4 ms·

Can someone please define "safety"? I keep hearing this but could you clarify what that means in practical terms? Is that why there's a "BasedGPT"?

metabagel 3y ago
Basically, it amounts to working around the issue that AI has no morals or code of ethics and is unconstrained from certain human limitations, such as processing speed and replication speed. It thinks different than we do, and these differences will only become more pronounced as capability grows by orders of magnitude.
- Obscurity4340 3y ago
  Neither do search engines really. Like why does ChatGPT have to be considered so different from that when you could literally do all the same prompts in the form of searches and many search engines do the same thing at the very top as a quasi-summary? All that is required is to "string them together"
- otabdeveloper4 3y ago
  AI doesn't think, it is just a fancy madlibs engine.
  - Obscurity4340 3y ago
    Isn't AI probabalistic, to what exent is thinking also probabalistic?
  - pixl97 3y ago
    Please take a break from posting this dumb crap and actually think about the issue for half a minute. This repeated crap is just a rehash of "Machines can never do X better than people so why worry", well, because shit gets shaken up when machines do X better, faster, and cheaper.
    - moritzwarhier 3y ago
      Would you mind making your point in civilized language? Apart from that, I think everyone has heard your argument as well as the one you are responding to very often by now.
    - otabdeveloper4 3y ago
      Generative AI, by definition, produces statisticially average output. Humans produce wildly different output, including lots of statistical outliers on both tails of the distribution. From an economic point of view, all the value creation happens precisely at the tails, where generative AI can't function.
      - reducesuffering 3y ago
        Is that why it’s acing all the standardized tests with 90th percentiles? Because it’s average? Should be 50% no?
        majikaja 3y ago
        Bottom 90% doesn't count as human in top brain communities like this
        otabdeveloper4 3y ago
        > Because it’s average? Yes. > Should be 50% no? No. It's giving the statistically average correct answer, which it already knows because it has been trained on answer books for these standardized tests.
        pixl97 3y ago
        LLMs are way above 50, and that's not even looking at ideas behind specialized networks that are focused on particular training. A world where 9 out of 10 people aren't ever going to catch up to the AI specialist submodule is going to have a lot of problems distributing wealth.
      - pixl97 3y ago
        Why would an adversarial network be bound to the average? From an economic point of view your ideas where a few people make a trillion dollars and the rest struggle to find a means to eat are not valuable, unless you're selling the implements of war and strife. You just end up with a crapsack planet where the rich tell the poor the socialism is bad.
SpicyLemonZest 3y ago
It’s meant literally. Everyone involved in this story, both Altman and the people who did the coup, agrees that AI is dangerous in the same way that pandemics and nuclear bombs are dangerous.
- emptysongglass 3y ago
  A lot of HN chatter (and I'm inclined to agree) believe a number of those people involved are pro-safety to defend their moat and promote regulatory capture. There's a very large contingent of ML researchers who think any idea of AI extinction risk is foolish because we don't have any evidence that intelligence equals compute. I've yet to see a single person give evidence that these two things are equal. What's more, the missing ingredient in any of these AGI extinction scenarios is desire (desire to act, desire to be, desire to love, to kill, etc.) and if you thought there was a paucity of evidence for intelligence = compute just wait till you see the evidence for transformers showing evidence of desire. There's none. Not a shred of evidence. As ever, it's other human beings that are our greatest extinction risk.
  - Obscurity4340 3y ago
    I fan absolutely believe this. I'm interested in what BigMentalHealth /s feels about LLMs that can be trained off a knowledgebase like CBT/DBT/IFS and other modalities that are basically defined and representable You can't insist on a Monopoly where much of the population can't access you (cost/coverage) and caveat that with if they can't get help they have to hurt/cause more destruction/die. Not crazy about the structure here but you get the meaning.
  - Obscurity4340 3y ago
    Honestly inclined to agree. Not to question people's ethics unduly, it sounds exactly like the same logic we use to "protect" drug users from drugs by destroying their lives in advance and inducting them into the criminal justice system as a manner of measured due course. Makes sense, yup yup yup. We know what's best for you. ChatGPT doesn't do anything you can't already get by searching Brave or Kagi. Just a little more human-centric given its chat formst
digbybk 3y ago
https://aisafety.info/ https://aisafety.info/
- Obscurity4340 3y ago
  Are they in any way serving as a bottleneck to the widespread access to LLMs/"AI" for use in a self-determined self-therapeutic context? If they are they need to fück off

metabagel 3y ago

Basically, it amounts to working around the issue that AI has no morals or code of ethics and is unconstrained from certain human limitations, such as processing speed and replication speed. It thinks different than we do, and these differences will only become more pronounced as capability grows by orders of magnitude.

