Sure, but I prefaced my statement by saying that I am only looking at a subset of "low-impact business queries and random “lifestyle” queries".

So what are the benefits with respect to local LLMs in the context I described?

Pretty good in absolute sense, and closing the gap compared dGPUs on a relative basis.

~5-7 years ago even the most high end iGPU was basically unusable for even light 3D applications.

Today, we'll be testing the larger Galaxy Book4 Edge 16 with the Snapdragon X Elite (X1E-80-100), 16 GB RAM and 512 GB of storage. Its MRSP is US$1,449.99, which is around US$65 more than the comparable Galaxy Book4 Pro 16 with the Intel Core Ultra 7 155H.

I can't believe the X Elite version is more expensive than the x86 variant. I don't understand their logic for this.

I've been waiting for the 0.27.1 update on F-Droid and there has been nothing so far.

Did not know you had to add the New Pipe repo.

Microsoft CTO Kevin Scott is of course not a reliable source due to conflict of interest and his position in the US corporate world.

If anything, the fact that he is doing damage control PR around "LLM scaling laws" suggests something is amiss. Let's see how things develop.

It's too bad they couldn't get the second disabled core in the Sempron 130 working (with the overclock). I am curious what the performance would be like in GTAV.

For the average enthusiast, CPU overclocking is dead. Unless you really know what you are doing and you have a powerful liquid cooling system, it's just not worth it to risk CPU overclocking. Modern desktop CPUs have very little headroom left.

In a sense this is a good thing, we get max performance by default.

This is so depressing.

It's like MLM / Pyramid scheme insiders set a political donation group to "promote independent entrepreneurship", while in reality they are just looking for marks.

I was curious what the "Fairshake" organization was about. This is the main text on their landing page:

Fairshake supports candidates committed to securing the United States as the home to innovators building the next generation of the internet.

Providing blockchain innovators the ability to develop their networks under a clearer regulatory and legal framework is vital if the broader open blockchain economy is to grow to its full potential here in the United States.

Fairshake is a federal independent expenditure-only committee registered with the Federal Election Commission and supports candidates solely through its independent activities.

And other than a few links to news articles about Fairshake, there is nothing more. They can't even be bothered to link to "paid" research or articles about the economic potential of blockchain. Why bother?

Progress is definitely happening. One area that I am somewhat knowledgeable about is image/video upscaling. Neural net enhanced upscaling has been around for a while, but we are increasingly getting to a point where SD (DVD source, older videos from the 90s/2000s) to HD upscaling is working almost like in the science fiction movies. There are still issues of course, but the results are drastically better than simply scaling the source media by x2.

The framing of LLMs as some sort of techno-utopian "AI oracle" is indeed a damning reflection of our society. Although I think this topic is outside the scope of current "AI" discussions and would likely involve a fundamental reform of our broader social, economic, political and educational models.

Even the term "AI" (and its framing) is extremely misleading. There is no "artificial intelligence" involved in a LLM.

I am increasingly starting to believe that all these rumors and "hush hush" PR initiatives about "reasoning AI" is an attempt to keep the hype going (and VC investments) till the vesting period for their stock closes out.

I wouldn't be surprised if all these "AI" companies have come to a point where they're basically at the limits of LLM capabilities (due to problems with its fundamental architecture) while not being able to solve its core drawbacks (hallucinations, ridiculously high capex and opex cost).