How Marc Andreessen says he sets up his AI chatbot — and why critics are skeptical it works

Marc Andreessen, co-founder of Andreessen Horowitz, has expressed his desire for a more assertive and less courteous AI chatbot. In a recent post on X, he shared his customized AI prompt, emphasizing a need for systems that are 'provocative, aggressive, argumentative, and pointed.' This statement highlights Andreessen's growing criticism of what he perceives as overly cautious 'woke' restrictions within the AI field. His vision for a more combative AI is not without its detractors. Renowned psychologist and neural science expert Gary Marcus, who has often taken a critical stance on mainstream AI developments, pointed out the prompt’s expectation for flawless accuracy. Similarly, Zach Tratar, an AI engineering leader at Notion, critiqued the prompt as outdated, noting that the tactics Andreessen advocates may not yield the desired results with modern large language models. These critiques underline a significant limitation in today’s AI technology: even the most precise instructions do not ensure reliable performance. Current language models can still produce inaccuracies, overlook directives, or fail to verify their responses, particularly when faced with complex or contradictory instructions. The conversation also sheds light on a deeper schism within the AI community. Organizations like OpenAI and Anthropic have invested considerable effort into creating safeguards for their models, striving for safety, reliability, and broad applicability. In stark contrast, Andreessen’s approach calls for a reduction of these constraints, even suggesting that the AI should steer clear of moral or ethical discussions unless specifically prompted.