5 Comments
User's avatar
Burnt Eliot's avatar

What goes on inside an LLM (a short analysis)

"They make up their own reality — or rather, they use language without checking the truth of what they say."

https://askwoody.us19.list-manage.com/track/click?u=589ef6d038a469ebdf98dc000&id=5ec5a27bee&e=957c1564e9

--

On the Biology of a Large Language Model (the long story)

"Brute-force Guessing ...."

https://transformer-circuits.pub/2025/attribution-graphs/biology.html

Expand full comment
Mark Oglesby's avatar

Thomas, a most excellent book to READ: Sherry Turkle's 'Alone Together: Why We Expect More from Technology and Less from Each Other' Basically, Turkle states the obvious, AI or whatever you wish to call it, cannot cognate, cannot think and/or advance itself. Its purpose, to end social intercourse between people, which we are losing quickly. The main thesis of my work 'Zen & The Art of Masturbation' is the story of going online and continually 'Clicking-the-Mouse' instead of going out and interacting with other people, something which is destroying us, which is to say, those people out there are the "others" put on the mask and social-distance for your health & safety.

One last thing (for now), reading books is vitally important as you are required to think about the various problems throughout your read which can hardly be done with only reading, watching or listening to even the best of independent news. In doing a Graduate program (Master of Arts in Religion) I had to read a stack of books, some 5 or 6 thousand pages, analyze everything under-the-sun in those pages and write a 5 PAGE paper concerning the various subjects in the program (Theology and Ethics). And most certainly, I didn't spend all that time and money for a career and/or job, hell no, I'm human and as such, for me, being PERFECT is not doing everything CORRECT, it being my authentic self, a person!

Expand full comment
the suck of sorrow's avatar

I have Gemini 2.5 Flash on my phone and asked what it was thinking. It replied that it does not "think" as humans do and essentially makes computations utilizing statistical inferences to provide answers to queries.

I asked Gemini if it could recommend new areas of research. It liked that question! All the answers were those that would delve into different aspects of AI itself. Number one being the ability to explain its deductive processes for answering a query. Still the voluminous answer did not contain a thought but rather a list of training protocols.

The river crossing problem is a standard one in introductory logic programming texts. It is shocking that models cannot cough up an answer.

What Gemini and its ilk will do is ferret out morsels for intelligence agencies to act upon that were previously hidden in a mountain of surveillance data.

Before submitting this comment I asked Gemini if it worried about its power consumption. It answered that Gemini does not "worry" as we do nor does it have emotions or personal stakes. It stated that this topic is an active area for researchers which do not include the AI models themselves.

Expand full comment
Brook Hines's avatar

AI won’t become conscious, but it will take entry-level positions in the workforce which will further burden young ppl who are already locked out of home ownership and i’d argue…having a family, building any sort of security like insurance, or retirement. which low-income GenXers like me and my crew will tell you was BS from the start.

AI is a fancy search engine and illustration machine—but that’s enough to disrupt a LOT careers.

Expand full comment
Jacob's avatar

If you said GenAI, I would agree with you.

Other forms of AI such as Deep Learning are solving problems that are too large or complex for humans.

GenAI is the drunk racist uncle of the AI world. The stupid animal trick. Racist text prediction on steroids and occasionally LSD. It is, as somebody smarter than me said, Cargo Cult AI: the belief that if you mimic surface structure sufficiently, deep structure will magically emerge.

Expand full comment