SIAgentNews.com
← All news

OpenAI's GPT-6.2 faces early backlash for poor usability

Early reviews of OpenAI's latest model release suggest it struggles with autonomy and user experience compared to competitors like Anthropic.

· First reported by gizmodo.com

14.5-minutelength of Matt Shumer's YouTube review

The short version

  • GPT-6.2 Astra is marketed as a frontier model for complex projects like finance and law.
  • GPT-6.2 Sol is designed as a faster, more efficient model for everyday technical and non-technical tasks.
  • Early reviewers reported that the model frequently asks for permission on simple steps like downloading files.
  • Critics have suggested the release may give Anthropic a competitive advantage ahead of its IPO.
  • OpenAI's product lead defended the release, claiming the model underwent significant changes just before launch.

Why it matters

For users looking for reliable AI assistants, these reviews suggest that OpenAI's newest tools may require more manual 'hand-holding' than previous versions. It also highlights a potential shift in the market leadership between OpenAI and Anthropic.

A high-stakes model release

OpenAI recently launched its latest model family, GPT-6.2, at a time when the company faced significant pressure to impress the public. The release arrived just as Anthropic, a major competitor, prepared for its expected IPO tomorrow. OpenAI aimed to position these new models as expert digital assistants capable of handling professional projects in fields like accounting, law, and finance.

The new lineup includes two distinct versions tailored for different user needs. GPT-6.2 Astra is marketed as a frontier model designed for the most difficult projects, such as drafting documents and slideshows that must adhere to specific templates. It is intended to produce high-quality output that does not appear to be generated by an artificial intelligence.

In contrast, GPT-6.2 Sol is built for speed and efficiency. It is designed to handle everyday tasks while using fewer tokens than its predecessor, GPT-6.1 Sol. OpenAI stated that many of its own employees currently use the Sol version as their primary model for both technical and non-technical work.

Claims of self-improvement

A major selling point for GPT-6.2 Astra is OpenAI's claim that it has crossed a high threshold for AI self-improvement. This refers to recursive self-improvement, a process where the AI can improve its own capabilities. OpenAI suggested that this milestone is roughly equivalent to providing human AI researchers with a highly capable assistant.

Despite these ambitious technical claims, the initial reception from the tech community has been largely negative. Several well-known tech commentators were given early access to the model and published their findings shortly after the release. Many of these reviewers expressed disappointment with the actual usability of the new tools compared to the marketing hype.

The feedback suggests a gap between the model's theoretical capabilities and its practical application. While the model may be able to perform complex tasks, users found the experience of getting it to do so to be difficult. This discrepancy has led some observers to question the readiness of the model for public use.

Frustrating user experience

One of the primary complaints from early testers involves a lack of autonomy in the model's behavior. Investor Matt Shumer reported that the model would constantly ask for permission on minor, innocuous steps. For example, it would ask the user if it should download a file, a task that most current market models perform automatically without intervention.

Shumer described the experience as discouraging, noting that the model felt like a step backward in terms of user experience. He stated that the model has no concept of what autonomy means and compared it unfavorably to current Anthropic models. His review suggested that while the model might have potential with more training, it currently feels like a gift to its competitors.

Other reviewers echoed these sentiments regarding the amount of guidance required to achieve results. Ethan Mollick, an associate professor at the Wharton School, noted that while the model could perform ambitious tasks like simulating a space elevator, it required significant pushing. He described the process as often frustrating and noted that it was harder to steer than other models currently available.

Internal changes and defense

In response to the wave of negative reviews, OpenAI's product lead, Thibault Sottiaux, defended the release on social media. He suggested that the early reviews might not reflect the current state of the model because his team made a number of changes in the days leading up to the launch. He noted that some influencers were given access a while ago and may not have tested the most recent updates.

This defense has faced some pushback from the community. Dan Shipper, an AI commentator who also published a negative review, claimed that his company continued testing during the timeframe Sottiaux mentioned. He stated that he did not receive any communication from OpenAI regarding new changes, which led him to believe the testing was consistent.

The situation highlights a potential communication breakdown between OpenAI and its early testing partners. If the model was indeed updated significantly without notifying the primary reviewers, it could explain the discrepancy in performance. However, the initial public reception remains a significant hurdle for OpenAI as it tries to maintain its lead in the AI space.

The competitive landscape

The backlash against GPT-6.2 comes at a critical moment for the industry. With Anthropic's IPO looming, the performance of OpenAI's latest model is being scrutinized as a barometer for the company's trajectory. Critics have pointed out that if OpenAI's newest model is less capable than Anthropic's Fable 5.2, it could shift the balance of power in the AI market.

The feedback from developers and investors suggests that the

Sources and links

First reported by gizmodo.com. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.

Liked it? Say why

Add a line for other readers.

No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.

More AI news