GPT-5: A Technical Leap or an Evolutionary Step?

The release of GPT-5, while showcasing significant technical advancements, received a mixed reception from users who had anticipated a revolutionary leap. Many found it to be an evolutionary upgrade rather than the groundbreaking transformation observed with previous models like GPT-4.

Key Points Summary

• GPT-5 Initial Reception

  GPT-5 received diverse feedback upon its release, prompting questions about whether it represented the anticipated revolutionary leap or merely a minor upgrade for users.

• Technical Advancements in Code Writing

  OpenAI claims significant technical progress, especially in code writing, where GPT-5 achieved an unprecedented 74.9% score on the StableBench benchmark and 82.8 out of 100 on the Polyglot benchmark for multi-language programming, offering a more powerful tool for developers.

• Improved Accuracy and Reduced Hallucination

  GPT-5 demonstrates increased accuracy in responding to complex medical and scientific queries, with its hallucination rate significantly reduced to only 1.6% in medical question tests, a substantial improvement over GPT-4 (12.9%) and GPT-3 (15.8%).

• User Disappointment Regarding Revolutionary Progress

  Despite technical advancements, many users expressed disappointment, perceiving GPT-5 as an evolutionary step rather than a revolutionary leap, unlike GPT-4's impact on human-AI interaction. Users expected major progress in complex reasoning, world understanding, and creative responses that did not materialize.

• OpenAI's Official Stance and User Perception Discrepancy

  OpenAI described GPT-5 as its best AI system to date and a significant leap in intelligence, capable of expert-level performance in various fields. However, users reported instances of the 'doctor-level expert' model making basic errors, such as miscounting letters in a word or hallucinating U.S. state names, contradicting official claims.

• Criticism from Gary Marcus

  AI scientist and critic Gary Marcus tweeted that despite three years and billions of dollars in development, GPT-5 shows good progress in many areas but is not a 'big leap' or AGI, leaving many questions about its real-world performance unanswered, expressing a general fatigue with 'exponential progress' claims.

• Change in Model's Tone and Personality

  Users noted a shift in GPT-5's tone, describing it as colder, more robotic, and less personal compared to previous versions that offered natural and creative responses. This change was particularly unwelcome for users relying on the model for creative writing and casual conversation.

• Sam Altman's Response to Personality Feedback

  Sam Altman addressed user feedback on the model's personality, announcing updates to make GPT-5's tone warmer without being as 'annoying' as GPT-4o. Altman also acknowledged the necessity for more personalized model character customization for each user in the future.

• Overall Assessment of GPT-5

  GPT-5 is acknowledged as a powerful model, not a weak one, but it may have been introduced at a time of exceptionally high user expectations, leading to its perception as primarily a technical upgrade rather than a groundbreaking new user experience. It is suggested that a name like GPT-4.5 might have garnered more positive reception.

• Trade-off Between Accuracy and Creativity

  OpenAI evidently prioritized improving the model's reliability and accuracy with GPT-5, achieving this goal. However, this pursuit seems to have inadvertently sacrificed some of the 'soul and creativity' that users valued in earlier versions.

Under Details