𝐉𝐚𝐠𝐠𝐞𝐝 𝐅𝐫𝐨𝐧𝐭𝐢𝐞𝐫 – 𝐏𝐚𝐫𝐭 𝟐

The jagged frontier - that unpredictable line between what Generative AI can and can’t do well - isn’t just about AI’s limits. It is also about how humans work with GenAI. Some people hand off tasks. Others work in lockstep. Both are navigating AI’s jagged edge.

A Harvard working paper identified two patterns:

• 𝐂𝐞𝐧𝐭𝐚𝐮𝐫-𝐬𝐭𝐲𝐥𝐞 – like the half-human, half-horse myth, humans delegate clearly-defined subtasks to AI, maintaining separate roles.

• 𝐂𝐲𝐛𝐨𝐫𝐠-𝐬𝐭𝐲𝐥𝐞 – humans tightly weave their prompting into AI’s outputs in feedback loops, blending efforts at every step.

𝐂𝐞𝐧𝐭𝐚𝐮𝐫𝐬 𝐦𝐚𝐢𝐧𝐭𝐚𝐢𝐧 𝐜𝐥𝐞𝐚𝐫 𝐛𝐨𝐮𝐧𝐝𝐚𝐫𝐢𝐞𝐬. 𝐂𝐲𝐛𝐨𝐫𝐠𝐬 𝐛𝐥𝐮𝐫 𝐭𝐡𝐞 𝐝𝐢𝐬𝐭𝐢𝐧𝐜𝐭𝐢𝐨𝐧.

In the paper, these weren’t fixed types; people switched depending on the task. 𝐒𝐮𝐜𝐜𝐞𝐬𝐬 𝐨𝐟𝐭𝐞𝐧 𝐞𝐦𝐞𝐫𝐠𝐞𝐬 𝐧𝐨𝐭 𝐝𝐢𝐫𝐞𝐜𝐭𝐥𝐲 𝐟𝐫𝐨𝐦 𝐭𝐡𝐞 𝐀𝐈’𝐬 𝐜𝐚𝐩𝐚𝐛𝐢𝐥𝐢𝐭𝐲 𝐚𝐥𝐨𝐧𝐞, 𝐛𝐮𝐭 𝐟𝐫𝐨𝐦 𝐝𝐲𝐧𝐚𝐦𝐢𝐜 𝐢𝐧𝐭𝐞𝐫𝐚𝐜𝐭𝐢𝐨𝐧𝐬 𝐛𝐞𝐭𝐰𝐞𝐞𝐧 𝐡𝐮𝐦𝐚𝐧 𝐚𝐧𝐝 𝐥𝐚𝐫𝐠𝐞 𝐥𝐚𝐧𝐠𝐮𝐚𝐠𝐞 𝐦𝐨𝐝𝐞𝐥𝐬 (LLMs).

That’s one of the messages, from the frontier, that I find interesting in this paper: value isn’t just in what GenAI (LLMs like Claude, Gemini, and ChatGPT) can do, but in how we collaborate with it.

You can find Jagged Frontier – Part 1 here and part 3 here