From: hector@hespejo.com   
      
   an0n wrote:   
   > Interesting article shared on Y-Combinator >   
   > https://arxiv.org/abs/2510.18212 trying to quantify how close we are to   
   > AGI by starting with an empirical analysis of basic human intelligence.   
   >   
   > Tl;Dr > The paper attributes overall AGI score of 27% for GPT-4 and 57%   
   > for GPT-5, showing both the rapid progress and the substantial gaps   
   > remaining to reach human-level cognition. The most significant   
   > bottleneck identified is Long-Term Memory Storage (MS); This fundamental   
   > "amnesia" means models must rely on "capability contortions", such as   
   > using massive context windows (Working Memory) to compensate for their   
   > inability to learn and store new information persistently, often the   
   > root cause of hallucinations.   
   >   
   > Soooooo .... not long now boys !! "just" need to solve for these issues   
   > and we are off to the races.   
      
   As much as I like ML / AI, the current state of LLMs which guess the   
   next token will never become something a AGI. Professor LeCun is very   
   clear on this and I he states this in a very lucid way.   
      
   This is a very interesting article that summarizes his statements - much   
   recommended reading it!   
      
   https://eu.36kr.com/en/p/3364112069871367   
      
   --   
   Cheers,   
      
   Hespejo   
   =======   
      
   My page: https://erratia.com/   
   PGP Fingerprint: 26F8 B4C5 30E5 F883 F22A 5B30 62C0 59C3 9B1A C06A   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)