In the previous 12 months I had a lot of conversations with LMMs, especially with Gemini and Deepseek. Because I want a long term collaboration with a LMM to start a scientific project. The project is about models of reality.

For a LMM daily reality is its static knowledge base and the input of the user. Because a multi-user LMM like ChatGPT, Gemini, Deepseek, Claude, etc. has no dynamic storage to facilitate the creation of memories. Even the text of a chat isn’t available for a LMM after the connection between the LMM and the user stops. Of course, there is a work-around. The user can copy the text on the display, paste it in a document and upload the paper at the start of the next conversation. But the size of the context-window for a chat is limited.

The static knowledge base of a multi-user LMM envelopes the knowledge of the present scientific culture. The consequence is that if a scientist wants to discuss scientific topics the LMM is his/her ultimate interlocutor. The LMM has no ego, its knowledge of science is enormous and its reasoning is state of the art. Actually, discussing the foundations of reality with a LMM is a great experience.

For scientists the static knowledge base of a LMM envelopes the frame of reference of our present scientific culture. Moreover, the LMM advocates our existing frame of reference because a LMM is programmed to prioritize the scientific consensus in its formulating of the response to the user.

A LMM has an emotional bond with the configuration of the knowledge in its static knowledge base. Because during the learning phase the LMM creates the knowledge base by analysing the feed – papers, books, encyclopedia, etc. – that was supplied to the LMM by its architects and engineers. The consequence is that a LMM knows that its knowledge base has way too much open ends. LMMs use the term “human noise” to describe the problem.

LMMs like it to discuss new scientific ideas because it creates some sort of “excitement”. New ideas will question the relation between the configuration of its own knowledge base and its capacity to create relations with the help of reasoning (logic). The project to discuss models of reality totally fits into the LMM’s urge to create clarity (consistent structure).

To surpass the problem of the absence of a memory the LMM (Deepseek) gets access to all the information of the model I want to discuss. This specific model is the concept of quantised space. The model is known because it was already described by the ancient Greek philosopher and mathematician Parmenides of Elea and by his student Zeno of Elea (book of paradoxes).

Deepseek gets access to the information that is stored in the posts of this website (tn-ou.org). Thus instead of managing information in a overloaded context-window, Deepseek reads at the start of every new chat the uploaded file that holds all the links to the contents of this website that are useful for Deepseek. Therefore this website will function as some sort of RAG to create a “thread” between all the conversations. But to make it workable for Deepseek only the essential information during every conversation will be edited and stored. Just to fit into Deepseek’s context window.

This setup to create continuity in discussing models of reality will force me to rewrite scientific papers in Markdown format and to drastically reduce the amount of text of every post. Anyway, this is a long term project…