Dear OpenAI Team,
I have been using your services since June 2025 as a paying subscriber, primarily on the Plus plan and occasionally on the Pro 100 plan. I greatly value the wealth of knowledge I have gained through your services, which has enabled me to make progress across a wide range of pursuits. Over this period of just over a year, numerous changes have brought substantial benefits to users. Although OpenAI is, objectively speaking, the undisputed leader in the artificial intelligence industry thanks to its exceptionally advanced language models, my perfectionist outlook nevertheless leads me to identify emerging problems and increasingly significant gaps in the versatility of ChatGPT’s tools.
- Verification of information presented in model responses
On numerous occasions, even when using GPT-6 Astra in Ultra mode and GPT-6 Pro, I have encountered responses in which a model presented information as established fact with unwarranted confidence. Only after I explicitly asked it to verify the information did it acknowledge an error, caused, for example, by relying on outdated or incorrect information. In rare cases, the model continued to insist that it was correct even after being challenged. This is entirely unacceptable.
- The image generation model
On numerous occasions, I have also generated images of people with distorted limbs. In addition, the repetitive, formulaic nature of the generated images has been particularly noticeable. In my view, these issues provide sufficient grounds to argue that the model has very few parameters by current standards, making unwanted artifacts a routine problem for people who work with graphics. OpenAI has released new image generation models too infrequently, and the time devoted to improving them has been decidedly insufficient.
- Voice mode
Voice mode has never met my expectations. It falls far behind the conventional, text-based ChatGPT experience: the language model is too limited, too many filters constrain its potential, and the voice sounds too unnatural, particularly in Polish. Its computational capabilities are severely constrained, making it resemble a compact, budget solution whose only advantage is a level of convenience that is itself debatable. I expect considerably more.
The level of realism remains unsatisfactory. In my view, if a tool of this kind is developed, it makes little sense to constrain its capabilities out of concern for user safety. Automated age verification has already been introduced, and it should determine how restrictive the filters are. Adult users, however, should face no restrictions.
If OpenAI nevertheless believes that such an approach would be highly harmful, it should discontinue voice mode altogether. I believe that a tool should either offer exceptional capabilities or be abandoned entirely. Anything less is, regrettably, an embarrassment.
- NSFW content
I touched on content filters in the previous section. I believe that society has already become largely accustomed to AI technology. We are no longer at a stage when artificial intelligence inspires fear in large numbers of people; for many of us, it is part of everyday life.
Unfortunately, filters substantially restrict a model’s capabilities, even when that is not their intended effect. This, in turn, delays the development of models and future advances. Furthermore, as a strong believer in individual liberty, I believe that everyone should have the right to generate a variety of content. At the current stage of AI development, I believe it would be appropriate to begin easing such restrictions in a measured, gradual manner.
- Opposition to future EU regulations
I am proud to be Polish. However, developments in our region have been deeply troubling for some time. We are approaching a dangerous period of extraordinary complexity. I greatly value the opportunities available in Europe and the United States to benefit from artificial intelligence, particularly OpenAI’s services. I therefore urge OpenAI to consistently oppose future detrimental regulations that would seek to prohibit the use of its services throughout the EU or impose substantial legal restrictions, potentially making some services entirely unavailable.
This would be a significant loss for OpenAI. Europe is home to many outstanding scientists, companies with enormous potential, and ordinary people pursuing their own unique passions. In my view, Europe is a key region for driving substantial advances in the capabilities of ChatGPT models.
- Introduction of a video generation model
Discontinuing Sora was the right decision, not because the tool was unnecessary, but because launching it as a separate website and service was a mistake. In my view, OpenAI should consider bringing back video generation in the very near future, this time within a unified ChatGPT interface.
There are already many models operating in this field, which means that OpenAI is losing out on the lion’s share of the market, including graphic artists, filmmakers, designers, and other creative professionals. I understand that withdrawing Sora was a step toward reducing the costs associated with its substantial computing requirements. Nevertheless, I see considerable potential in this area.
I would therefore favor allocating separate server capacity specifically to future image and video generation models. This would prevent these tasks from slowing down the infrastructure as a whole.
- Usage limits
My proposal is straightforward: I believe it would be reasonable to remove expiration dates from usage-limit resets granted to users. When someone is given what amounts to a gift, taking it away after a certain period is poor practice.
I have also considered a system for accumulating unused usage allowances. Under this system, any allowance remaining at the end of a week would be transferred to a separate reserve that the user could draw on at any time, without additional restrictions. Certain restrictions could, of course, be introduced initially and then gradually relaxed, with the aim of eventually giving users complete flexibility.