Live data from Hacker News

DALL·E 2 prompt book [pdf]

dallery.gallery

141–150 of 157 posts

Re: DALL·E 2 prompt book [pdf]

#142

OpenAI has pretty much been ruined for me after they sold their souls to Microsoft, stopped releasing all their source code, and then dishonestly refer to their sad practice of censoring the training data as "AI safety/alignment" when in fact it will never be a reasonable AI safety technique in the long run, and is only done to avoid bad PR. Clearly OpenAI is no longer a company worthy of its founding principles of o…

I feel like they have no choice but to do some heavy-handed censorship strategy. People who have zero understanding of technology but an infinite supply of outrage will get the whole concept legislated out of existence on “think of the children” or “this computer program is _____ist” grounds if they allow anything even mildly disturbing.

Re: DALL·E 2 prompt book [pdf]

#143

I just got access to the DALL-E 2 beta, and it's a ton of fun to make pictures out of everyday occurrences as prompts. Someone else here on HN observed that everyday people don't "get" how huge this all is. I experimented with asking random acquaintances at a local cafe for prompts and showed them the generated pictures. All but one person was totally unimpressed. If everything feels like magic, then what's one more…

The thing is, regular people have been under the impression that computers are capable of generating complex imagery by themselves for quite some time already. About 10 years ago I was visiting family, and my uncle -- an older guy who enjoys tinkering with technology -- asked me to draw a picture on his new tablet. Now, I've spent a fair chunk of my life drawing, so I was able to sketch a pretty decent face quite quickly with just the basic pencil tool. When I handed it back, he was so amazed that he grabbed my grandmother as she was walking past. "Look at this! Look at what your grandson drew!" She glanced over, completely unimpressed, and said "yeah, but that's on the computer".

Re: DALL·E 2 prompt book [pdf]

#144

Earlier quoted context omitted.

After sharing and enjoying people share their prompts in various communities for 2-3 months, I hate this business model.

Why? You can still do what you want to do - this sort of thing just helps people who perhaps don't have the ability or inclination to capably form their own concepts. It's much like hiring a painter, a writer or a web designer... . At least, it's interesting for me to see what may be the beginnings of a new job class. I've been wondering where all this ML/AI business might take us.

Because it provides an incentive to not share knowledge, but instead hide it behind micro transactions.

Re: DALL·E 2 prompt book [pdf]

#145
post #142

OpenAI has pretty much been ruined for me after they sold their souls to Microsoft, stopped releasing all their source code, and then dishonestly refer to their sad practice of censoring the training data as "AI safety/alignment" when in fact it will never be a reasonable AI safety technique in the long run, and is only done to avoid bad PR. Clearly OpenAI is no longer a company worthy of its founding principles of o…

I feel like they have no choice but to do some heavy-handed censorship strategy. People who have zero understanding of technology but an infinite supply of outrage will get the whole concept legislated out of existence on “think of the children” or “this computer program is _____ist” grounds if they allow anything even mildly disturbing.

The thing is, they release just enough information for people to eventually remake their models. Someone with no scruples will make a uncensored version eventually and then their work will have had the effect they claim to want to avoid anyway, but at least it won't reflect back on them.

That's why it's really nothing but a PR exercise. I honestly don't think they care much past that.

Re: DALL·E 2 prompt book [pdf]

#146

I just got access to the DALL-E 2 beta, and it's a ton of fun to make pictures out of everyday occurrences as prompts. Someone else here on HN observed that everyday people don't "get" how huge this all is. I experimented with asking random acquaintances at a local cafe for prompts and showed them the generated pictures. All but one person was totally unimpressed. If everything feels like magic, then what's one more…

>asking random acquaintances at a local cafe for prompts and showed them the generated pictures. All but one person was totally unimpressed.

I think this is because people reason this way: the computer has many pictures in its archive, saved as "dog.jpg" and "hat.jpg". If your prompt is "dog with hat", the computer just combines them. I think this is what people think is happening, so they are less impressed.

Re: DALL·E 2 prompt book [pdf]

#147
I don't know whether I'm happy to see that living creatures are still mostly nightmare fuel. In these images DALLE seems to only really get faces right. Hands, horns, etc. are either contorted, blobs, or have the wrong number of appendages.

The images give a good first impression. Which is... impressive in itself. But they won't fool anyone who's studying them for even a few seconds.

Re: DALL·E 2 prompt book [pdf]

#148
post #135

Earlier quoted context omitted.

If we assume intelligence is a normal distribution, then we can say half of the population is below average. I think it's fair to categorize half as most.

“Half = most” is… innovative. I like to think of it as “half = so many that you’ll run into a lot of them.”

By "most" definitions, "most" means more than half. So we're literally arguing over a single person.
Post reply on HN