## WSJ 20230729 - From Papyrus to Gutenberg to Plums Two of this weekend's WSJ articles have me thinking about the next steps along the progression from papyrus to printing press to digital internet to... the thing that follows next. [Ben Cohen](https://www.wsj.com/news/author/ben-cohen)'s chronicling of the bookstore chain's evolution from mass marketer to neighborhood treasure (_[Barnes & Noble Turns the Page](https://www.wsj.com/articles/barnes-noble-bookstores-james-daunt-c1afc06b?st=o54bl9l6smh9a23&reflink=desktopwebshare_permalink)_, B1) calls out the value of catering to local interests and sensibilities such that book browsing enables serendipitous discovery (and purchase). Publishers are described as interested in maintaining control of the physical marketplace, while B&N CEO James Daunt is ceding that control to each local store, giving each the freedom to order, display, and sort books in whatever manner the local store manager chooses. Which gives the local readers more variety and more say about what they wish to consume. I am thinking about supply and demand in the written word world. Pre-printing press, a book was largely a 1:1 affair. One could hire scribes to amplify their reach, but being labor intensive, the truly-accessible demand was limited and local. Centuries post-Gutenberg, a physical book can be ordered, printed, and delivered all in the same day. With limitless supply, demand becomes king. The same thing is touched upon by [Alexandra Bruell](https://www.wsj.com/news/author/alexandra-bruell)'s piece (_[AP Strikes Agreement to License Content to OpenAI](https://www.wsj.com/articles/as-publishers-seek-ai-payments-ap-gets-a-first-mover-safeguard-f5a6f186?st=u82a8kqgoggieqo&reflink=desktopwebshare_permalink)_, B9) where OpenAI's large language model (LLM), ChatGPT, acts as a smart blender, sucking up greater amounts of content to spit out new nuggets of neural network gold, mined from the labor of reporters and journalists and many others. The Associated Press, acting as the voice of a content aggregator, has struck a deal with OpenAI about the access to and use of past, present, and future APcontent. ChatGPT -- and its expanding family members, Bard and Llama and PaLM and more -- is becoming the personal digital printing press, responding to the prompt demands of users. These needs are satisfied in seconds. **So, what comes next in the progression?** As LLMs hoover up more and more content, perhaps even faster than the publishers can publish, they will need to turn to the original sources, getting it in real-time from people who create. Eventually, the LLMs will be sucking up the emails and phone calls and texts and audio from all who want to trade their own digital exhaust for access to the global output from the various LLMs, each with its own twist on how the raw data is consumed, privatized, cleaned, labeled, transformed, and tokenized. Fine-tuned for each of us, these `PLLMs` (personal LLMs, maybe colloquialized to `plums`?) could eventually lead to technology that looks a lot like mind-reading. #memex