On 2022-11-04, Elon Musk was the “Surprise Guest CEO” at the 29th Annual Baron Investment Conference in New York City. Baron Capital founder Ron Baron interviewed Musk on a wide variety of topics, including the rationale for his investment in Twitter and where he intends to take the company. Interestingly, Musk described Twitter as fitting into his original plans for “X Corporation”. Starting at 42:00, he describes his original plan to make X Corporation “the most valuable financial institution in the world” and how he sees Twitter as part of “executing the X game plan”.
At 52:20 Elon fields a question from an attendee who asks why an investor should pay 70 times earnings for Tesla while they can buy Mercedes-Benz for 6 times earnings and collect a 5% dividend. His answer is…intriguing.
Musk needs to see this video and reorient Twitter (hence news content) accordingly:
Even though it “levels the playing field” for news filters, it is still consistent with his business strategy with “X Corporation” as a network rent business (as was PayPal and as Starlink promises to become).
Musk’s distinguishing businesses, SpaceX and Tesla, are almost pure free enterprise plays – not network rent businesses. Silicon Valley is so full of shit now (sometimes literally right on the sidewalks) that it has lost the ability to pursue anything but network rents. Musk should be able to eat their lunch on a level playing field. Moreover, if we are to believe he is sincere about trying to save humanity, getting freedom of expression back on track is more important than maintaining whatever monopoly rents Twitter can provide during its transition to 2 separate businesses: X (network rents) and optimal news filtering.
My thinking on optimal news filtering has evolved since 1982, but even that ancient idea would be a tremendous advance and as simple as falling off a log. At present, I’d advise Musk to take an approach to filtering out noise with the Algorithmic Information Criterion for model selection to disclose what “is” or a rigorous definition of “truth” within the natural sciences. As people are obviously aware, the large language models are being paraded around as though the parameter count is some sort of virtue that indicates degree of quality in its responses when the Algorithmic Information Criterion warns us that this is bullshit.
A “truth” filter based on AIC (A = Algorithmic, NOT Akaike) would consist of engaging readers as first-stage lossless compressors of the unfiltered “news”, with a hierarchy of humans (such as the “editorial staffs” I suggest in the video) that compete in a lossless compression marketplace for truth discovery. The kernel of this Algorithmic Information would be something like a “large language model” – encoding virtually the entire content of the current corpora used by the LLMs – except that:
It would aim for minimum parameter count.
It would not be permitted to throw away any “news” in its (singular) compressed executable archive.
It would make extensive use of human intelligence.
It would not rely on specious statistical – the dog ate my homework – models like transformers (which are really the devil spawn of the social pseudoscience reliance on statistical models) but insist on recurrence so that logical reasoning is inherent to the model.
In terms of competing for customer engagement with the news service, it boils down to customizing the presentation to the customer based on a comparison between the customer’s knowledge and the news service’s knowledge in a manner not unlike computer based education’s use of drill and practice: All interactions perform 2 functions: Placement and education. This iterates under some model of the “student” that includes things like the short term memory to long term memory transition of key facts as well as an understanding of what the student’s goals are.