momenti

It's so valuable because knowing lots of people's passwords is useful for mass surveillance, cyber attacks, industrial espionage etc.

I suspect the change we are seeing is that Google is now using a neural network to (re-)interpret the search query.[1] Presumably that neural network also calculates some sort of neural hash for document/paragraph/sentence similarity. The upside is that it can correct more typos, intelligently drop irrelevant terms and understand the meaning of the query to some extent. But the downsite is less precision when you know exactly what you want. It sometimes even seems to ignore the quotes syntax for exact string matching. Very frustrating and very poor quality control.

[1] I bet they train these models based on unsuccessful queries as inputs (in which the user did not click any search result), and then the final search query after which the user left as desired output.

You can also enforce most of these settings in all profiles with a policy: https://github.com/mozilla/policy-templates

What does it help if the creepy intelligence services are in the hardware?

The 1st-gen Zero had the same resonance cavity antenna on-board,[1][2] but it had a nicer design for the external antenna, namely a 0-Ohms(?) resistor that could be desoldered and resoldered at an angle (and be reverted if need be),[3] rather than crudely cutting the trace.

[1] https://i.redd.it/fido28zdv1331.jpg

[2] https://www.embeddedcomputing.com/technology/analog-and-powe...

[3] https://www.briandorey.com/post/raspberry-pi-zero-w-external...

You are not the first "genius" realizing this. Look up Ted Kaczynski. But he was smarter than you because he considered what this reality entails for the ruling classes: Namely the decision whether they shouldn't give up on maintaining the masses once technologically feasible. It is for this reason that it is a taboo and has to remain one. Everything else is propaganda and chatter.

> no matter how you want to define "generalisation"

... and how you want to define "task". For some prompts/"tasks", GPT-3 does generate impressive (more than trivial) outputs that cannot be found on the internet and that are indistinguishable from what a human would respond, so it generalizes in that sense. Maybe human ingenuity and generalization is also just slightly perturbed interpolation? It is very difficult to produce something truly novel, so we are also rather tightly limited by prior experience. Who knows? Also, who cares if submarines swim? Anyhow, it seems 50% of the internet is bikeshedding about definitions.

> ANNs generalise atrociously badly

This statement is meaningless without controlling for model complexity and data type. For their simplicity, ANNs generalize well on a wide variety of data. GPT-3 yields almost human-level generalization ability for some tasks.

I also clarified that the generalization is probably not far. There is not much complex "realism" to be found in low-to-mid-level features; they're almost mathematical in their simplicity, similar to basis functions.

The title is click-bait then because it is not only unit/pixel-level noise, but also "noise" at the gestalt-level (e.g. large geometric shapes).

It is basically an experiment how the mid-to-low-level features in NNs generalize from random abstract generated images to real images.

This is not a very surprising result, because it is known NNs generalize fairly well and at these levels you mostly only have blobs and edges that do not look much different from artificial ones.

I think they lack talent to really do it competitively. Korea has crazy amounts of cheap talent!

The comparison is also flawed in that Korea at least makes cheap and good stuff.