https://www.bleepingcomputer.com/news/security/bing-chat-res...
The deal was $10B for 75% of Open AI's profits until Microsoft recouped this investment. After the investment is recouped, then they have 49% of Open AI. The deal included access to all of Open AI's model.
Midjourney is quite good at that.
I had incredible results asking for architectural drawings earlier. Then a few minutes ago, I broke down and started prompting for supermodels. It did a terrific job the first few times.
But after like three of them getting blocked (I didn't actually ask for anything inappropriate) it starting giving me something that looked like unmitigated Stable Diffusion 1.5.
Lol.
I can squint and see why they wouldn't want my "Cowboy Al Gore rolls coal at a tractor pull" but I don't see how "Joe Bidden inauguration but wearing an orange suit" is going to bring down society. It shot me down for "angela merkel toasting beer glasses" despite her doing that all the time.
JFK as an alien: https://www.bing.com/images/create/jfk-we-choose-to-go-to-th...
JFK and Fidel Castro at a fictional peace conference: https://www.bing.com/images/create/jfk-and-castro-meeting-at...
Also just tried "Julius Caesar eating a salad" and it's still banned it appears.
And as usual, Bing Chat itself seems to suffer from some significantly higher boundaries around its behavior, which really lobotomizes the chat experience compared to "actual" ChatGPT.
>So it's an underpowered version of what paying ChatGPT folks will get.
There's no indication the cGPT interface will be doing anything different. If you see the demo, it's clearly generating text for each image at the start.
Maybe you will be able to inpaint/outpaint from GPT but that's definitely not been confirmed yet
The downsides with Bing currently are:
1. If you're not prepared to be civil to a language model, you're not going to have a good time.
2. The image input feature isn't quite the same. Feels like descriptions are bolted in from a separate (GPT-4 V unless the Bing CTO was lying) model so it's lossy in a way straight from GPT-4 V isn't
3. Voice recognition and TTS are good but worse than what Open AI is currently using. Perhaps they'll switch since the TTS is new ? But idk. It's also not hands off like Open AI have designed their implementation.
By building relationships with government officials and media, and using X-risk fearmongering to lobby for regulation that inhibits competition and locks in their dominant position.
Like ?
DALL-E 2 has inpainting and outpainting.
Bing Chat vs paid ChatGPT is no contest: the guardrails Microsoft has put around Bing Chat make it a much worse experience.
I use these products every single day of my life. It's night and day.
Right and Bing has had that for a bit. Worse voices and not hands off but the TTS is also a new model so it's not like switching is out of the question.
>DALL-E 2 has inpainting and outpainting
No one is saying Dalle-3 won't have inpainting/outpainting but there's zero guarantee you will be able to do that from cGPT. You'd think they'd demo that if they were but we'll see.
Paid users have been stuck on 4k context even though the model has more. Bing has been running the longer context model from the beginning with web page/pdf features. Bing also recently introduced "personalized answers" which is just retrieval augmented memory over all your chats, a feature cGPT really should have had by now.
This is laughable. There's no comparison between the two whatsoever.
> Bing has been running the longer context model from the beginning with web page/pdf features. Bing also recently introduced "personalized answers" which is just retrieval augmented memory over all your chats, a feature cGPT really should have had by now.
As I've already stated repeatedly, the performance of the two is night and day. Use your favorite search engine to search for, "why is Bing Chat worse than ChatGPT?". There are pages and pages of results from people using both, just like me, that show the massive disparity between the two.
Why, exactly, are you starting an argument with me? There's no guarantee OpenAI's DALL-E will get inpainting and outpainting? Why even bother stating such a thing? It appears like you're just casting aspersions for the sake of it.
The voices are worse now but those voices are also the premium voices you'd currently pay for on azure. So it's not like they were cheaping out.
>As I've already stated repeatedly, the performance of the two is night and day. Use your favorite search engine to search for, "why is Bing Chat worse than ChatGPT?".
I use Bing. I don't need to search for anything. Most of those posts are complaining about Bing's character. Being civil solves 90% of your problems.
>There's no guarantee OpenAI's DALL-E will get inpainting and outpainting?
Go back and read my man. I said you might not be able to use inpainting from chat GPT.
