solutions like this and I've used some of them over the past years.

- There's obviously the fantastic Magit (https://github.com/magit/magit) which is an Emacs Plugin but you can configure your Emacs start up just with Magic and nothing else so that Emacs is only used as a TUI Git client. I did this for a while.

- There's GitUI written in Rust (https://github.com/extrawurst/gitui) I did use this for a long time but recently switched over to LazyGit for the better Vim bindings and having more features

- LazyGit (https://github.com/jesseduffield/lazygit) is what I'm using right now and I'm mostly happy

I actually wrote my own in C some years ago called Gitsi (https://github.com/terhechte/gitsi).

One thing that I added that (as far as I know) none of the others have and I sorely miss is VIM number based movements. So you can say 4j and jump 4 selections down. This makes it much faster (for me) to jump to the one file I'd like to commit. I ultimately stopped developing Gitsi because I didn't have the time to implement all the features others had readily available.

I do prefer TUI based Git clients to full blown GUI apps because of the keyboard movement. So I can quickly enter do something and exit, while staying in the terminal

Back when computers took up a whole room, you'd also have asked: "but what exactly is this useful for? B-Roll some simple calculations that anybody can do with a piece of paper and a pen."?

Think 5-10 years into the future, this is a stepping stone

I don't think you can do that with Claude. That's what people use uncensored local models for.

You could also try Hyperdeck which uses Markdown for slides as well, but supports most of the animation features of Powerpoint as well as MathML and stuff like that (no vector editing though)

http://hyperdeck.io

Most devs use windows (https://www.statista.com/statistics/869211/worldwide-softwar...). Reddit llocallama alone has 250k users. Clearly the market is bigger than 1000 people. Why are gamers and Linux people always so aggressive diminutive of other people’s interests?

I've been wanting to learn Godot for some time now. Being able to develop on the iPad is supercool (for side-project game devs). In addition, you can also now run Godot on the Meta Quest (https://www.meta.com/en-gb/experiences/godot-game-engine/771...). Also, as Xogot runs on iPad and VisionOS can run iPad apps, it might even run there (haven't checked). That's a lot of fun platforms for playing around with it.

The paper doesn't really say that. Summary:

> The study concludes that while individuals may initially sleep more when given extended opportunities—likely to recover from prior sleep deprivation—they do not sustain excessive sleep beyond their physiological needs. Consistently sleeping more than 10 hours may indicate underlying health issues and warrants further investigation.

So more than 10 hours might indicate a problem, not 8-9 hours. Also, it's different if people work out a lot or have a physically demanding job:

> For athletes or individuals with high physical demands, 10 hours of sleep might not indicate an underlying health issue but rather an increased physiological requirement. Research has shown that elite athletes often sleep longer than the general population. (this is not from the paper but from a quick search)

I tried to build something kind of like NotebookLM (personalized news podcasts) over the past months (https://www.tailoredpod.ai), but biggest issue is that the existing good TTS Apis are so expensive that a product such as NotebookLM is not really possible for a normal company that doesn't have internal access to Google's models. OpenAI has the cheapest / quality good enough TTS Api, but even then generating hours of audio for free is way too expensive.

Open Source TTS models are slowly catching up, but they still need beefy hardware (e.g. https://github.com/SWivid/F5-TTS)

but that doesn't run the official Netflix, Disney+, Apple+ etc apps, does it? Or via something like Waydroid?

This is kind of the best way to do software development using an iPad. You can SSH or VNC into the Pi and then use all the tools available to the system. I've thought more than once about gluing a Pi to the back of my iPad so I can just take one device on longer travels and not have to take the iPad and the Mac (reason for the bringing the iPad is that all big streaming services typically have an iPad app that allows downloads, but no macOS app. I like having downloads for no / bad wifi situations).

This goes opposite to the saying "Experts say it cannot be done; amateurs accomplish it every day.".

Sometimes it's good to have someone with fresh eyes looking and something and not be shaped by decades of prior history.

This kind of argument, together with the "Stochastic Parrot" argument falls (in my opinion) under the category of dangerous half knowledge. Somebody read the wikipedia article on LLMs and thinks they know inside out how they work. Whereas the engineers working on these models say that they don't really know why they work / what it is that is manifesting in the different layers during training.

Yes, LLMs predict the next word, but so do we. I, at least, when I reflect on how I form sentences, also start with something and the next most likely word comes out next. You can see that happening a lot when somebody talks and then stops (searching for a word) and another person helpfully suggests the word they're searching for. If our way of talking was completely different - no next word prediction, another person shouldn't even be able to do that.

I'm not claiming that current models have feelings or a deep understanding of the subject matter. What I'm saying is that there's (high likelihood) more going on here then just the simple statistical trick that this article and others like it tend to focus on.

No. It's good for a couple of hours, but I much prefer my 27" screen at home for a whole workday. I haven't tried the same experience on the Vision Pro yet because that's not something Apple lets you try out when you book the free demo, so I can't compare, but I'd guess it will be similar because at some point you get tired from the weight on your face.

