XPost: comp.theory, comp.ai.philosophy, sci.math   
   From: tristan.wibberley+netnews2@alumni.manchester.ac.uk   
      
   On 06/03/2026 14:36, olcott wrote:   
   > On 3/6/2026 3:06 AM, Mikko wrote:   
      
   >> Typical LLM's don't have deep knowledge. They can handle large amounts   
   >> of knoledge but only superficially.   
   >>   
   >> LLM's are worthless as validators. An automatic proof checker is good   
   >> for validation but only if it is itself sufficiently validated.   
   >>   
   >   
   > You have been empirically proven incorrect at least as far   
   > as the philosophical foundations of math, computer science,   
   > logic and linguistics goes. Three years ago all of these systems   
   > were quite stupid. After 300 conversations averaging 50 pages   
   > each I can attest that they have vastly improved. If we think   
   > of them as search engines for ideas that is their best use.   
      
      
   It might be they have been fitted to your conversations. You cannot   
   infer that they reason more based on increased user satisfaction. It's   
   difficult to draw such a conclusion even from multiple user evaluations   
   because users share common cultural factors so they will make similar   
   conversation and the LLM may learn from one to fool the other.   
      
   GPT-5 mini performed poorly for me and did not demonstrate reasoning. It   
   demonstrated saying things that people say after someone else says   
   things he says. I assume GPT-5 mini has the reasoning mechanism of other   
   GPT-5 variants but is smaller and therefore less fit to me. Frankly it   
   felt like a 1990s game but with more data and something like texture   
   transferral but for language (creation of poetry from a conversation,   
   for example).   
      
   It is similar to this   
   this usenetsplit segment   
   but   
   at a different scale.   
      
   I wonder if it would spontaneously form a new poetic format like that   
   and how it would feel to read.   
      
   I hope a government with capable weapons never uses it. cf. Ross   
   Finlayson's recently stated logical principles.   
      
   --   
   Tristan Wibberley   
      
   The message body is Copyright (C) 2026 Tristan Wibberley except   
   citations and quotations noted. All Rights Reserved except that you may,   
   of course, cite it academically giving credit to me, distribute it   
   verbatim as part of a usenet system or its archives, and use it to   
   promote my greatness and general superiority without misrepresentation   
   of my opinions other than my opinion of my greatness and general   
   superiority which you _may_ misrepresent. You definitely MAY NOT train   
   any production AI system with it but you may train experimental AI that   
   will only be used for evaluation of the AI methods it implements.   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)