Wormholes, psychadelic drugs, and a dryer - 16 July 2023

 Well, this looks interesting.   

July 17 at 9am ET, a new web-only series by Steven Soderbergh called Command-Z premieres in which Michael Cera leads a team using a wormhole in a washing machine to alter the present by traveling back in time.

https://extension765.com/blogs/soderblog/command-z

Is it time to unretire? - 15 July 2023

Retirement is a gift.  No, really.  If you are healthy and wealthy enough, retirement is a great thing.  I can see there is a definite down-side for those who cannot afford to retire, but my only regret is that I did not try a half-time work schedule before I retired.  

Now we find that the universe might be older than conventionally thought.  Rather than 13.7 billion years of universal evolution, it may be as old as 26.7 billion years.  Fortunately, the age is added to the early years of the universe (to allow galaxies to form).

https://phys.org/news/2023-07-age-universe-billion-years-previously.html

Trouble in AI Land - 13 July 2023

Forgetting.  As I understand the European data rules, one has the right to be forgotten.  At first, this sounds odd.  Although it can be abused, it makes sense when you think of it.  If a mistaken or outrageous screed about you gets injected into the Internet, you should have the right to get it removed - to be forgotten.  Now think about AI engines.  

Similar to search engines, AI engines today will crawl the web looking for input materials.  No single AI company can generate the vast amount of input data needed for AI training, so they pretty much are left with crawling the web.  The information they find is incorporated into the training models and used in the "predictive" part when the AI model generates output.  It is likely that somewhere in the depths of ChatGPT that there is a copy of Moby Dick, Shakespeare, and your master's thesis, all wrapped up in "the model" and existing as a latent copy of your work.  

Now, suppose you wish for your thesis to be forgotten, removed from the Internet, relegated to nothingness.  You can get the data pulled from a search engine, but how does one get data pulled from an AI model?  The developers do not know how information is incorporated into the model (whle they know in vague terms, "the algorithm", they cannot trace any particular data item into the model).  I suppose they would have to retrain using the old input dataset but without your work, and that would be very expensive.  Training is often the most expensive part of developing an AI model.

To emphasize this problem, there is already a lawsuit by someone who claims that their copyrighted material has been included in an AI model without their consent (ref. Sarah Silverman).

So the AI folks are in a pickle.  

https://www.theregister.com/2023/07/13/ai_models_forgotten_data/?td=rt-3a