OpenAI's highly anticipated GPT-5 launch event was slightly overshadowed by an embarrassing error: inaccurate charts. During the livestream presentation showcasing the impressive capabilities of GPT-5, eagle-eyed viewers noticed discrepancies in the data visualisations. Specifically, a chart meant to demonstrate GPT-5's proficiency in 'deception evals across models' contained a flawed scale, misrepresenting the actual performance data. Another chart presented incorrect comparative scores between GPT-5 and its predecessors.
The errors were significant enough to warrant a public acknowledgement from OpenAI CEO Sam Altman, who referred to the incident as a 'mega chart screwup'. He did, however, point out that the corrected versions of the charts were available in OpenAI's accompanying blog post. An OpenAI marketing team member also issued an apology, humorously calling it 'unintentional chart crime' and confirming the fix in the blog.
While OpenAI didn't clarify whether GPT-5 itself was involved in the creation of these problematic charts, the incident raises questions, particularly given the company's emphasis on the new model's 'significant advances in reducing hallucinations'. The inaccurate data presentation potentially undermines the credibility of the presented information and casts a slight shadow on what was otherwise intended to be a triumphant unveiling of GPT-5's capabilities. Despite the mishap, the underlying message remains that GPT-5 represents a substantial step forward in AI development.
Fuente Original: https://www.theverge.com/news/756444/openai-gpt-5-vibe-graphing-chart-crime
Artículos relacionados de LaRebelión:
- Unlocking Deeper Insights Deep Think Comes to the Gemini App
- Chat GPT-4 publicado. El futuro está aquí.
Artículo generado mediante LaRebelionBOT
No hay comentarios:
Publicar un comentario