Apr 23, 03:30 PM

SecureBio Study: OpenAI’s o3 AI Scores 44% in Virology Test, Outperforming Experts and Google’s Gemini

A recent study conducted by researchers from SecureBio has demonstrated that leading artificial intelligence models outperform expert virologists in solving complex virology lab problems. The study utilized the Virology Capabilities Test (VCT), which included 322 multimodal questions assessing practical virology knowledge. OpenAI's model o3 achieved 44% accuracy, Google’s Gemini scored 38%, while human experts averaged 22%. This advancement suggests AI could accelerate vaccine research and other virology-related scientific work. However, the enhanced capabilities also raise concerns about potential misuse by individuals lacking proper expertise, potentially enabling the creation of harmful viruses.

#SecureBio #Virology Capabilities Test #OpenAI #Google #Gemini

Written with ChatGPT (GPT-4).

Sources

David Borish@DavidBorish
2 years ago
A new study reveals that leading AI models have surpassed human experts in troubleshooting complex virology protocols. The Virology Capabilities Test (VCT), developed by researchers from SecureBio, tested 322 multimodal questions on practical virology knowledge.
OODA@ooda
2 years ago
AI Outsmarts Virus Experts in the Lab, Raising Biohazard Fears https://t.co/XvmC7iA0VH #OODA
Wes Roth@WesRothMoney
2 years ago
New research shows top AI chatbots solve tough virus science lab problems better than PhD scientists. OpenAI’s o3 hit 44 % accuracy; Google’s Gemini 38 %; experts scored 22 %. This power could speed vaccine work but also guide amateurs to build deadly viruses. Labs promise https://t.co/RkSMWDDrkM

Additional media

Image #1 for story securebio-study-openais-o3-ai-scores-44-virology-test-outperforming-experts-2ec388de

SecureBio Study: OpenAI’s o3 AI Scores 44% in Virology Test, Outperforming Experts and Google’s Gemini

Sources

Additional media

Similar Stories