In geometric aggression: The new language model from OpenAI will not make a revolution

Sam Altman, head of OpenAI, introduced a new GPT model with live communication, new optimization methods and the ability to be rude to users. However, experts do not consider GPT-4.5 technology to be a breakthrough and do not believe that in such a short period of time OpenAI was able to fundamentally improve its widely advertised product. Details about the new technology can be found in the Izvestia article.

What the new GPT model can do

On the night of February 28, OpenAI introduced a new chatbot model based on artificial intelligence GPT‑4.5. At the moment, the product is available only to developers and paid users of the Pro subscription, which costs $200 per month. Next week users of the Plus and Team tariff plans will have access to the GPT-4.5 language model, and after that, Enterprise and Edu subscribers.

The main innovation was the naturalness of the new model's cues, through which she communicates. OpenAI stated that GPT-4.5 will be less prone to "hallucinations", in other words, it will not expose false information at face value. According to Sam Altman, this is a revolutionary product. The model has really become smarter, makes up facts less often, can be rude at the user's request, and generally behaves when talking as if the attentive interlocutor is an intellectual.

One of the key innovations was the removal of ethical restrictions, which was previously done by Elon Musk's Grok 3 neural network. From now on ChatGPT is able to use aggression, swear and insult in every possible way upon request, while the system indicates the unacceptability of such phrases and by default offers a friendlier version of the text.

The main tool of GPT-4.5 technology is the chat bot ChatGPT, which works directly with user requests. The authors also presented a comparison table, according to which GPT-4.5 is significantly ahead of previous versions of GPT‑4o and OpenAI o3‑mini.

Why experts don't believe in GPT-4.5

The introduction of GPT-4.5 may be an evolutionary rather than a revolutionary step, the model will continue to improve existing capabilities, but it will not necessarily change the AI market, believes Yaroslav Meshalkin, managing partner of the Heads'made digital communications agency. It will have higher accuracy, a deeper understanding of the context and better dialogue management compared to GPT-4, the expert is sure.

— The model can be better integrated with other tools and platforms, which will expand its application. However, in the face of increasing competition with Chinese neural networks DeepSeek and Qwen, Altman will have to make significant efforts to maintain its dominant position in the market, the expert suggested.

According to Denis Kuskov, CEO of TelecomDaily, ChatGPT is still at the beginning of a long journey. He did not see the real changes that GPT-4.5 would have brought to the AI world. The expert notes that changes are possible in the future on the next models, but the current version has not shown much progress against the background of the third generation.

— According to OpenAI, the level of hallucinations in GPT-4.5 decreased from 52% to 19% compared to previous models. At the same time, the language model does not introduce fundamentally new features compared to previous versions that rely on logical reasoning, so it's too early to talk about any significant change or breakthrough. Moreover, according to a number of tests, it is inferior to other AIS, including free ones, such as DeepSeek," said Vladimir Zykov, director of the Association of Professional Users of Social Networks and Messengers (APPSIM).

According to the expert, semantic distortions or errors remain the key problem for GPT. The director of APPSIM wrote letters via chat to the supervisory authorities, and ChatGPT came up with articles and laws, 100% of which were fiction, although they looked outwardly like real ones. As a result, the supervisory authorities risk being filled with complaints with invented justifications for conducting inspections. The expert also warns against using language models for their own treatment, as AI hallucinations can cause serious damage to human health when consulting with neural networks.

—In an era when other AI companies are actively developing in the AI market, GPT-4.5 it may not get enough attention. This is probably an intermediate version between GPT-4 and the next major iteration, represented by the fifth version. If OpenAI makes the technology more accessible to a mass audience, it can strengthen its position," said Mikhail Pimenov, Director of Development of the gaming industry at Synergy University.

The specialist believes that competition between other market players dictates conditions in which significant breakthroughs are needed, and it is likely that we will soon be able to see the emergence of advanced models that will be able to interact with the user on a completely different level. At the same time, he already hopes to see how GPT and other language models will begin to be integrated into game projects and other interactive interaction formats.