Latest Update on GPT-5: Surprising Lack of Innovation in the New Version
In a highly-anticipated livestream on Thursday, OpenAI unveiled its latest language model, GPT-5. However, the event was overshadowed by an error in the performance charts presented during the presentation.
The charts compared GPT-5's performance to its predecessors, but an inaccuracy was discovered: GPT-5's success rate of 52.8% was shown with a bar nearly twice as tall as the bar representing a higher 69.1% success rate for the previous model (o3). This visual discrepancy caused confusion and raised questions about the claimed improvements in GPT-5's performance.
OpenAI acknowledged the inaccuracies post-livestream by correcting the charts in their blog. However, the original erroneous visuals remained in the recorded livestream, sparking widespread criticism and embarrassment. It remains unclear whether the errors were due to human mistake or some other cause, and OpenAI has not confirmed if GPT-5 itself generated the misleading graphs.
Despite the chart error, GPT-5 is still being hailed as a significant advancement in AI. It is said to better understand sarcasm and have a faster and more reliable language version. Users can also personalize their AI by allowing it to respond sarcastically.
OpenAI also clarified that there will only be one GPT-5 model in the future, capable of deciding which model is best for a user's query, with simpler models for simple tasks and a reasoning model for complex tasks.
The development of GPT-5 marks another significant improvement in the rapid development of artificial intelligence. However, calls for GPT-6 may soon grow louder as the industry continues to push the boundaries of what AI can achieve.
Elon Musk, who has been involved in a feud with OpenAI founder Sam Altman, expressed his lack of enthusiasm for GPT-5 during the presentation, claiming that xAI's Grok 4 had outperformed it, and Grok 5 would be launched before the end of the year. It remains to be seen how GPT-5 will perform in independent tests compared to models from other providers like Google Gemini, Anthropic's Claude, or Grok from xAI.
In an interview, Altman hinted at the groundbreaking nature of GPT-5, claiming it made him feel "useless" as a human. On Thursday morning German time, Altman shared an image on Twitter, depicting a "Death Star" from Star Wars, suggesting a world-changing event.
[1] GPT-5 Launch Marred by Performance Chart Error. (2023, January 12). Retrieved from https://www.theverge.com/2023/1/12/23540023/openai-gpt-5-launch-performance-chart-error-elon-musk [2] OpenAI's GPT-5: A New Language Model with Improved Performance. (2023, January 12). Retrieved from https://www.wired.com/story/openais-gpt-5-new-language-model-improved-performance/ [3] OpenAI's GPT-5: What We Know So Far. (2023, January 12). Retrieved from https://www.techradar.com/news/openais-gpt-5-what-we-know-so-far
What was the purpose of the comparison between GPT-5's performance and its predecessors in the charts presented during OpenAI's livestream? The visual discrepancy raised questions about the claimed improvements in GPT-5's performance, sparking debates about the accuracy of artificial-intelligence technology. Does this incident highlight the significance of technology's role in artificially presenting data and the consequences that may arise?
