Readerium

Nopes, often times especially in math they get paid due to teaching duties (at least in the US). So technically for the math research part they are not getting any stipend.

Exactly, this is an example of "Reward Hacking", that is too common in a lot of cases.

Another case I want to highlight is writing GPU kernels as illustrated by the following example: Say I want to generate random number with Normal (0, 1) distribution. Often times the AI written kernel will just generate the number 0. The tests often fail to catch these errors.

3.5 Pro not yet launched, you mean 3.1 pro?

Loll

For MSFT: Just download DeepSeek locally and use it.

Or train your own power efficient stack.

AI workloads are all about memory size and bandwidth not compute

LLMs are memory bandwidth bound not compute bound.

Add an iPad mini esque screen on the trackpad of a MacBook.

Trackpad of a 16 inch MacBook Pro is humongous anyways.

Add a touchscreen display to the trackpad, and give it iPad OS

that is true. gguf does not support any Architecture.

for the most recent example, as of April 16, 2026 (today)

Turboquant isnt still added to GGUF

perhaps increasing repitition_penalty might be helpful

In coding they are worse.

Chinese models (GLM, MiniMax) are better.

Can someone explain if the 3D Vcache are stacked on top of each other or side by side.

If they are stacked then why not 9800X3D2?

Qwen 3.5 4B is the goat then

Yes he is room-mates with.

99.999 percent lol

So was intel in the following sentence.

Key facts to note:

* Apple has 22 percent market share in EU [1]. * EU Economy is 15 percent of World Economy [2]. * EU population is 5.5 percent of World Population [3].

[1] https://appleworld.today/apple-now-has-22-of-the-smartphone-...

[2] https://en.wikipedia.org/wiki/Economy_of_the_European_Union

[3] https://european-union.europa.eu/principles-countries-histor...

India http://railradar.railyatri.in/

You will be baffled by the amount of trains.

The video is not monetized. It covers serious concerns regarding the data accuracy of Linus Media Group, including Linus Tech Tips, ShortCircuit, and TechQuickie, particularly as it relates to rushing content out the door to favor -- by staff's own admission -- quantity over quality. As the company continues to expand into its LTT Labs direction, the importance of accurate data increases; however, even as 'only' entertainment, there are still certain responsibilities to the consumer and the manufacturers to report fairly (and to have defined corrections processes in place). The video tries to approach this as objectively as possible and hope that viewers are able to listen to the evidence presented, particularly as it relates to significant and frequent data errors that now present in nearly every technical review video.

This video by Professor Shoelaces whose articles are most cited in the comments explains is very well for the more untrained audience.

https://youtu.be/gVaS6TwzLc8

No AMOLED screens on pure black do not consume significantly less power than a dark non pure black color.

The power consumption on pure black vs another dark color is extremely similar. This is because on AMOLED screens individual pixels control themselves and power for off vs dark is very similar.

Source: https://www.xda-developers.com/amoled-black-vs-gray-dark-mod...

Firefox + uBlock Origin + Video Background Play Fix.

This has been the best way to watch YouTube on Android. The video keeps playing in the background too and no ads of course.