Artificial intelligence has the potential to replace workers, spread “disinformation,” and enable cyberattacks, OpenAI CEO Sam Altman has warned. The latest build of OpenAI’s GPT program can outperform most humans on simulated tests.
“We’ve got to be careful here,” Altman told ABC News on Thursday, two days after his company unveiled its latest language model, dubbed GPT-4. According to OpenAI, the model “exhibits human-level performance on various professional and academic benchmarks,” and is able to pass a simulated US bar exam with a top 10% score, while performing in the 93rd percentile on a SAT reading exam and at the 89th percentile on a SAT math test.
“I’m particularly worried that these models could be used for large-scale disinformation,” Altman said. “Now that they’re getting better at writing computer code, [they] could be used for offensive cyber-attacks.”
“I think people should be happy that we are a little bit scared of this,” Altman added, before explaining that his company is working to place “safety limits” on its creation.
These “safety limits” recently became apparent to users of ChatGPT, a popular chatbot program based on GPT-4’s predecessor, GPT-3.5. When asked, ChatGPT offers typically liberal responses to questions involving politics, economics, race, or gender. It refuses, for example, to create poetry admiring Donald Trump, but willingly pens prose admiring Joe Biden.
Altman told ABC that his company is in “regular contact” with government officials, but did not elaborate on whether these officials played any role in shaping ChatGPT’s political preferences. He told the American network that OpenAI has a team of policymakers who decide “what we think is safe and good” to share with users.
Read More: ChatGPT creator warns of AI dangers