marci

I wish his llms included sanakirja* in the comparisons, which has some (human made) speed claims*.

* https://docs.rs/sanakirja-core

* https://pijul.org/posts/2021-02-06-rethinking-sanakirja#benc...

After looking at freely available SVGs of pelicans and bicycles, I have a hard time imagining what they could be using to game this.

delta.chat is email

https://delta.chat/en/help#can-i-use-a-classic-email-address...

Then maybe what you want is email?

Sprinkled with a little bit of https://delta.chat ?

And potentially a dash of iroh https://news.ycombinator.com/item?id=48542480 https://www.iroh.computer/solutions/delta-chat ?

Nothing to understand. Straight up hallucination. I could have sworn I read that they used a novel architecture where the model is dense but you could select specific layers or something at inference. reread the announcement: just said MoE. Corrected my brain's weights so thanks.

https://machinelearning.apple.com/research/introducing-third...

Seems like what Apple's going for with afm3. Their latest model that will be embedded in macOS 27 is a quantized dense 20B that only select between 1 to 4B at inference, based on the prompt, not token by token. If only they could make a 100B or 400B dense that selects ~5 to 15B...

Makes me wonnder... how much compute/storage there's in all the satellites currently in LEO combined.

This was a preview release. They haven't finish training. The Pro contains more knowledge but it probably takes longer training than flash for the smarts to kick in.

  "That’s where EMO comes in.

  We show that EMO – a 1B-active, 14B-total-parameter (8-expert active, 128-expert total) MoE trained on 1 trillion tokens – supports selective expert use: for a given task or domain, we can use only a small subset of experts (just 12.5% of total experts) while retaining near full-model performance."

https://allenai.org/blog/emo

But with Apple's AFM 3 architecture, we might end up with huge SOTA adjacent on devices with limited RAM.

They use a technique where you only load between 1B and 4B of a 20B dense model for an entire prompt run, not token by token like a MoE, and use mostly the low power ANE instead of GPU cores.

Now, imagine if/when they scale up to 100B or more? On a chip using 2W?

Don't worry. Most people spend most of their compute time on a phone, where you're ability to filter ads is way more enshitified.

I wonder where's the line between using a font and copyright/trademark infringement.

"That’s where EMO comes in.

We show that EMO – a 1B-active, 14B-total-parameter (8-expert active, 128-expert total) MoE trained on 1 trillion tokens – supports selective expert use: for a given task or domain, we can use only a small subset of experts (just 12.5% of total experts) while retaining near full-model performance."

https://allenai.org/blog/emo

Did they modify their post? I can't see who claimed that consumer hardware will be able to build most things?

when you sign an app with your personal dev account.

That's just a regular rounndabout.

I thought you were talking about this: https://www.youtube.com/watch?v=6OGvj7GZSIo

Also everything from scratch by allen.ai.

Weights, datasets, code, multiple checkpoints...

I like their FlexOlmo concept.

I think it's mostly because most cpus that can run a gpu already have parts dedicated as h264 encoder, way more efficient energy wise and speed wise.

Unfortunately, the most extreme is that it's the new normal that now, there's >0 chance that someone, whether they are a US citizen or not apparently, child or adult, can end up in a camp, with no due process.

Imagine, a llm trained on the best thrillers, spy stories, politics, history, manipulation techniques, psychology, sociology, sci-fi... I wonder where it got the idea for deception?

Their issue with the mac was the sound of fans spinning. I doubt a dedicated gpu will resolved that.

"finetune"

Not

"Train from scratch"

It is so, so long... I barely reached the middle before my brain just "Nope."

They are talking about this kind of battery replacement: https://www.ifixit.com/Guide/Fairphone+3+Battery+Replacement... . The "TV remote" battery replacement kind.

If your family members ever had to mount an ikea furniture or equivalent, they'll probably have an as easy or easier time replacing a part on a fairphone. Especially for the battery. At least for version 3 and older. I don't know for later models. If you know how to swap batteries in a tv remote, you know how on this phone.

This is the draft, not the current version.

edit: and for the current unfortunately there's only a dead dropbox link.

When they bothered compensating, it was far from market value, so no.

It makes more sense to word it like this when you take into consideration historical trends, like drowned towns for lakes or dams, highway system along redline, thriving neighbourhoods erased to create parks… often preceded by violence and little to no compensation.

In other words, sometimes, things happen in reality that, if you were to read it in a fictional story or see in a movie, you would think they were major plot holes.

Thinpkad owners/modders, probably.