OpenAI’s highly anticipated GPT-5 launch has backfired spectacularly. Days after unveiling its flagship AI model during a glitzy livestream, the company faces a user revolt over what subscribers call a “personality-deficient,” overly restricted upgrade. The backlash forced CEO Sam Altman to publicly walk back key limitations and restore access to the older GPT-4o model, a rare retreat for the AI giant.
User Fury Erupts
ChatGPT Plus subscribers, paying $20 monthly, slammed GPT-5 as “colder,” “mechanical,” and “an overworked secretary” compared to GPT-4’s conversational warmth. Complaints flooded Reddit and X, with users threatening mass cancellations. Many lamented shorter replies, abrupt tone shifts, and a perceived drop in contextual accuracy, especially jarring after Altman’s Star Wars-themed teaser hinted at revolutionary gains.
Compounding frustrations, OpenAI initially imposed a strict 200-message daily cap on GPT-5 interactions—lower than prior models while removing GPT-4o access entirely. “*We for sure underestimated how much some things people like in GPT-4o matter to them*,” Altman admitted.
Technical Troubles Amplify Distrust
Behind the scenes, a critical technical failure worsened perceptions. An “autoswitcher” system meant to route queries between GPT-5’s reasoning tiers broke during rollout, making the model appear “way dumber,” Altman confirmed. Simple tasks like counting letters in “blueberry” failed until users demanded it, “think harder,” triggering a shift to advanced reasoning modes.
The fumble fueled existing skepticism. AI researcher Gary Marcus dubbed GPT-5 a “major letdown,” arguing its incremental gains damaged OpenAI’s brand. Meanwhile, benchmarks showed mixed results: GPT-5 topped some leaderboards but trailed xAI’s Grok in AGI-oriented testing.
OpenAI’s Damage Control
Facing defection threats, Altman announced three fixes within 48 hours:
-
Restoring GPT-4o: Now available for Plus users “while monitoring usage” 26.
-
Doubling GPT-5 rate limits: Addressing “compute strain” from surging demand.
-
UI transparency upgrades: Indicating which model answers the query.
“We are significantly increasing rate limits for reasoning,” Altman tweeted, adding legacy model limits would soon exceed pre-GPT-5 levels.
The Personality Paradox
The revolt highlights a growing tension in AI development: raw capability versus emotional resonance. Users didn’t just critique GPT-5’s accuracy; they mourned a lost “friendship” with GPT-4o’s relatable responses. As Wharton professor Ethan Mollick observed, next-gen AI must balance “extraordinary” skills with human-like warmth.
Altman acknowledged unease about users forming emotional dependencies, noting OpenAI is tracking “subtle cases where AI could reinforce delusion”. Yet restoring GPT-4o concedes that user loyalty hinges on more than benchmarks.
Broader Implications
OpenAI now faces tough “capacity tradeoffs” as demand strains infrastructure. Rivals like DeepSeek, whose app saw 75M downloads in 2025, are gaining ground with transparent, open-weight models. While GPT-5’s reasoning power may yet shine, its launch underscores a pivotal lesson: for AI, feeling human matters as much as thinking smart.
— Additional reporting from TechCrunch, Bloomberg, and CoinCentral.
Subscribe to my whatsapp channel
Comments are closed.