GeekyBear

The on-device models that Apple distilled from Google's server based models are already available to developers to use in their apps at no cost.

We're in a time where devices are becoming more capable and developers are figuring out how to do on-device inference using less resources.

The companies that spent a fortune on AI data centers might as well have built a huge pyre of money for all the "moat" they built with it.

Mac revenue being up 24% vs last year makes me think that users have figured out that they can just plug a Mac Mini into the monitor they already have.

In return for Google getting a discount on how much they pay Apple to be the default search engine in Safari, Apple has the right to distill Google's Gemini server models and to create their own models that fit on-device.

> Apple is cutting down Google Gemini's massive models into smaller and more secure parts through distillation, to create elements that are more suited to on-device Apple Intelligence processing.

https://appleinsider.com/articles/26/03/25/apple-is-distilli...

Apple doesn't pay Google.

Google gets a small discount how much they pay Apple to be the default search engine on Safari.

It's on most of YouTube streams because Google forces it on devices that lack hardware support, even if it breaks the functionality of the device.

Mainstream support doesn't have to be forced.

If the common hardware customers already have fails when you attempt to use a newer codec (but works fine otherwise), that can't be said to be a mainstream codec yet.

Sorry, but YouTube began thermal throttling on a Firestick the moment they started forcing AV1 regardless of the presence of on-device hardware support.

The M7 generation is said to be another huge advance on local LLM performance, so Apple is skipping the M6 Pro and M6 Max to bring out the M7 generation early.

A name like M5 Pro denotes the generation and the number of cores.

What does a name like Snapdragon 8 Elite Extreme Gen 6 even mean?

> without 3rd party software

At least you can finally acknowledge reality.

Also, VLC over a wired connection and streaming from one Mac to another using WiFi (AirPlay) are built into the OS.

There is literally nothing stopping you from using a modern iMac as an external display using a thunderbolt connection today.

If you're OK with video compression artifacts, you can even do it over a WiFi connection.

In a world with the chips themselves named things like:

Ryzen AI Max+ Pro 495 or Snapdragon 8 Elite Extreme Gen 6.

Your distaste seems misplaced.

The iMacs before the 5k version did support Target Display mode.

They dropped it when the cable and connector standards of the day didn't support a display that big.

That's not laziness, it's a technical limitation.

Now that we have newer high bandwidth connection standards, you can once again use a newer iMac as an external display over a thunderbolt connection.

President Obama was the President of the United States when he went on late night TV and assured everyone that the government didn't spy on Americans at all, after Wyden's warnings and the Snowden revelations.

https://www.npr.org/sections/thetwo-way/2013/08/06/209692380...

> President Obama defended the U.S. government's surveillance programs, telling NBC's Jay Leno on Tuesday that: "There is no spying on Americans."

https://www.npr.org/sections/thetwo-way/2013/08/06/209692380...

I seem to remember that Senator Wyden's public warnings (over a decade ago) that the government was spying on Americans was one reason for President Obama to go on national TV and lie to Americans.

Senator Wyden has a long history of warning the public on surveillance abuses.

There are third party apps like Luna Display that do this now.

You can't continue to support Target Display mode when the connectors and cables of the day don't yet support 5k displays.

If you have an older 5k iMac, there is an inexpensive third party hardware kit to convert it into a display with modern connectors.

https://www.youtube.com/watch?v=0VXQu0fAc8s

DeepSeek designs models that require less hardware resources for inference.

Remember when people thought that board level repairs were a perfectly normal thing for repair shops to do?

Until very recently, the low power version of RAM had to be soldered to the motherboard.

If you want the battery life improvements, bear in mind that there is a hard limit on how large a battery you are allowed to fly with, so low power components are important.

> But in the search for even thinner laptops

The LP in LPDDRx isn't about creating thinner devices.

The LP stands for Low Power and is aimed at making devices that draw less power resulting in a longer battery life.

SRAM doesn't shrink as well as other types of circuitry do when process nodes advance.

They didn't allow unlimited cloud photo storage for all those years out of the goodness of their hearts.

That was all used as training data.

The last two non-evil employees at Google.

The neural engine is optimized for energy efficiency, and the GPU is optimized for performance.

You're going to get much better battery life and much less heat running on the neural engine.

How many 5080 16 Gig cards would you have to buy to run the same model, and what would that cost?

Is there a naming scheme dumber than Ryzen AI Max+ Pro 495?

Touch support was added this year.

https://support.apple.com/en-us/102597

The High Yield YouTube channel has a good early look at die shots of the new A20 chip used in the iPhone Duo and discussion of the changes from the prior generations.

https://www.youtube.com/watch?v=t8bxkCKNiuA

They have a similar video comparing the M6 to the M5.

https://m.youtube.com/watch?v=uA7WLmlg1kM

Apple research has also posted information on the topic.

> In this post, we show how to run the latest frontier-class model — Muse Spark Open, with nearly a trillion parameters — right from your desk, using the latest version of MLX, released today. On a cluster of four Mac Studios, Muse Spark Open generates over 36 tokens per second

Beginning on a single Mac Studio (up to 512GB of unified memory), then scaling out across multiple Macs with the RDMA over Thunderbolt driver, below we share an overview of the range of models you can now run on Mac Studio.

https://machinelearning.apple.com/research/mlx-distributed-i...