In a notable achievement in the field of artificial intelligence, the GPT-6-Astra model has solved 49 out of 50 unpublished theoretical physics calculations on the TheoPhysEval benchmark. This performance showcases the model’s capability to compete with leading specialists in the field.
Interestingly, in instances where GPT-6-Astra appeared to falter, it was discovered that the reference solutions provided by human experts contained errors. Researchers have acknowledged that while this breakthrough is significant, the realm of physics has not yet experienced the same level of advancement seen in mathematics, primarily due to the agents’ perceived lack of ‘taste’ in understanding complex concepts.
Compiled automatically by the Tech AI Newsdesk from public AI-news sources and summarised in our own words.
