Understanding the GPT-5 Debacle: What Went Wrong
The highly anticipated launch of GPT-5 has left many users feeling underwhelmed. Unlike the revolutionary changes that accompanied the release of its predecessor, GPT-4, the new model has not elicited the expected “wow” factor. CEO of OpenAI, Sam Altman, expressed excitement about this release, likening it to a technological marvel akin to a “death star.” However, the reality is that many users are longing for the familiar warmth of GPT-4.
The Introduction of the Router System
So, what led to this disappointment? One major factor is the introduction of the router system in GPT-5. OpenAI has adopted a unified approach by launching all sub-models of GPT-5 simultaneously. This system is designed to choose the most appropriate model for each query based on the conversation’s complexity or the tools required. However, this automated decision-making process has frequently opted for the basic GPT-5 model rather than more advanced reasoning models. As a result, users have reported that GPT-5 seems far more simplistic than expected.
Efficiency or Cost-Cutting?
The underlying problem appears to be related to efficiency and perhaps even cost-cutting. OpenAI’s routing mechanism aims to determine which model is not only efficient for the user but also for their server infrastructure. At launch, OpenAI was not transparent about the criteria used for selecting a model, leading to suspicions that the company was prioritizing lower operational costs over delivering the most sophisticated responses.
Aidan McLaughlin, a member of the OpenAI team, defended this strategy, stating that the system prefers the basic model based on user interaction patterns. According to McLaughlin, they do not employ the router solely for cost management but to improve overall efficiency. Despite the rationale, users have expressed dissatisfaction, feeling that their expectations have not been met.
Altman’s Acknowledgment of Issues
During a recent Ask Me Anything (AMA) session on Reddit, Sam Altman acknowledged the backlash from users who felt that GPT-5’s performance did not measure up to GPT-4. He remarked that many users found the new model to be “more silly,” attributing this to the router’s premature deployment. Altman’s admission highlights that even the company behind GPT-5 is aware of its shortcomings.
Managing Expectations
In an effort to temper expectations prior to launch, Altman had previously warned that GPT-5 would incorporate experimental features that might not deliver the gold-standard performance users hoped for. His warnings included statements like, “This is an experimental model that incorporates new research techniques,” indicating that significant advancements would take time to materialize. Unfortunately, these cautious words did little to quell users’ enthusiasm.
OpenAI’s Response to User Feedback
In light of the widespread frustration, OpenAI decided to allow subscribers of their Plus plan—priced at approximately €20 per month—to continue using GPT-4, giving them a familiar option amid the chaos. Furthermore, the company took swift action to adjust the routing algorithm, aiming to improve model selection without clarifying what prompts the decisions.
Finding Workarounds
In a bid to enhance their experience with GPT-5, users discovered methods to push the router towards the more advanced reasoning models. For Plus subscribers, employing phrases like “think through your response in-depth” in their requests proved effective in activating the Thinking Model. This trick allows for better performance without exhausting the weekly limit of 200 messages.
Integrating Reasoning Options
In a further attempt to rectify the situation, Altman confirmed the reintroduction of a hidden reasoning button for free users, allowing access to the GPT-5 Thinking Mini model. This update aims to enhance engagement with reasoning capabilities, which many felt were lacking with the basic version. This new feature enables users to generate more complex responses while maintaining limited message access.
Quantifying User Engagement with Reasoning
To provide insight into the effectiveness of these changes, Altman shared statistics indicating a marked increase in reasoning usage among both free and Plus subscribers. Free users jumped from less than 1% to 7%, while the use of reasoning among Plus subscribers soared from a mere 7% to an impressive 24%. This substantial improvement suggests that the reasoning function is now seen as a vital component of GPT-5’s appeal.
The fallout from the GPT-5 launch has prompted OpenAI to make significant adjustments, showing their responsiveness to user feedback. As users navigate this evolving landscape, they remain hopeful for a future where AI can meet their high expectations.