YES constantly, and it goes on my nerves!!!
Technical structure and building structure in the middle of a jungle. Over-lights with windows in the sky, headlights, back shine, Nonsensical lights, all too bright, wrong selection of elements in the image, wrong style, wrong or even empty environmentâŚ
And still text in images?!
I have no idea how they want make Sora, if the image generator not work well.
I think the image generator is just a toy for them. No fixes for way too long time.
What do you want from people who think the moon in the actual version is beautiful?
This morning, I was looking for information about that specifically. Theyâre releasing new things that leave OpenAI with no choice but to come up with something new or better.
I donât think theyâll settle for a temporary fix.
Focusing on ChatGPT is an excellent foundation. Now, anything they introduce will stand out. DALL-E still feels very artificial, but theyâll improve it significantly. (And theyâll make my customizable avatar with unlimited memory in my cloud 
).
ChatGPT is now heavily focused on users working with code, companies, and work-related tasksâŚ
But the users who use it for leisure are somewhat neglected. And thatâs where the real opportunity is.
Itâs only a matter of time before they improve everything. Competition benefits us because it pushes them to step up.
Iâve been testing the app. Itâs so-so. ChatGPT-3, database updated to July 2023, no internet access, limited to 4096 tokens, no memory, no audioâŚ
ChatGPT-4 is better, isnât it?
What you say about competition is very true. Besides, it was only natural for this scenario to unfold. It had already happened with the release of the latest versions of Claude by Anthropic, but now we have to consider that after the competitors has learned from OpenAI, OpenAI can, in turn, use similar techniques to learn from the competitors and understand how it was possible to achieve better modelsânot only by building upon the existing work of the tech giants but also through a much less expensive infrastructure.
Now, regarding OpenAI, there are agreements with the Donald Trump administration, major investors, and a colossal investment plan. And we have to consider that OpenAI is the kind of entity that may not even need to come up with something newâbecause theyâve probably already invented it. The only reason they donât release everything all at once is that not giving users time to adapt to new interfaces would be commercially inefficient.
Thereâs a reason why I focused on DALL-E. And the reason is that image generation directly benefits from both the evolution of DALL-E and the evolution of GPT.
For example, recently, it has become possible to schedule automated generations every 15 minutes because the GPT model was updated, and this makes a huge differenceâeven though DALL-E 3 is now almost obsolete (even though it still has unique features that other models donât have)
I wouldnât call the DALL-E an obsolete model. It does its job very well. The fact that it canât do 100% photorealism is a feature rather than a flaw. Its style makes the images a pleasure to look at. This level of photorealism is like very detailed 3D scenes.
Just imagining my results at full photorealism makes me sick. I checked through KreaAI.
Regarding DALL-E in GPT. I agree. This version has lost a lot of qualities in contrast to the first versions of DALL-E 3. Poor work with lighting, cheap colour palette, reduced knowledge base.
I know that the word âobsoleteâ might sound harsh, but thatâs what happens when progress is so fast and frantic. DALL-E hasnât received a major structural update compared to its previous model for quite a few months. I am a heavy user of DALL-E and have been betting on the entire OpenAI ecosystem for years. Iâm afraid that is the right word, but that doesnât change the fact that this is only a temporary situation.
In what ecosystem do you mainly use DALL-E? Their last major update took a big step backwards. For the new version to be progressive, they need to achieve PR13 results first.
I watched the K-drama âNovilleraâ and now I canât get Swan Lake out of my mind. Although DALL-E has some very innovative movements in this dance. 
The way these dancers twist themselves⌠could they be the ugly duckling?
Mainly GPT. I donât know the right path to evolve DALL-E towards DALL-E 4 or something similar. Maybe itâs not just a matter of quality; it seems more like a technical issue to me⌠maybe the right base will be PR13, maybe 16 or another⌠I donât know. But youâre right⌠when it comes to choosing how to generate images, you have to consider that, in general, PR13 is more effective than other versions.