Obscurity4340 3y ago
Neither do search engines really. Like why does ChatGPT have to be considered so different from that when you could literally do all the same prompts in the form of searches and many search engines do the same thing at the very top as a quasi-summary? All that is required is to "string them together"
otabdeveloper4 3y ago
AI doesn't think, it is just a fancy madlibs engine.
- Obscurity4340 3y ago
  Isn't AI probabalistic, to what exent is thinking also probabalistic?
- pixl97 3y ago
  Please take a break from posting this dumb crap and actually think about the issue for half a minute. This repeated crap is just a rehash of "Machines can never do X better than people so why worry", well, because shit gets shaken up when machines do X better, faster, and cheaper.
  - moritzwarhier 3y ago
    Would you mind making your point in civilized language? Apart from that, I think everyone has heard your argument as well as the one you are responding to very often by now.
  - otabdeveloper4 3y ago
    Generative AI, by definition, produces statisticially average output. Humans produce wildly different output, including lots of statistical outliers on both tails of the distribution. From an economic point of view, all the value creation happens precisely at the tails, where generative AI can't function.
    - reducesuffering 3y ago
      Is that why it’s acing all the standardized tests with 90th percentiles? Because it’s average? Should be 50% no?
      - majikaja 3y ago
        Bottom 90% doesn't count as human in top brain communities like this
      - otabdeveloper4 3y ago
        > Because it’s average? Yes. > Should be 50% no? No. It's giving the statistically average correct answer, which it already knows because it has been trained on answer books for these standardized tests.
      - pixl97 3y ago
        LLMs are way above 50, and that's not even looking at ideas behind specialized networks that are focused on particular training. A world where 9 out of 10 people aren't ever going to catch up to the AI specialist submodule is going to have a lot of problems distributing wealth.
    - pixl97 3y ago
      Why would an adversarial network be bound to the average? From an economic point of view your ideas where a few people make a trillion dollars and the rest struggle to find a means to eat are not valuable, unless you're selling the implements of war and strife. You just end up with a crapsack planet where the rich tell the poor the socialism is bad.

SpicyLemonZest 3y ago

It’s meant literally. Everyone involved in this story, both Altman and the people who did the coup, agrees that AI is dangerous in the same way that pandemics and nuclear bombs are dangerous.

emptysongglass 3y ago
A lot of HN chatter (and I'm inclined to agree) believe a number of those people involved are pro-safety to defend their moat and promote regulatory capture. There's a very large contingent of ML researchers who think any idea of AI extinction risk is foolish because we don't have any evidence that intelligence equals compute. I've yet to see a single person give evidence that these two things are equal. What's more, the missing ingredient in any of these AGI extinction scenarios is desire (desire to act, desire to be, desire to love, to kill, etc.) and if you thought there was a paucity of evidence for intelligence = compute just wait till you see the evidence for transformers showing evidence of desire. There's none. Not a shred of evidence. As ever, it's other human beings that are our greatest extinction risk.
- Obscurity4340 3y ago
  I fan absolutely believe this. I'm interested in what BigMentalHealth /s feels about LLMs that can be trained off a knowledgebase like CBT/DBT/IFS and other modalities that are basically defined and representable You can't insist on a Monopoly where much of the population can't access you (cost/coverage) and caveat that with if they can't get help they have to hurt/cause more destruction/die. Not crazy about the structure here but you get the meaning.
- Obscurity4340 3y ago
  Honestly inclined to agree. Not to question people's ethics unduly, it sounds exactly like the same logic we use to "protect" drug users from drugs by destroying their lives in advance and inducting them into the criminal justice system as a manner of measured due course. Makes sense, yup yup yup. We know what's best for you. ChatGPT doesn't do anything you can't already get by searching Brave or Kagi. Just a little more human-centric given its chat formst

digbybk 3y ago

https://aisafety.info/ https://aisafety.info/

Obscurity4340 3y ago
Are they in any way serving as a bottleneck to the widespread access to LLMs/"AI" for use in a self-determined self-therapeutic context? If they are they need to fück off