kytazo

> Running LLaMA-12 7B on a contact lens with WASM (arxiv.org) How cool is that?

500 Internal Server Error.

I use 3 banks, they all work as well. Plus they're all on a separate user profile, which makes it even more secure.

Why was this flagged?

Impressive! Glad to be able to use Revolut again. Wondering, is this a change or their end or some workaround implemented by Graphene?

My impression would be the opposite. IPv6s get constantly rotated by most ISPs.

MACs are always randomized, even when connecting to the same network. At least as far as modern devices go.

Am I wrong?

Graphene builds are verifiable, go build them yourself if you feel like.

Public code is definitely safer than binary blobs.

You don't trust the devs, you trust the public code

> Customize to your heart’s content

Ok Sir https://ibb.co/fGx3Qf55

Wait until you see AI

Yes they forgot that while trying to make a statement or something

Just happened to notice how the author contradicts himself by stating "I love to distro hop" but later in the article "I don't just like to distro hop".

A good article nonetheless.

Hey, I've been wanting to do something similar for some time now for my domestic audience.

Putting it all together is quite an effort, are you satisfied on the returns considering the time/effort you've put into it?

I would assume people using Linux more or less care to an extend

USB-PD as of its last revision can deliver up to 240W.

I assume this is a yes

Since when fluoride in water is considered a good thing by any means?

Arm based laptops weren't really a thing when apple launched apple sillicon back in the m1 days.

I suppose this isn't the case anymore

Constant SAI aren't really helping either, on the contrary...

I make heavy use of after:2023 type of filter in google

I ended up just resorting in bash with an exec bash inside my mkshrc iirc.

I switched to nushell a couple of months back and I haven't got the opportunity to rock it on android, really curious how it does.

It is supposed to support android as well if I'm not mistaken.

This is wrong in many ways. First off you don't necessarily need a high-end consumer GPU, look at llama.cpp.

Or even better look at something like phi-2. It's likely to go even lower. I'm sure there are people here who can detail more specifics.

Its 1/24 and we're already there more or less.

I picked it up right around when this got posted https://news.ycombinator.com/item?id=37576303

More or less a month ago. As of speaking, I'm almost entirely familiar with the standard API, comfortable with the language enough to implement a wrapper around openai with built-in tooling and have a configuration which closely resembles my previous zsh interactive setup.

All and all, after a month of usage, I have no regrets apart from reasonable ones like investing time into what could be categorized as personal delight but the power that lies beneath your fingertips grows exponentially with all of this functionality available out of the box.

Its been a pleasure so far and witnessing its incredible pace makes it even more exciting with pleasant surprises from features introduces waiting in every second corner

Yes, similar to how they caught them by surprise in the first place. Not buying most of it.

This is a dead horse but they keep still beating it

If we assume this is true: https://iv.nboeck.de/watch?v=K5iDUZPx60E&t=2989

Then there isn't anything in particular which makes their model(s) stand out. On the contrary, they seem rather inefficient, which is probably reflected on the inference cost this gargantuan conglomerate takes to run.

As of passing by: 666 points by luu 6 hours ago

You might want to reach for some application which serves some unique functionality without having to leave your keyboard and grabbing your phone.

Probably there are also people who'd like to get rid of their smartphone entirely so this pose as a solution to the ever growing dependency on such devices, be it exclusive bank or other apps, some forms of verification and others.

Admittedly I didn't look into it much but I assume we're talking about physical devices, which likely holds true by the cost of the subscription as well as the considerable challenge of misrepresenting a virtual device for a real one, in which case the service looses any actual appeal.

Is assuming the sequence length is directly correlated to the context window a meaningful thought?

Does this imply similar increases in context in practice?

My exact thoughts