Borges' cartographers and the tacit skill of reading LM output
7 hours ago
- #artificial intelligence
- #digital literacy
- #critical thinking
- Borges' story illustrates that maps are useful because they are reductions; when they lack compression, they lose their purpose.
- Language models (LMs) are compared to maps, having evolved beyond mere representations to actively reshape the territory they describe, requiring users to develop skills to engage critically with them.
- Baudrillard's four stages of representation apply to LMs: from faithful copies to distorting reality, masking its absence, and potentially detaching completely, with transitions between stages being slippery and hard to detect.
- Unlike traditional maps, LMs are interactive and personalized, with outputs varying based on prompts and user backgrounds, offering flexibility but also challenging shared understanding.
- LMs serve as useful tools for summarization and reflection, clearing clutter and aiding in tasks like research or coding, but they also become objects of study themselves, shifting focus from the territory to the map.
- Effective use of LMs involves developing tacit knowledge—similar to clinical intuition or code smell—to sense when outputs are unverified, too smooth, or in need of verification, a skill learned through practice rather than rules.
- The analogy of LMs as maps is limited; they are also tools, economic forces, and research subjects, with future improvements making the skill of critical engagement even more important.
- A balanced approach is needed between obsession with LMs and indifference, emphasizing the delicate practice of using them well while staying honest about their evolving role in our lives.