solomatov

>Just asking out of curiosity: Why not a Ryzen 16-Core CPU + 128GB RAM and then VMs / containers everywhere?

It will be harder to simulate failure there to see how the system behaves in such a case. You could just remove ethernet cable or remove a power cable, and see how it reacts.

I don't see why this can't be done with transformers. I guess, somebody already tried doing this.

As far as I remember in RNN times, the best models were RNNs with attention. Does this thing has any attention mechanism? If it does, then it has the same problem with the O(n^2) computation where n is the window size. My understanding is that transfers are superior due to the fact that they are much faster to train/evaluate than RNNs.

It happens the same way if you train a small transformer. It learns to produce words very quickly.

>I honestly can't think of a single incidence of the patent system promoting innovation in the last 20 years.

Without patents pharmaceutical industry would be nearly impossible, i.e. it's very hard to come up with a compound but very easy to synthesize it as soon as it's known.

In my experience, it's not the case. I tried periods when I drink coffee, and drink no coffee for months. I was much more productive when I did drink some coffee.

P.S. Of course it's a sample size of one.

I am not a lawyer, but it seems the issue is the following:

- For a contract to work it should be accepted

- OSS projects which are just available online, could be download without performing or even reading a license

- One of the important clauses to shield from legal liability is limitation of liability clause, i.e. you use the software as is, and won't be able to make the author accountable for damages caused by software

- MIT license has such a clause

- We now have a precedent when a developer is held liable due to damage allegedly caused by such software

Consequences of this precedent are super scary

As far as I see it, that was his employment contract. Good for him. Companies can't retroactively change employment contracts in case of contingencies, and that's very good for all of us.

(I am not a lawyer).

In addition to security and privacy I have two more concerns:

- Replacement of nice physical controls with touch screens and touch panels. It's so nice to have tactile feedback when you turn a knob compared to pressing some virtual button.

- Introduction of new failure modes which might make devices almost unusable without finding their instruction and following obscure reset procedures, i.e. you turn on some settings, and it's not so easy to turn it off.

I suspect that the main problem with these systems is that Google doesn't know how it will affect Google revenue, i.e. if there's a possibility of destroying a lot of revenue without replacing it with larger revenue. I.e. it's what's known as an innovator's dilemma.

I suspect that other methods of CPU segmentation, i.e. RAM channels, and PCIe lane are artificial in a similar way, i.e. cost of producing a die for a CPU is more or less the same, but pro CPUs have more of them.

Then, the competition is working as expected, i.e. Ryzen had ECC unofficially for some time, now Intel has it. There're plenty of other ways to segment users, i.e. Memory Channels, PCIe lanes, etc.

Naive RNN have vanishing gradient, but LSTMs and GRUs are much better in this respect.

My understanding that LSTM is a kind of RNN.

My understanding is that RNNs aren't worse than Transformers per se, they are just slower to train, and use GPU much more efficiently, i.e. much more stuff could be run in parallel.

There's definitely huge value in the system, but there's a lot of work to do to make it even more valuable. Hope they will get close to the premium product soon. Would be happy to pay for it even in its current form within reasonable price range.

IANAL, but the US federal law roughly consists of USC, i.e. US code written by congress, and CFR, code of federal regulations, written by executive branch of the government. They can't write anything they want, only using powers given to them by congress in USC.

If anyone wants to read the rule as written by FTC, could do it here: https://www.ftc.gov/legal-library/browse/federal-register-no...

This document only briefly mentions interior mutability, which IMO, is one of the most important things to become productive in Rust.

@nine_k shared this https://programming-language-benchmarks.vercel.app/wasm-vs-r... in the comment tree. The results are pretty bad. You could lose 2x or more in cpu perf. There're cases where wasm is pretty close to native though.

Thanks a lot. The difference is huge unfortunately :(

But how large is this gap? I.e. 10%, 50%, 100% or 1000%. Size of this gap affects trade offs of using WASM vs native code a lot.

> It's also like 2x as slow,

Do you have links to recent benchmarks? My understanding is that there were investment in bridging the gap recently.

Does anyone has any links to recent benchmarks which show how large is the gap between WASM and native code?

In theory WASM looks really great, but the last time I looked at it, the gap was big enough to be a concern.

Good to see that it's on the apple side and not on the side of my router configuration. It's really frustrating, and I have it several times a day.

Though, I wouldn't use their solution with the script.

It's not just a pointer, but a pointer + metadata.

See this paper for details: https://www.cl.cam.ac.uk/research/security/ctsrd/pdfs/201904...

> The extremely common scenario where you include a photo on your blog or social media post that isn't yours, is virtually never fair use.

Yep. There's a lot of case law, and there're a lot of details, so if your business relies on fair use, you need a very good lawyer (I am pretty sure the companies using AI for creative purposes have them).

P.S. I am not a lawyer

Concerning my opinion, we need to use criteria which separates fair and not fair use. It should be the same as with human artist. If you use others' work as inspiration or in a transformative way it's fair use, but if you plainly copy others' work it's a copyright infringement. We need to teach this to ML models like we teach it for artists.

P.S. I am not a lawyer.

>There are websites that offer free images but either require attribution of the source of the image or else require paying a premium membership or subscription.

There're no prerequisites for fair use. I.e. licenses don't apply in such cases.

You could read more about it here: https://en.wikipedia.org/wiki/Fair_use

P.S. I am not a lawyer.

Sine there're a lot of discussion in this tread. My opinion about category theory and programming is:

- You don't need category theory to be a programmer (or good or 10x or 100x programmer)

- It makes sense to learn category theory because its beautiful like it makes sense to study arts, literature or philosophy. Not because it useful, but because it gives you pleasure.