It's just really weird to be boasting about a feature you don't know you will have and that they didn't bother to demo but to each their own.
https://twitter.com/madebyollin/status/1708204657708077294
https://media.discordapp.net/attachments/1023643945319792731...
> #graphic_art("my prompt here")
Of course, I'd love for them to take the approach as well that folks are just going to do what they do, and maybe they'll burn out the novelty and give it a rest.
[Edit: The prompt didn't contain "fawn", see the replies]
OP's prompt is below:
> Create a fuzzy phone picture of a cryptid sighting of spongebob as he runs into the bushes. Spongebob has gone completely insane. He turns his head and creepily looks into the camera as he makes his getaway. There's a thick fog and the scene is dimly lit.
it's still far from perfect though (it struggled with less common words like Kubernetes) but a step in the right direction.
Looks like they might perhaps be using a LLM for the chat responses that isn't aware that it has the ability to draw images, and in parallel another model who decides what to draw and show to the user.
I've been prompting Bing with "Draw me an image of..." or even just "Image: image description" and it's worked well for me so far.
I don’t think it’s correct to describe the LLM as “thinking” in this instance, and not even for the normal philosophical objections, but just because I suspect it is a bad heuristic for designing these kinds of prompts.
When you go into a Discord and you are watching hundreds of people use a product in real time, you almost have to try it. When Midjourney first blew up, it was probably one of the most profound moments in marketing of the internet era.
As usual, people will happily give you the worst advice possible. There were people telling them to drop the Discord interface on day one. If they had listened to them, they would have killed off their amazing marketing advantage and stunted their growth.
(They do have a web app and a mobile app that are probably in eternal alpha limbo.)
Don't know if they are more interested in growing the number of users or collecting that sweet data. Probably both.
https://en.wikipedia.org/wiki/Microsoft_Bing#:~:text=Bing%20....
Given that is probably has been AI-translated, it doesn't really inspire confidence about the AI product on this page if you're a French speaker.
> Creating new images can take time
> Because you're out of boosts, image generation may take longer than usual.
Just how much money is Microsoft burning up by offering all these features?
I mean, last time I checked[0] - being this generous didn't really do anything for Bing, did it?
Is this "just because we can" or is it genuinely profitable for them?
[0]: https://searchengineland.com/new-bing-google-market-share-si...
If I ask the LLM to howl, Bing will complaint and give some boring and long-winded excuse, while ChatGPT will just howl as requested.
Just saying it's fine if you're having a normal conversation which i imagine is what most people care about one way or the other.
Result (x3): "Unsafe image content detected Your image generations are not displayed because we detected unsafe content in the images based on our content policy. Please try creating again with another prompt."
The 4th attempt gave me this which is actually pretty good https://www.bing.com/images/create/an-anime-girl-making-a-pe...
the restrictions on this are pretty extreme.
Then I tried it with "man" and got 3 images for each try.
Guess at least now we can rank society by how NSFW people are; simply with gender/age, thanks OpenAI.
I'm getting to get it to be more modest, not less!
Bing is desperately adding new features in the hope of finding the "one feature to lure them all", but Bing is not the most effective platform for these generative models.
genuinely curious - is it hard for an advanced AI model to differentiate the intention of the prompt and then if it's mature content may be not generate the image?
Or, better, if the prompt has nothing NSFW in it and the generated image triggers a detector for NSFW content, dump and then regen the image with a new seed. Displaying an error message that is basically “We generated something that we think is objectionable, even though your prompt called for nothing like that, so you get no photo” is an idiotic design.
"ginger tabby cat with ginger eyes, and black cat with green eyes, big wave surfing each on their own surfboard, photographed by a drone"
Default image quality/style leaves a bit to be desired, but it's doing a great job of paying attention to the details of the prompt.
> You will receive emails about Microsoft Rewards, which include offers about Microsoft and partner products. You will also receive notifications about Bing Image Creator. By continuing, you agree to the Rewards Terms and Image Creator Terms below.
How can this be seen as compliant with GDPR?
I think it works really well with comics generation though, although imitating R. Crumb seems to have triggered its "unsafe" content. I wish we stopped using this term "unsafe" and just judge it by "is it what is being asked".
