More accurately, you are fine until you try to distribute them, or display them. Giving out free copies would not be commercial, but would definitely be a violation.
Most attacks are remote, not involving physical access to a specific (compromised) device. Think a compromised website exposing 1000s of peoples passwords, and trying those passwords against other services. For that threat vector, it is more secure.
The way I would look at it is if that threat concerns you, don't use it for high-value accounts (email, bank, etc). Still is probably worth using for all those other low-value accounts; if an attacker has your phone, and has broken in to it, them having access to your Netflix account is probably the least of your worries.
I believe the other big thing is it allowed you to parallelize the training as well, so you could split those colossal training sets across many machines in a way you could not do with LSTMs.
On the one hand, yes. On the other hand, if they are both going to have liability issues, choosing the one that isn't literally full of lawyers seems like a smart play.
Most still believe that "really intelligent AI" is still a long way off, from what I have seen. Many have started to believe there can be a lot of harm caused by the systems well before then, however.
I think that was more because places like Reddit just linking directly to the image meant they couldn't serve ads to pay for it. So they needed to make an incentive for people to directly visit the site, and the most obvious way was to just cut out Reddit as a necessary component to a user discovering more images.
Bard would not trick anyone into thinking it was sentient, yet something they have supposedly did. I just think Google has far more to lose than Bing, so they are being more cautious.
Kaniko is what I use as well, since I initially had some containers that were not building correctly with buildah. Not quite as seamless as buildah promised to be, but it can read Dockerfiles with no modifications, so it was easy enough to move to.
I remember VMWare solved this with a balloon driver, that basically asks the guest for a lot of memory at a very low priority, so it has precedence over the page cache, but under any program that actually needs it.
Same as it ever was. What is acceptable to make fun of and what is not is constantly shifting with the times. What seems more constant is people being upset that what was once acceptable is now frowned upon, and the friction that causes them.
Such as Twitch's opinion that this joke was a violation of their TOS, and thus deserving of a temp ban? That opinion seems to be making a lot of people here pretty uncomfortable.
But why? I am under the impression the Q crowd doesn't tend to be the wealthiest bunch. Tesla was also already selling every car they could make, with like a 6 month backorder. Expanding, or switching, your target market makes no sense if they cannot afford your product, and you have no inventory to sell them anyways.
As far as I can tell, because it allows one-man production teams. A single person can do everything, without needing to find a vocalist. It allows them to try more stuff quicker and easier, and anything that happened to get popular ends up covered by vocalists anyways (and the producer gets to sell those rights).
Signal also has a grand total of 40 employees. Keeping the application running on the various platforms takes a decent chunk of their time. Difficult development to add something like federation cannot be expected to happen quickly.
You can copyright derivative, transformative work though?
I don't think current legal frameworks are equipped to handle these types of generative works. There will need to be either new laws, or at least some legal precedence established. For example, it would make total sense for stuff posted publicly to optionally have a "not to be used to train models" license. But models are being trained on public data that pre-dates anyone thinking that is something that they should worry about.
One of the more interesting things I have seen recently is the combination of different domains in models / datasets. The top network of Stable Diffusion combines text-based descriptions with image-based descriptions, where the model learns to represent either text or images in the same embedding; a picture, or a caption for that picture, lead to similar embeddings.
Effectively, this can broaden the context the network can learn. There are relationships that are readily apparent to something that learned images that might not be apparent to something trained only on text, or vis-versa.
It will be interesting to see where that goes. Will it be possible to make a singular multi-domain encoder, that can take a wide range of inputs and create an embedding (an "mental model" of the input), and have this one model be usable as the input for a wide variety of tasks? Can something trained on multi-domains learn new concepts faster than a network that is single-domain?
That is a bit of an oversimplification. The presence of money is no sufficient, that money must be in circulation. For example, they can print all the money they want, if they just toss it all in a big bank account and never remove it, there will be no inflation.
That was a big reason why printing a lot of money in the past did not contribute to a lot of inflation: most of that money ended up sitting in the accounts of a few big institutions, who then did not end up spending it on anything.
This time around, we both gave that money directly to people (who then spent all of it), but also were dealing with global supply issues, an energy crunch, a food crunch, etc. Everyone had more money to spend, but there were less things to spend it on, leading to dramatic price increases.
The reason I saw mentioned is that NSFW content was so prevalent in their training set that, if they did not explicitly exclude it, it was actually difficult for it to NOT produce NSFW results even from normal inputs. For example, if you ask for anything with a woman in it, it would probably give you a naked woman in the results.
At some point, these discussions always get to some question like "what is 'knowing'?"
Would a person who only ever read articles and looked at pictures of giraffes have a better understanding of them than Dall-e does? At some level, probably, in that every person will have a similar lived experience of _being_ an animal, a mammal, etc. that Dall-e will never share. Is having a lesser understanding sufficient to declare it has no real understanding?
I am not really sure that this is a basis to say human intelligence is fundamentally different. If you ask a human to draw a picture based on nonsensical input, they probably wont. That might not be a difference in process, but in agency.
So is a single-purpose AI equivalent to the entirety of the Human Experience? Of course not. But can it be similar in functionality to a small sliver of it?
More, the author saw mention of it pop up on thier Twitter feed, so wrote an article to cash in on the current "controversy" with no actual research beyond Twitter.
We are going to be seeing this more frequently in cars, and probably hardware overall. Having a single hardware spec that is license-limited to various levels is just cheaper for the manufacturer in how it simplifies their logistics. It also provides additional revenue options from existing customers that might want to "upgrade" at a later date.
On the plus side, if you can figure out how to sideload your car you get a free range extension on your battery.
The individual ops could be faster, so even though there are more of them, the overall speed is quicker. The authors mention using more 16bit ops, so that might be part of the reason?
All YOLOs past v3 will be, to some extent, since the original author wont be releasing anymore versions. This one looks to be in pytorch, so it is also not in darknet. I am not really sure what makes it "YOLO" anymore other than being a single-shot detector, but the author claims they took inspiration from the techniques in the original YOLO papers.
The fact that we can say a person revolutionized the field is specifically because their work, at some point in time, got recognized by a larger audience. The fact that an artist had pushback is also context that is used to determine something is revolutionary; you cannot have subversive art if there is nothing to subvert.
I would actually say that humans have an enormous extra set of data that we, as people, are "trained" on. We walk around in our daily lives, seeing things constantly, and that influences our perception of art. Art is always a product of the broader context it was made in (social, environmental, etc). Something that gets accepted or praised today might very well not have been 200 years ago.
One of the things that is interesting with these new big models is it is dramatically broadening the context in use. The models are learning both the textual representation of a concept, as well as the artistic/visual representation and the relationship between the two domains.
The generation side is actually probably easier than the critic side. I don't know of any network that is capable of determining if something is aesthetically pleasing in an abstract sense, instead they always compare to known examples.