Most openclaw users are not running the models locally.

This is what I thought. The iMessage integration makes sense I guess though.

The entire marginal cost to serve AI models is paid for by the API costs of all providers by nearly every estimation. The cost not currently recouped is entirely in the training and net-new infrastructure that they're building.

And the open source models are only months behind, so the big AI companies need to keep burning money on R&D with no end in sight. If OpenAI took a quarter off from model development, they might fall behind forever.

So why are they banning people from using it in systems like claw?

The reason Macs get recommended is the unified memory, which is usable as VRAM for the GPU. People are similarly using the AMD Strix Halo for AI which also has a similar memory architecture. Time to first token for something like '1+1=' would be seconds, and then you'd be getting ~20 tokens per second, which is absolutely plenty fast for regular use. Token/s slows down at the higher end of context, but it's absolutely still practical for a lot of usecases. Though I agree that agentic coding, especially over large projects, would likely get too slow to be practical.

We are getting into a debate between particulars and universals. To call the 'unified memory' VRAM is quite a generalization. Whatever the case, we can tell from stock prices that whatever this VRAM is, its nothing compared to NVIDIA.

Anyway, we were trying to run a 70B model on a macbook(can't remember which M model) at a fortune 20 company, it never became practical. We were trying to compare strings of character length ~200. It was like 400-ish characters plus a pre-prompt.

I can't imagine this being reasonable on a 1T model, let alone the 400B models of deepseek and LLAMA.

With 32B active parameters, Kimi K2.5 will run faster than your 70B model.

Here's a video of a previous 1T K2 model running using MLX on a a pair of Mac Studios: https://twitter.com/awnihannun/status/1943723599971443134 - performance isn't terrible.

Is there a catch? I was not getting anything like this on a 70B model.

EDIT: oh its a marketing account and the program never finished... who knows the validity.

I don't think Awni should be dismissed as a "marketing account" - they're an engineer at Apple who's been driving the MLX project for a couple of years now, they've earned a lot of respect from me.

Given how secretive Apple is, oh my, its super duper marketing account.

Jeff Geerling and a few others also got access to similarly specced mac clusters. They replicated this performance.

The tooling involved has improved significantly over the past year.

Not too slow if you just let it run overnight/in the background. But the biggest draw would be no rate limits whatsoever compared to the big proprietary APIs, especially Claude's. No risk of sudden rugpulls either, and the model will have very consistent performance.

Animals become a fine red mist when presented with these sorts of forces. The train feels a bump, but will not crash. I'm unsure at what size a rock will cause issues, but I would expect in most cases they would be kicked away by the train without issue, if a person can move them.

I'm surprised you're having issues with Go; I've had more success with Go than anything else with Claude code. Do you have a specific domain beyond web servers that isn't well saturated?

This is absolutely not an objective review. The person who wrote this is a very particular type of person who Alpha School appeals strongly towards. I'm not saying anything in particular is wrong with the review, but calling it unbiased is incorrect.

Because there have been different FDA food pyramids since then. The one people popularized hasn't been the recommendation for decades

The 90s food pyramid lasted until 2005, so decades is just about correct. Then it was some myolate something or the other.

But people used the 90s food pyramid everywhere and that was the only one popularly known. The myplate stuff, I guess it wasn’t advertised well by the government, who knows.

Chip factories need years of lead time, and manufacturers might be hesitant to take on new debt in a massive bubble that might pop before they ever see any returns.

I bought a ReMarkable 2 a couple of years ago with far higher hopes of hackability than it ultimately ended up supporting. Ended up selling it a few months ago after it just couldn't fit into any of my usecases.

I think ReMarkable is wasting a TON of potential at their price/form factor/ux. A device can be powerful without sacrificing simplicity and singularity of purpose.

Have you seen these? https://usetrmnl.com/developers

I was thinking about picking one up and giving it a shot.

cheers, im on the TRMNL team.

looks like reMarkable runs linux so in theory our OSS lib would do the trick: https://github.com/usetrmnl/trmnl-display

which can then point to one of several OSS server clients we offer: https://docs.usetrmnl.com/go/diy/byos#implementations

I did end up buying one! I really love it, and though I haven't hacked much on it, I really haven't needed to. If it's something you're interested in, you won't be disappointed.

I have been rocking a Daylight computer for about a year now. It is my primary mobile device, and I am writing this very comment from it. I highly recommend it.

That's not an e-ink device though, right?

I was trying to find an e-ink tablet, amazon kept recommending me the magic notepad from xppen. It looked good, but I wasn't sure what that cryptic "x-paper display" was. The wording is just vague enough to make you think it's an e-paper display, without committing to that detail.

It took going through comments to find out that it's not an e-ink display.

The Daylight computer seem like that too. So what do you think of the display? Is it just another LED screen, or does it approach e-ink in any way?

I really hated how they marketed that tablet. Some weird statements about how it will transform your life, all while making meaningless comparisons about framerate by just not acknowledging the difference between e-ink and transflective LCDs (to the point I found it intentionally misleading).

Might still be a good product though, of course.

It's selling point is a lack of features. There's no web browser, no Instagram, no Facebook or Slack. No messaging. Just a digital piece of paper. No distractions. If you give me a way to get reddit on there, the device is ruined.

That's a clear case of PEBTSAC [1] as it is not the device which decides whether those mentioned time wasters are able to do their master's bidding but the person using the device. Sellers of 'premium' devices which are specced-down so as not to be able to run those things are guilty of the same crime as sellers of 'light' products which replace nutritious ingredients with water or air while selling at a 'premium' price. Just eat less of the good stuff instead of being suckered into paying more for less. The same goes for these 'premium´ devices which only do one thing and often do that thing badly.

[1] s/Keyboard/Touch Screen/

> same crime as sellers of 'light' products which replace nutritious ingredients with water or air

The analogy, in this case, would be replacing NON-nutritious ingredients with water or air.

I didn't want a web browser either. I wanted access to my calendar, or some way to set the lockscreen to my calendar. I wanted a live syncing folder of images/pdfs that wasn't tied to the subscription remarkable walled garden. I wanted a way to read rss content, instead of setting up a complex automation to sync things over ssh.

That wasn't its selling point for me. I installed KOReader on mine and have been quite happy with the result, but expect to move to a PineNote in the near future, as I'm tired of jumping through someone else's hoops to control my own hardware.

It's BS.

It doesn't support non-Latin languages, not even having a keyboard for them. Its handwriting recognition barely works, and it lacks a good system to organize notes.

The hardware is awesome, but their software is terrible.