Iv

The thing that gives me incomfort is the knowledge that we have a turing-complete language to manage display preferences.

I feel that there has to be a better way.

[insert meme "Always has been"]

As a foreigner I find that great to be able to know if some of the more uncommon American names are male or female.

AI companies are running against the clock to normalize training against copyrighted data.

Let me tell you the story of Google Books, also known as "Authors Guild Inc. v. Google Inc"

https://en.wikipedia.org/wiki/Authors_Guild,_Inc._v._Google,....

In 2004, Google added copyrighted books to is Google Books search engine, that does search among millions of book text and shows full page results without any authors authorization. Any sane lawyer of the time would have bet on this being illegal because, well, it most certainly was. And you may be shocked to learn that it is actually not.

in 2005 the Authors Guild sues for this pretty straightforward copyright violation.

Now an important part of the story: IT TOOK 10 YEARS FOR THE JUDGEMENT TO BE DECIDED (8 years + 2 years appeal) during which, well, tech continued its little stroll. Ten year is a lot in the web world, it is even more for ML.

The judgement decided Google use of the books was fair use. Why? Not because of the law, silly. A common error we geeks do is to believe that the law is like code and that it is an invincible argument in court. No, the court was impressed by the array of people who were supporting Google, calling it an invaluable tool to find books, that actually caused many sales to increase, and therefore the harm the laws were trying to prevent was not happening while a lot of good came from it.

Now the second important part of the story: MOST OF THESE USEFUL USES HAPPENED AFTER THE LITIGATION STARTS. That's the kind of crazy world we are living in: the laws are badly designed and badly enforced, so the way to get around them is to disregard them for the greater good, and hope the tribunal won't be competent enough to be fast but not incompetent enough to fail and understand the greater picture.

Rants aside, I doubt training data use will be considered copyright infringement if the courts have a similar mindset than in 2005-2015. Copyright laws were designed to preserve the authors right to profit from copies of their work, not to give them absolute control on every possible use of every copy ever made.

The progressive end of slavery is often quoted as a possible cause for the sudden interest in moving machines.

Another important factor was interest. It is noteworthy that fast development in the steam engine came after slavery started being outlawed everywhere. Before, moving machines powered by wind or steam were just an intellectual amusement (that was actually known since antiquity).

Gotta love open source

Areva can't really bankrupt: it is backed by the French government, its main owner. It still exist and is basically a public company that critics argue is structured in order to dismantle the industrial companies that were still under public control. But nuclear energy is very unlikely to be really privatized in France. That's a very touch political subject.

I need something like that for my open manufacturing project. Openship sounds like one of the missing piece I was looking for! Thanks!

How would you argue he is secular? Last time I checked KSA had an official religion, apostasy was punished (in theory) by death and one of its main cities (Mecca) was forbidden to non-muslims.

If you know that image generators are not it, then why talk about it here? Do you get this kind of angst at every technological increment?

Apart from nuclear scientists I don't know a field where participants are as conscious of the risks as AI research.

People will generate creepy porn and fake pictures. Humanity will survive this.

Matrix is the protocol I think one should go to if Signal's requirement of phone numbers is a turn down.

It will get cheaper. On 5 years it will run on your phone

You are not a journalist and I am not a state.

> An aerial shot from 1989 of the squat enclave Kowloon Walled City, which Gibson contrasts favourably with Singapore

Huhu

Try living in both.

“psychologist/neuroscientist developing computer models to understand how intelligence works” is a very niche domain of artificial intelligence which has, for half a century, meant "the attempt to duplicate intelligence in algorithms"

Conservatives all over the world dream of setting up a chinese-style firewall in their countries. They are angry that it was first done by a clearly authoritarian regime which brings some resistance to the idea, but otherwise, in the name of fight against terrorism or child pornography, they would have set it up already.

Japan is similar. I keep marveling at the stories about the US system and wonder how it is supposed to be considered friendly for businesses...

18% of the world population, but what is relevant is the % of web users. This country denies its citizen the benefits of an open web and treats the ability to route around censorship like a bug to fix. Why give someone a seat at a committee steering something they openly want to sabotage?

More than the military budget, the reluctance to let the population access an unfiltered internet is problematic. The CCP is trying to create a separate network from the global one, this raises many questions.

I am not sure if I am cynical or humble here, but my question is "Why would anyone care?".

First, the company is hiring an engineer, not a CTO or a head of marketing. Why would they consider my input valid? Even in my field of expertise I have no idea of the size of markets or of the big players strategy there. Trying to keep afloat of the tech is hard enough.

Second, if interviewing as an employee and not a business associate, you don't really care if the company becomes the next Facebook or, as is likely, goes into oblivion in 5 years. Know your incentives: if you don't have stock, you don't care about the company's success. Want employees that care? Give them incentives.

Third, half of the founders I have met have inflated egos, delusions and/or personality issues. Never trust one who says they welcome honest feedback, especially during interviews. There is nothing in for you to gain and everything to lose.

"These questions are direct, but a company that reacts badly to them may not be a good place to work." Disagree. A company with a non workable business plan can be a fantastic place to work in as the investor money burns through the various vanity project of middle managers. There are a lot much more relevant red flags to look out to anticipate a toxic workplace.

In general, only give feedback when explicitly asked for it or once you understand the inner power dynamics of the place.

This is actually from a paper published last year:

https://www.reddit.com/r/MachineLearning/comments/pffoo8/r_m...

A few questions:

- Do some ML frameworks implement it already? - It promises up to 200x compression, is it reasonable to expect it to allow us to run GPT-3 on smaller mainstream GPUs?

Maybe half of the python code I use comes from single files. Your approach works for a module.

> If you have a large collection of mostly-dense vectors and can tolerate lossy compression, Bolt can probably save you 10-200x space and compute time.

Space. It can save space.

The main limitation of fast ML models nowadays is how much parameters you can load in your GPU memory, and these are usually matrices.

200x would allow me to run GPT-3 on my old GTX 1050.

Frameworks, please implement this NOW!

In almost all practical uses of matrix multiplication, we have rounding errors. For example, in 3D it is hard to reverse exactly a rotation and get the exact initial position back.

I don't know what amount of losses we are talking about but in deep learning, several operations don't require a crazy level of compression, and it led to some lightweight float implementations (bfloat, on 16 bits, being the most common but there are also 8 bits floats for extreme cases)

If that's really a 10-100x speed increase at the cost of a bit of loss, I am sure machine learning will love it.

Wow, such a good and useful idea!

If only that could make them support a bit more their surprisingly cheap and good robot arm, the Braccio!

Thanks I'll give it a try!

Oh adoption does not have to be that good. It is for researchers right now, to exchange results. So the only requirement is that it is possible to use it with a somewhat mature lib in python.