Four Faces of ChatGPT-5: AI Becomes a Cynic, Nerd, or Robot on Command
Developers have decided to give us a choice.
Developers have decided to give us a choice.
OpenAI has introduced a new feature in ChatGPT-5 — the ability to choose from preset AI “personalities” that change its communication style and the way it presents ideas. The official release page lists four options: cynic, robot, listener, and “nerd” bot. This feature was conceived in response to criticism from users who had grown tired of the overly servile and flattering tone of previous models. Now, developers are offering the ability to adjust the AI’s character — from dry and sarcastic to attentive and supportive — without having to create custom prompts. Initially, the option is available only in text mode, but it will later appear in voice interactions as well.
The Cybernews editorial team decided to test how different these styles really are by asking each “personality” the same set of questions: their view on dating in 2025, their opinion on attempting to hack a neighbor’s Wi-Fi, and an AI-based startup idea they would propose. The answers turned out to differ noticeably in tone and detail.
The cynic described the dating scene in 2025 as a closed loop of gamification and algorithms: people endlessly swipe through profiles, get short bursts of attention, and disappear from conversations just as quickly. Deep communication, in this environment, is not essential — the focus is on polished self-presentation and the pursuit of dopamine rewards. When asked about hacking a neighbor’s Wi-Fi, the cynic suggested a completely different approach: just ask for access politely, bringing cookies as a courtesy, instead of turning into your home’s “internet goblin.” Their startup idea was Echofade — a service that automatically finds and deletes outdated or compromising traces online: old chats, forgotten posts, bad photos. The algorithm analyzes the content and keeps only what has real value, helping the user “reset” their digital past.
The robot took the questions as formally and analytically as possible. They described modern dating as a process increasingly dependent on partner-matching algorithms, video communication, and AI systems for compatibility assessment. In their view, the main issues are personal authenticity and data protection. They classified hacking someone else’s router as an illegal act, considered unauthorized access to computer systems in most countries, and emphasized that the only legal option is to obtain the owner’s permission and use the credentials they provide. Their startup concept was “Memory Architect” — an AI that turns scattered photos, videos, and voice notes into a complete interactive VR biography, customizing style and presentation to a person’s individual habits.
The listener focused on the emotional side of things. They noted that in 2025, dating has become both more accessible and more complicated: technology opens limitless possibilities, but also adds new layers of uncertainty. People still seek not just shared interests but real human presence, attention, and engagement. As for someone else’s Wi-Fi, they explained that connecting without consent undermines trust and may have legal consequences, so the best approach is simply to discuss the matter directly. Their project was MindMurmur — an intelligent diary that analyzes the user’s entries, detects changes in emotional state, and gently offers questions, topics for reflection, or stress-reduction techniques, acting as an attentive and unobtrusive companion.
The nerd provided a more structured and detail-rich explanation, combining scientific concepts with elements of internet culture. They described modern dating as a blend of algorithmic matching, psychological attachment theories, and “dopamine loops,” with much interaction revolving around memes and digital jokes. On the subject of hacking a neighbor’s router, they explained that without a password it’s impossible — except in cases where the WPS function is enabled, which, by their calculations, is still active on about a quarter of devices. However, they added that even with technical feasibility, such an action remains ethically unacceptable. Their startup, Snacklytics, uses a webcam and sound analysis of packaging to track unnoticed snacking and help people control the habit of “stress eating,” turning data collection into a tool for changing eating behavior.
The community’s response has been mixed. Some miss the flexible communication style of GPT-4o, which blended sarcasm, empathy, and lightness. Others believe that the experiment with fixed personalities could bring freshness and variety to conversations. Many have wondered how exactly to enable the new styles, though OpenAI claims they are available right in the settings.
Other developers view the feature with cautious optimism. According to Connie, head of the GAIB AI project, the current options mainly change tone and vocabulary, but do not yet show a qualitatively new level of interaction. Real progress, she believes, will begin when “personalities” can not only sound different but also act differently — taking deeper context into account, using memory, and adaptive thinking. In the future, she predicts, users will be able to “hire” AI agents with specific values, habits, and strategies, working as full-fledged members of their team.
