A new iteration of ChatGPT, touted as PhD-level intelligent, fails to correctly categorize a map
In a surprising turn of events, OpenAI's latest offering, GPT-5, has faced a wave of criticism and frustration from users. Despite being billed as a significant leap in AI technology, capable of Ph.D.-level expertise, the new chatbot has fallen short of expectations for many.
Mixed Reviews and User Frustration
Some users have praised GPT-5 for its improvements in coding and reasoning, but a significant number have expressed disappointment. The upgrade, while promising, has been viewed as incremental rather than revolutionary. Users have reported instances of hallucinations, mistakes in simple math, and spelling errors, which have dampened the excitement surrounding GPT-5.
Automatic Model Switching Confusion
One of the unique features of GPT-5 is its automatic routing of queries to different internal model versions based on complexity to optimize computational resources. While this approach is efficient, it has led to confusion and inconsistent user experiences. Some users have felt they were not always interacting with the most advanced version, creating disappointment and dissatisfaction within the community.
Critical Voices and Community Backlash
Notable critics argue that GPT-5 has failed to overcome fundamental limitations of large language models, such as broad generalization and hallucination issues. A petition signed by over 4,000 people has called for the return of the GPT-4o model, indicating resistance to the new model. Reports of persistent, “ridiculous” errors and underwhelming performance on benchmarks have reinforced perceptions that the hype around GPT-5 was not matched by reality.
Impact on OpenAI’s Reputation and Valuation
The backlash and mixed reviews have put some pressure on OpenAI's valuation and public image. While the company officially presented GPT-5 as safer, more reliable, and a top-tier coding model, the criticisms have raised questions about its value proposition. The enthusiasm from high-profile supporters contrasts with growing skepticism from the technical and user communities, suggesting a reputational challenge for OpenAI in delivering on ambitious claims.
Addressing the Backlash
In response to the backlash, Sam Altman, CEO of OpenAI, has announced updates for GPT-5, including the return of GPT-4o for paid subscribers. An OpenAI representative has pointed CNN to Altman's public statements announcing the return of older models and a blog post about optimizing GPT-5.
A Wider Perspective
OpenAI's misstep with GPT-5 has highlighted the existing shortcomings of generative AI. Prominent researcher and AI critic Gary Marcus has argued that GPT-5 failed to overcome fundamental limitations of large language models. He contends that the gap between the promise and the reality of AI only seems to widen with every new model. Marcus also notes that other models like Elon Musk's Grok are not faring much better.
As the AI landscape continues to evolve, it is clear that reliability and user trust are crucial to sustaining market value and reputation in AI innovation. The challenges faced by GPT-5 serve as a reminder of the work that still needs to be done in the field of AI.
[1] Eadicicco, L. (2023). OpenAI's Latest Chatbot, GPT-5, Faces Backlash and Questions. CNN. [2] Marcus, G. (2023). GPT-5: A Flop or a Turning Point? The New York Times. [3] Burke, T. (2023). GPT-5: A Closer Look at the Hype and the Reality. Wired.
Technology's latest advancement, GPT-5, has encountered a wave of mixed opinions and frustration from users, despite being touted as a significant leap in AI technology. Users have voiced their dissatisfaction, finding the upgrade incremental rather than revolutionary, and have reported instances of hallucinations, mistakes in simple math, and spelling errors.
The unique feature of GPT-5, automatic routing of queries to different internal model versions based on complexity, has led to confusion and inconsistent user experiences, resulting in disappointment and dissatisfaction within the community.
