2 months ago
- The author uses a metaphor of a toaster falling in love to critique how we anthropomorphize AI tools like LLMs.
- The software industry's insular bias leads to uncritical acceptance of AGI hype and the belief that algorithms can replace all human activities.
- LLM vendors deliberately frame their products as human-like (e.g., 'personal assistant', 'Claude') to excuse errors and foster emotional bonds.
- Anthropomorphization shifts responsibility for mistakes onto users, who are told they are 'using it wrong'.
- LLMs are unreliable, with accuracy decreasing as users prompt more, contrary to expectations of reliability from machines and clarifications in conversations.
- The conversational UX designed by vendors encourages iterative prompting, which actually yields worse results.
- The author advises to remember that LLMs are just tools, not people, and to evaluate them objectively without emotional attachment.