OpenAI's recent GPT-5 livestream showcased charts that presented misleading data representations. There were discrepancies in the scales used for their deception evaluation charts, causing confusion over GPT-5's performance compared to previous models. CEO Sam Altman labeled one of the charts as a "mega chart screwup," while a marketing staffer acknowledged the mistake as an "unintentional chart crime." The situation casts doubt on OpenAI's claims regarding advancements in the new model, particularly in regard to reducing hallucinations.
The scale of the deception evaluation charts for GPT-5 is inconsistent, with bars misrepresenting the scores of GPT-5 and its predecessors.
CEO Sam Altman referred to the chart inconsistencies as a "mega chart screwup," highlighting the lack of clarity in the presented data.
Collection
[
|
...
]