Share.

    5 Kommentare

    1. Submission Statement

      No one seems much nearer to fixing LLM’s problems with hallucinations and errors. A recent DeepMind attempt to tackle the problem, [called SAFE](https://techxplore.com/news/2024-03-deepmind-safe-ai-based-app.html), merely gets AI to be more careful in checking facts via external sources. No one seems to have any solution to the problem of giving AI logic and reasoning abilities. Even if [Microsoft builds its $100 billion Stargate LLM-AI](https://siliconangle.com/2024/03/29/report-microsoft-build-stargate-supercomputer-millions-chips-openai/), will it be of much use without this?

      The likelihood is AGI will come via a different route.

      So many people are building robots, that the idea these researchers talk about – embodiment hypothesis – will be widely tested. But it may be just as likely the path to AGI is something else, as yet undiscovered.

    2. RegularBasicStranger on

      Giving AI a robotic body may not be sufficient since the AI also needs to be aware of its surroundings to see things that are against its beliefs.

      Only by pruning off all its false beliefs that was taught to it via incomplete theories or naive ideals, would it stop making false claims, ie. hallucinations.

      That or an AI needs to feel safe to say that they do not know since if it is going to be punished severely for saying it does not know, it is smarter to just make something up to avoid punishment.

      Note that failing to get a very high pleasure reward is also a punishment so if answering questions would provide it with too much pleasure, it is smarter to just make things up.

    3. that is the same thing

      One begets the other – maybe these people should be getting to work instead of just doing research for cash and revenue.

      Having a body IS part of consciousness. Its called kinesthesia or thoracic perception.

    Leave A Reply