khalic

1. Not a tangent, it’s related to the very first content visible on that link. 2. Not a dismissal, an advise from an expert 3. Not complaining, as stated, giving an advise about the optics of using clear LLM prose on the first paragraph

I strongly recommend the authors rewrite the readme by hand. It’s kind of a snif test for how much care someone put into this project.

yeah gonna correct to first one

lidar = depth sensor

300 messages and counting, and you're the first one I see making the comment. Shame on everybody here for not doing their basic due diligence of reading the fkn article before posting or reacting

Ah, the ai panic has reached scientists, queue in the 1000s of articles about how it was about people all along and doom is coming to our civilisation because of it.

I very much hope that apple is using lidar data to determine if it's a flat surface being screened

I can only talk for Switzerland, but here you can’t systematically create a file on people without breaking privacy laws, this would probably qualify

It’s probably for legal reasons

Nobody has breached the Secure Enclave, not sure it’s doable without investing millions in direct circuit manipulation. Seems good enough

I don’t think it bothers me, as long as the audio isn’t stored… but I can definitely see why it would bother some people.

The device microphones are already always listening, this just makes use of that sound a lot more than just listening for “hey siri”.

Clickbait title…

That’s a lot of misconceptions for one paragraph…

lol somebody didn’t hear the news about Kimi and GLN 5.3 cyber

I can't wait for Qwen 27B on an ASIC

Why is the headline trying to make it sound like it's new or a big deal?

That's a terrible sensationalist website btw...

Retaliation by Hegseth for not allowing Claude to be used for weapons systems.

Fascinating read thank you

"Nobody could have predicted this"

It's getting really tiresome

Indeed, but the fact that it moves the needle at all is enough to be skeptical of the anthropomorphic interpretation of the intermediate tokens. It's waaaay too early into the discovery process to take anything for granted, even more when it's related to our very strong bias towards giving inanimate things human qualities.

Why does injecting random text increase the precision then?

Scientific papers aren't popsci articles. Constraining the environment (like with the token choice) is a basic method to reduce confounding factors...

The "toy model" part is flat out wrong, the models they use have _exactly_ fundamental mechanics then chatGPT or claude, they are used all the time for this kind of research for a reason.

I believe it because of scientific papers on the matter. It doesn't "think" that's the point, it's generating text that describes a thought process, it's related but not the same thing.

[2404.15758] Let's Think Dot by Dot: Hidden Computation in Transformer Language Models

https://arxiv.org/abs/2404.15758

I was under the impression that intermediate tokens (“chain of thought”) are _not_ a representation of a model’s logical path, with one study observing that you can replace intermediate tokens with single character chains and still get the increased precision…

Ah! verisign! Proving the world over and over what kind of scum they are

check their status page...

PSA you can switch to sonnet and compact as damage control

Are they crumbs? Why aren’t you answering?

Lol, are you saying the Erdos problems are a "crumb" that nobody bothered to empty? Are you doing a comedy routine?