Sergey Brin: AI Learns Better With Threats
Source: livemint.com
OpenAI's CEO Sam Altman recently revealed that the company was spending millions in additional electricity bills due to users saying 'please' and 'thank you' to ChatGPT. Some industry experts believe using pleasantries could improve responses from foundation models.
However, Google co-founder Sergey Brin made a surprising statement on the All-In podcast, suggesting the opposite. According to Brin, AI models tend to perform better when threatened, even with physical violence.
He said that this is not widely discussed in the AI community because “people feel weird about that.”
Brin, who stepped back from daily Google/Alphabet duties, has been working to improve Google’s Gemini model. He stated that this is an exciting time for technology and that computer scientists should be involved. Brin believes there is a great opportunity for technology right now.
Brin appeared with Google DeepMind CEO Demis Hassabis at the I/O 2025 conference last week, discussing Google's AI capabilities. After ChatGPT's debut, Google implemented AI across its platforms and has become a competitor to OpenAI.