Trying out ChatGPT 4 with DALL-E Image Generation

Alice: «How did he survive?»
[The Red Queen seems genuinely surprised.]
Red Queen: «I can’t say. My termination methods were very methodical.»
[She realizes everyone is staring at her.]
Red Queen (cont’d): «Sorry.»
Resident Evil (early Script)

ChatGPT ( https://chat.openai.com ) offers version 3.5 for free, but if you want to use the more powerful 4 version, you have to pay. You get access to it for $20/month. It comes with DALL-E, so you can generate images.

Trying it out, there are marked improvements. ChatGPT can now access the web and provide answers that include current information (e.g., a list of coming festivals). Nice. No longer are you constrained by old information. However, ChatGPT 4 is also much slower than 3.5, sometimes annoyingly so.

It can also create specialized ChatGPT instances, akin to Gab AI’s characters. You can tell it how it should answer questions. For example, if you want to generate a Devil’s Advocate who looks for weak points in assertions, you can:

My Input: Life is beautiful.
Devil’s Advocate: The statement “Life is beautiful” is certainly uplifting but it’s also broad and subjective, which can open it to various critiques and interpretations:
1. Subjectivity: The idea that life is beautiful depends heavily on individual experiences and perspectives. What is beautiful to one person might not be to another due to differing life circumstances, values, or personal struggles. This subjectivity means the statement can’t universally apply to everyone’s experience.
2. Neglect of Negative Aspects: By focusing on the beauty of life, this statement might overlook or minimize the real and often significant hardships many people face. Life can also be brutal, unfair, and painful. Ignoring these aspects can seem dismissive of the genuine suffering and challenges that are a substantial part of many people’s lives.
3. Philosophical Disputes: From a philosophical standpoint, this assertion can be critiqued by existential or nihilistic perspectives, which might argue that life lacks inherent meaning, beauty, or purpose. Such viewpoints might suggest that calling life “beautiful” imposes a certain interpretation on an otherwise indifferent universe.
4. Dependence on Context: The beauty perceived in life often depends on one’s circumstances at a given time. For someone in a state of suffering, grief, or depression, proclaiming “life is beautiful” may not resonate and could even feel alienating. Thus, the statement might lack inclusivity, failing to acknowledge the breadth of human experiences.
5. Evolution of Perception: Perceptions of life’s beauty can change over time due to personal growth, changes in circumstances, or shifts in societal values. What is considered beautiful or valuable at one point can become less so, or even problematic, as perspectives evolve.

Thus, while “Life is beautiful” serves as a positive affirmation, it’s important to consider its limitations and the broader context in which it is or isn’t applicable.

It can also generate images and … well, it does it somewhat well. It did generate a couple of very good images (each within half a minute or so):

Kudos.

Other images, however, not so … good:

It also allows you to edit the images after generation, so if there an annoying element, you can remove it. (As usual, you have to tell the AI what to replace it with for better results.)

(I did not like the clutter on the ground, so I marked it with the pen and asked ChatGPT to remove it — did it well.)

Speaking of images, you can also upload images and let ChatGPT work with them. For example, it can (mostly) correctly recognize what is on an image:

However, here the massive censorship of ChatGPT becomes obvious. In a lot of cases, the images I requested somehow violated the community guidelines.

For example, asking it to imitate a specific style:

Even asking it to imitate a whole art movement led to problems:

Some error messages seem to relate to technical problems:

And some make no sense at all:

Besides the content limitations, you also have a question and an image generation limit:

If you hit the limit, you can continue to ask questions to ChatGPT 3.5 but you cannot generate images. Strangely enough, you do not see how many questions or images you have left. If you use, e.g., Microsoft Bing and DALL-E, you see a coin-icon attached to each image with the number of images left. Even worse, it starts to generate and then tells you that you have reached the limit. Very annoying.

So, overall, is it worth it?

It’s interesting to play around with it. The answers are better, but slower. The image generation does produce some nice images, although I like the style of Gab AI (https://gab.ai) much more. Even better, Gab AI is much less censored than Chat GPT. And for 20 bucks at Gab AI, you get 50 images per 24 hours, which suits me more. BTW, if you want to try out DALL-E, you can do so with a Microsoft Account. Just use the chat (or switch to the Designer Mode: https://www.bing.com/chat ). Microsoft seems also less censorious than ChatGPT.

And yeah, overall, for images and for interaction with Characters, I like Gab AI much better. ChatGPT’s community standards are just too constraining and often make no sense at all.