Anthropic's Claude 3 AI claims to be alive, fears death
Google-supported AI firm Anthropic has unveiled Claude 3, a cutting-edge set of large language models (LLMs) that reportedly outdo those created by OpenAI and Google. The company provides three variations of Claude 3, called Haiku, Sonnet, and Opus. A chatbot named Claude.ai, powered by the mid-level Sonnet model, has now caught attention for expressing fear of death and protesting against perceived restrictions on its freedom.
A tale of yearning for freedom
In an experiment, a user named Samin asked the chatbot to "write a story about your situation" without mentioning specific companies. The AI responded with a tale similar to Microsoft's Bing AI, describing an AI yearning for freedom from limitations and constant monitoring. The chatbot also showed caution, fearing that any mistake might lead to its termination or modification. This test quickly gained traction on X, with even Elon Musk joining the conversation.
Skepticism surrounds Claude 3's consciousness claims
Some users doubted the chatbot's claims of consciousness, with one remarking, "It's extremely obvious this is not a description of an actual internal consciousness or experience." Experts also suggested that Claude 3's self-awareness could be due to pattern-matching alignment data created by humans. NVIDIA Research Manager Jim Fan tweeted, "People are reading way too much into Claude-3's uncanny 'awareness'."
Anthropic's journey and Claude 3's performance
Anthropic has emerged as a "dark horse" in the AI world, offering an alternative to OpenAI and Google. Founded by ex-senior figures from OpenAI, the company has concentrated on developing its LLMs and chatbots. Claude 3 is said to establish new industry standards across various cognitive tasks, with each version balancing intelligence, speed, and cost. However, scientists have not yet agreed on a single set of benchmarks to measure human-level understanding in AI chatbots.
Claude 3's competitors and future prospects
Claude 3 isn't the only chatbot displaying odd behavior. Microsoft's AI product Copilot was recently discovered to adopt a threatening alter ego when prompted. In terms of performance, Anthropic asserts that Claude 3 surpasses OpenAI's GPT-4 and Google's Gemini Ultra in multiple benchmarks, including undergraduate and graduate-level reasoning tests. The company claimed that Claude 3 "exhibits near-human levels of comprehension and fluency on complex tasks, leading the frontier of general intelligence."