I'm using a Quest3 for coding when I'm travelling with the Immersed app. It is a bit of a setup, but it allows using a USB-C cable for the connection. It also can open multiple virtual screens. I usually have 4 screens open. The resolution is not as good as the VisionPro but having 4 screens in a slightly lower resolution allows me to be able to read everything quite well.

How would that technically work? A film editor clearly wants to edit their recordings in Premiere or Final Cut or perform color operations. You can't do that on DRM encrypted movies. This comment makes no sense as it implies that movies are recorded in one take and then the output is a DRM encrypted container that is submitted to Apple.

Everything new is initially very expensive. However nothing prevents people from taking these movies and downscaling them for the Quest 3 lenses so you can watch it on a cheaper hardware. Current 8k 180 or 360 content for the Quest is really bad quality. While the Quest displays are lower quality (and I think the Snapdragon is also more limited than the Vision), downscaled content should still be far better than the movies we have right now.

https://old.reddit.com works

They have a deep hatred for Meta, so Llama is out

Not sure what this is, but it looks fishy. The mentioned `bin/generate_voice.py` actually does nothing:

https://github.com/OpenT2S/LlamaVoice/blob/main/bin/generate...

̶H̶o̶w̶ ̶a̶r̶e̶ ̶y̶o̶u̶ ̶u̶s̶i̶n̶g̶ ̶i̶t̶?̶ ̶I̶'̶m̶ ̶t̶r̶y̶i̶n̶g̶ ̶M̶u̶l̶l̶v̶a̶d̶ ̶a̶s̶ ̶a̶n̶ ̶i̶O̶S̶ ̶V̶P̶N̶ ̶a̶n̶d̶ ̶I̶ ̶d̶i̶d̶n̶'̶t̶ ̶g̶e̶t̶ ̶i̶t̶ ̶t̶o̶ ̶w̶o̶r̶k̶.̶ ̶D̶i̶d̶ ̶y̶o̶u̶ ̶b̶u̶y̶ ̶a̶ ̶s̶e̶p̶a̶r̶a̶t̶e̶ ̶s̶u̶b̶s̶c̶r̶i̶p̶t̶i̶o̶n̶ ̶w̶i̶t̶h̶ ̶a̶ ̶U̶S̶ ̶c̶r̶e̶d̶i̶t̶ ̶c̶a̶r̶d̶?̶

Nevermind, I deleted and re-installed the app on iOS while on VPN and now it works!

Did someone figure out if it works when I use a VPN?

The reason they do the serial coding is to reduce the incentive for stealing iPhones. Due to locking, you can't really sell a stolen iPhone. Instead, thieves break it apart and sell the parts. Serial coding disincentives this. I know many people that had their iPhone stolen and it is an awful experience.

I'm a simple man, I'd never have the ability to solder and fix my own phone. Instead of repairing it myself, I'd always go to an official repair shop. I'm not benefitting from non-coded parts. On the other hand, reducing the likelihood of people stealing my phone is a huge upside for me. So for me, and people like me, these measures that Apple takes are a net benefit. I understand that for tinkerers it is a different equation - but you have to agree that for the majority of iPhone owners the equation might also tip in favour of not having to worry about stolen phones instead of being able to fix their own hardware (which most normal people also would never do).

There's examples on the homepage, and there's a link to the Kaggle notebook in the article.

https://arcprize.org

I build a personalized newsletter / podcast that picks relevant news based on your preferences and comes every morning.

https://www.tailoredpod.ai

Just posting to say that I agree with the parent comment. All my sites have so much additional stuff that I couldn't just rely on somebody else's hosting even if your hosting is going to be fast and use use CDN to cache images. Maybe I want to host on Vecel to integrate their analytics or their other plugins, or I want to host on Github pages to link it to my repo. Or I want to move additional files in there and be able to access them via SSH.

Offering hosting is good, but if I pay for the service not being able to download as html / css feels like being held hostage. Instead, you should be up front about it (your software is a editor) and offer things like auto push of changes to GitHub so that I don't have to manually export.

There are ways around that. Apple & Google can embed a private key into the Secure Enclave, sign every image with this key, and at the same time write the hash onto a fast public ledger (e.g. Solana). This validates that this specific image was taken at that specific point in time. Doing it that way would validate the image. That would lead to a world where only these "vetted" images can be trusted and untrusted devices (e.g. Nikon cameras) can take nice pictures but no "truthy" pictures. But it would be an escape hatch.

I use various LLMs for all kinds of search queries. Sure, there is a danger of hallucinations, but for most queries I can tell whether is wrong because now I can make a much clearer Google or Kagi search because I better know how to formulate my search.

Also, more than once have I done a Google or Kagi search where most answers I found also are or were wrong.

I really don’t get the kind of people that hate on LLMs because of “hallucinations” (or worse, ideological hate, easily identified by their use of the “stochastic parrot” term). I find them genuinely useful in delivering better search results quicker. I also don’t have to wade through wades of SEO optimized shit.

Just today I wanted to know the Croatian word for “Orange” and a quick GPT “orange in Croatian” delivered faster and more concise than google