OpenAI’s latest GPT-5 chatbot has received a mixed reception, prompting CEO Sam Altman to issue multiple apologies on X. The launch video featured poorly presented charts that misrepresented performance metrics, notably the SWE-bench results. Initially, GPT-5 was shown to outperform OpenAI o3 and GPT-4o significantly, but errors in chart proportions misled viewers. Additionally, a second graph indicating GPT-5’s “coding deception” rate inaccurately presented data, creating confusion. Although OpenAI later corrected these figures and clarified the context behind the metrics, skepticism remains about the accuracy of the reported scores, particularly after comparisons with Anthropic’s Claude Opus 4.1 model. This situation raises concerns about OpenAI’s accountability and the rigor of its data presentation, reflecting broader issues within the AI industry regarding transparency and competence. The discrepancies have even caught the attention of industry figures like Elon Musk, leaving users anxious about AI safety management.
Source link
