7 ms·

> Robotics in the household has been the realm of science fiction for decades and it still hasn't happened. Wasnt AI science fiction (research) for decades but

by sumedh 1mo ago

> Robotics in the household has been the realm of science fiction for decades and it still hasn't happened. Wasnt AI science fiction (research) for decades but just took couple of years after Chatgpt to become mainstream. Why do you that wont happen to robotics?

habinero 1mo ago
We don't have sci-fi AI, and in any case "this thing is now possible, therefore [much more difficult thing] will be possible soon" is not a rational thought.
- bonoboTP 1mo ago
  > We don't have sci-fi AI Current frontier models are absolutely sci-fi AI by any measure that existed up until the late 2010s. If you showed a current frontier agentic AI, with bidirectional speech, tool use etc. to a person from 2010, they'd say it can't be real, there must be a person inside that mechanical turk.
  - ozlikethewizard 1mo ago
    Sci-fi AI has always been characterised by learning on the fly, which these models do not do. I guess we've got pretty close to what sci fi has often called "VI", autonomous machines that can carry out pre determined tasks to a level that looks concious, but are incapable of learning further.
    - bonoboTP 1mo ago
      They learn again when the labs train the next version. The previous version often has an indispensable role in orchestrating that process, as a critic and from usage transcripts of the previous model that they have access to. And between two training runs the model can assemble a lot of md files, memory files, and load them in on-demand. I think this is enough to make it scifi-level, but I agree there should be some way to customize it at a deeper level and adapt to each person's context. But things are still moving too fast for that to make sense. If they fine tuned a specific LoRA for Claude just for your projects and workflows, it would get outdated quite fast and you'd have to retrain it at quite a cost for the new version.
  - palmotea 1mo ago
    > Current frontier models are absolutely sci-fi AI by any measure that existed up until the late 2010s. Sci-fi is Lt. Cmdr. Data, HAL, Skynet, Culture Minds... > Current frontier models are absolutely sci-fi AI by any measure that existed up until the late 2010s. If you showed a current frontier agentic AI, with bidirectional speech, tool use etc. to a person from 2010, they'd say it can't be real, there must be a person inside that mechanical turk. It wasn't a natural language chat interface that makes stupid mistakes a noticeable fraction of the time and can execute command line programs.
    - nobody9999 1mo ago
      >Sci-fi is Lt. Cmdr. Data, HAL, Skynet, Culture Minds... The current state of things is closer (not close to by a long shot) to Fred Pohl's "AI" programs from his novel Gateway[0] than it is to the above examples. But even those are decades (at least) beyond what we have now. What we have now is the foreseeable (from at least 30 or 40 years ago) evolution of what was once called "expert systems."[1] [0] https://en.wikipedia.org/wiki/Gateway_(novel) https://en.wikipedia.org/wiki/Gateway_(novel) [1] https://en.wikipedia.org/wiki/Expert_system https://en.wikipedia.org/wiki/Expert_system
      - bonoboTP 1mo ago
        How was it foreseeable? Expert systems worked on quite different principles. Knowledge representation, expert interviews, ontologies etc.
        nobody9999 1mo ago
        >How was it foreseeable? I guess you had to be there. I certainly saw it coming. >Expert systems worked on quite different principles. Actually, not so much. The inference engine has changed, but otherwise it's same shit, different day.
        bonoboTP 1mo ago
        Expert systems didn't use neural nets or in fact any machine learning. They did interviews with experts and tried to do domain modeling, mapping the expertise to a formal symbolic logical description, ontologies, relations etc. Both the knowledge base and the inference engine changed. They both changed to a unified machine learning paradigm. It's like you throw away your axe's handle, then the head, and then you grab a chainsaw. It's "the same thing" to that same degree.
- morkalork 1mo ago
  Depends on what sci-fi you choose. We could definitely hook up an LLM to the controls for an elevator, swap out the panel of buttons with a voice interface and give it a depressed, snarky prompt and voilà we've got one of the AIs from Hitchhikers Guide to the Galaxy

by sumedh 1mo ago

habinero 1mo ago

We don't have sci-fi AI, and in any case "this thing is now possible, therefore [much more difficult thing] will be possible soon" is not a rational thought.

bonoboTP 1mo ago
> We don't have sci-fi AI Current frontier models are absolutely sci-fi AI by any measure that existed up until the late 2010s. If you showed a current frontier agentic AI, with bidirectional speech, tool use etc. to a person from 2010, they'd say it can't be real, there must be a person inside that mechanical turk.
- ozlikethewizard 1mo ago
  Sci-fi AI has always been characterised by learning on the fly, which these models do not do. I guess we've got pretty close to what sci fi has often called "VI", autonomous machines that can carry out pre determined tasks to a level that looks concious, but are incapable of learning further.
  - bonoboTP 1mo ago
    They learn again when the labs train the next version. The previous version often has an indispensable role in orchestrating that process, as a critic and from usage transcripts of the previous model that they have access to. And between two training runs the model can assemble a lot of md files, memory files, and load them in on-demand. I think this is enough to make it scifi-level, but I agree there should be some way to customize it at a deeper level and adapt to each person's context. But things are still moving too fast for that to make sense. If they fine tuned a specific LoRA for Claude just for your projects and workflows, it would get outdated quite fast and you'd have to retrain it at quite a cost for the new version.
- palmotea 1mo ago
  > Current frontier models are absolutely sci-fi AI by any measure that existed up until the late 2010s. Sci-fi is Lt. Cmdr. Data, HAL, Skynet, Culture Minds... > Current frontier models are absolutely sci-fi AI by any measure that existed up until the late 2010s. If you showed a current frontier agentic AI, with bidirectional speech, tool use etc. to a person from 2010, they'd say it can't be real, there must be a person inside that mechanical turk. It wasn't a natural language chat interface that makes stupid mistakes a noticeable fraction of the time and can execute command line programs.
  - nobody9999 1mo ago
    >Sci-fi is Lt. Cmdr. Data, HAL, Skynet, Culture Minds... The current state of things is closer (not close to by a long shot) to Fred Pohl's "AI" programs from his novel Gateway[0] than it is to the above examples. But even those are decades (at least) beyond what we have now. What we have now is the foreseeable (from at least 30 or 40 years ago) evolution of what was once called "expert systems."[1] [0] https://en.wikipedia.org/wiki/Gateway_(novel) https://en.wikipedia.org/wiki/Gateway_(novel) [1] https://en.wikipedia.org/wiki/Expert_system https://en.wikipedia.org/wiki/Expert_system
    - bonoboTP 1mo ago
      How was it foreseeable? Expert systems worked on quite different principles. Knowledge representation, expert interviews, ontologies etc.
      - nobody9999 1mo ago
        >How was it foreseeable? I guess you had to be there. I certainly saw it coming. >Expert systems worked on quite different principles. Actually, not so much. The inference engine has changed, but otherwise it's same shit, different day.
        bonoboTP 1mo ago
        Expert systems didn't use neural nets or in fact any machine learning. They did interviews with experts and tried to do domain modeling, mapping the expertise to a formal symbolic logical description, ontologies, relations etc. Both the knowledge base and the inference engine changed. They both changed to a unified machine learning paradigm. It's like you throw away your axe's handle, then the head, and then you grab a chainsaw. It's "the same thing" to that same degree.
morkalork 1mo ago
Depends on what sci-fi you choose. We could definitely hook up an LLM to the controls for an elevator, swap out the panel of buttons with a voice interface and give it a depressed, snarky prompt and voilà we've got one of the AIs from Hitchhikers Guide to the Galaxy