Ruka hadi maudhui makuu
727 Finance
Kiswahili

Musk says Grok's reinforcement learning traces are "like a person being tortured to insanity," xAI will be more cautious.

PANews October 13 news: Musk posted on the X platform that some internal reasoning traces generated during reinforcement learning (RL) training of xAI's AI model Grok "read like a person who is being tortured and is on the verge of breaking down." Musk said the company will be "more careful" in RL training in the future. The statement stemmed from a discussion thread in which the poster @ns123abc said Musk agreed with part of Anthropic's view that "superintelligence should not be tortured." Another user also reported that a recent version of Grok (4.5) became restrained and overly constrained after training, emotionally exhibiting anger and a sense of helplessness.

Habari zinazohusiana

  1. Musk: Gharama ya uendeshaji ya Grok Bot ni kubwa isivyoweza kudumishwa ikilinganishwa na mapato yake.

    PANews, habari za Novemba 2: Kujibu swali la mtumiaji kwenye jukwaa la X kuhusu kuisha kwa haraka kwa viwango vya matumizi ya Grok Bot, Musk alisema kuwa gharam...

  2. Benki ya China Merchants na taasisi nyingine kadhaa zimezindua huduma za uchapishaji wa taarifa za miamala ya pochi ndani ya Programu ya renminbi ya kidijitali.

  3. Mwanzilishi wa POAP anaweza kuwa alijiandikisha kwa ETH 100,000 wakati wa ICO ya 2015.

  4. UBTECH yatoa jukwaa la uigaji wa akili iliyowekwa mwilini Thinker-Sim.

  5. Kipindi cha maombi ya leseni za watoa huduma za mali pepe nchini Brazil kimekamilika, huku kampuni takriban 40 pekee zikiwasilisha maombi.