Theory Thursday: Are You Reading My Article, or a Model’s Reconstruction of It?

Art by @basilonmypizza: https://lnkd.in/eF8FkWzN - https://basilhefti.ch/

Sounds like a trivial question. Yet it is interesting from many angles.

The obvious one is distribution. For a long time, publishing followed a simple contract: one creates, others consume. The creator receives attention, feedback, attribution, maybe revenue.

Generative systems such as ChatGPT reshape that contract. The original source (a website, paper, image, song) is consumed indirectly: generative systems read it, learn and extract from it, summarize it, combine it with other material, and interact with the user on its behalf.

https://flipbook.page pushes this idea to the extreme: a web experience generated on request, demonstrating the power of today’s generative systems.

This distribution shift (from visit to inference) is amplified with a second, computational shift, from weights to search. Early LLMs were mostly closed-book: they generated answers from patterns stored in their weights. Newer systems search. Sometimes externally, across documents and the web. Sometimes internally, across possible reasoning paths. They sample, compare, verify, rerank, and construct the answer through his process.

It is tempting to reach for analogies such as music going digital or newspapers losing classifieds, and conclude that creators will simply find new income models.

Maybe. What is clear: the economics are shifting, and the position to capture the user’s attention, trust, and willingness to pay is being actively contested.

The slogan “knowledge becomes free” is incomplete. Knowledge still has to be created, edited, maintained, and updated. The opportunity is to find a sustainable economic model. The search is underway.

So, are you reading this? Let me know ;)

(Statistically, a model probably read it first.)

• Wishing you a very special birthday, Sara

• Art: https://lnkd.in/eF8FkWzN https://basilhefti.ch/

• Impact of AI Search Summaries on Website Traffic: Evidence from Google AI Overviews and Wikipedia https://lnkd.in/eXxmT5_g

• From weights to search: AgentIR: Reasoning-Aware Retrieval for Deep Research Agents, https://lnkd.in/ev9BNkn4

• Ideas for economic models are scarce. The most practical approach as of now is really simple licensing, https://lnkd.in/e-DMvxq5