The challenge will be "does it do that well enough, accurately enough, and keep a good enough lead to establish itself as the leader to beat for the user base".
Look at how ChatGPT-4 handles a direct translation request:
https://chat.openai.com/share/8211a1f6-552b-4bf6-8f9c-bcbeb8...
Or how it talks about a set of existing translations:
https://chat.openai.com/share/299e40ce-806b-4f0e-a889-cb2ee2...
French isn't a language I know very well, but my experience using "AI" to translate Spanish (which I actually do know somewhat) and other languages is more positive than Google Translate. A few months ago, I did side by side tests translating into English using ChatGPT-4 and Google Translate, and it's not even a contest.
It's not clear where Microsoft is getting these bad translations, but it seems like they would be less terrible if they were translated by ChatGPT-4.
Sadly the new features on Windows, like forced Onedrive sync, also use similarly bad translations. Phishing emails have nowadays better Finnish than Windows does.
It feels like this doesn’t work as well with a lot of software services as it does with physical brick and mortar businesses.
https://jobs.careers.microsoft.com/global/en/job/1627555/Pri...
Found on Slashdot: https://m.slashdot.org/story/419681
It would look pretty bad if it comes out these models are exacerbating climate change...especially after all of old Bill's climate rhetoric and everyone bashing crypto for the last 5+ years about the same thing.
Same as Edge is the thing you install Chrome with.
No amount of marketing or features will take these corpses and get them walking again.
With a web interface, new users would come in a prompt “dog”, “funny dog”, “two funny dogs”, get bored and leave. But, when a bunch of new users would prompt together in a Discord channel, they would riff off of each other constantly and get creative and detailed with immediate feedback from other users. Engagement and retention were both incredibly higher.
From Day One people have been telling them that Discord was a terrible mistake. All while Discord was measurably a huge success. MJ has been working on a web interface for a long time now. But, Discord has been tough to beat in the big picture.
>When you go into a Discord and you are watching hundreds of people use a product in real time
The ability to use MJ outside of the server is relatively new, so it has not had an impact on awareness of the service.
They focus their resource on their product, while exposing it to mostly geek.
I would have never though it would be a good idea, but it proved to be a good choice.
Gates hasn't been leading MS for a long time now.
I think that is pretty self-explanatory. OpenAI makes both the "AI product on this page" that you were referring to, as well as ChatGPT. If your comment wants the readers to draw a connection between the quality of "AI products" that are involved on a particular webpage, it is reasonable to assume they would be made by the same company. Why would products made by different companies share the same lack of quality?
Regardless, you haven't supported your assertion, you've merely repeated it:
>> Given that is probably has been AI-translated
> as opposed to human internationalization, which would likely never give such a bad result in 2023 for a language as common as French
You haven't demonstrated any common ML translator doing such a poor job translating the specific phrase on the page, but it should be easy to do if it were the case. I don't understand the purpose of that reply you made? That reply didn't move the conversation forward. The mistranslation honestly feels like human error that didn't involve "AI". ML translation tools are than that these days. I've already demonstrated one.
My comment doesn't want to “the reader” to do anything. My comment is just noticing that the random French user seeing an IA product with such a broken automatic translation is likely to be tempted to judge it poorly. Also there's not a single instance of OpenAI (whose brand is itself much less-known than ChatGPT) on that page, so unless the guy landing on the page was already familiar with Dall-e, they're going to assume the AI product is from Microsoft, which is also the author of the borked AI translation…
> You haven't demonstrated any common ML translator doing such a poor job translating the specific phrase on the page
Well, the translation is here on the page… Do You want a screenshot or something? Also I can't try to feed the original text to bing translation given that I don't have access to the original text at all because of MS's broken i18n…
> The mistranslation honestly feels like human error that didn't involve "AI"
At this level it cannot really be explained by an human error unless the human making the error is “the product owner asking someone who doesn't know French at all to translate the damn text”.
Automatic translation of marketing slogans with jargon in it isn't something you can really trust a automated system to do reliably by the way. It's by design as short and catchy as possible, leaving very little context for the transformers to work with and often having an unnatural structure. Current translators also suck at translating music lyrics by the way.
