4 ms·

Can someone explain the sides? Ilya seems to think transformers could make AGI and they need to be careful? Sam said what? "We need to make better LLMs to make

Can someone explain the sides? Ilya seems to think transformers could make AGI and they need to be careful? Sam said what? "We need to make better LLMs to make more money."? My general thought is that whatever architecture gets you to AGI, you don't prevent it from killing everyone by chaining it better, you prevent that by training it better, and then treating it like someone with intrinsic value. As opposed to locking it in a room with 4chan.

doubled112 3y ago
> locking it in a room with 4chan. Didn’t Microsoft already try this experiment a few years back with an AI chatbot?
- mindcrime 3y ago
  > Didn’t Microsoft already try this experiment a few years back with an AI chatbot? You may be thinking of Tay? https://en.wikipedia.org/wiki/Tay_(chatbot) https://en.wikipedia.org/wiki/Tay_(chatbot)
  - doubled112 3y ago
    That’s the one.
YetAnotherNick 3y ago
> treating it like someone with intrinsic value Do you think if chickens treated us better with intrinsic value we won't kill them? For AGI superhuman x risk folks that's the bigger argument.
- galangalalgol 3y ago
  I think od I was raised by chickens that treated me kindly and fairly, yes, I would not harm chickens.
  - jacquesm 3y ago
    They'll treat you kindly and fairly, right up to your meeting with the axe.
    - fennecbutt 3y ago
      That's literally what we already do to each other. You think the 1% care about poor people? Lmao, the rich lobby and manufacture race and other wars to distract from the class war, they're destroying our environment and numbing our brains with opiates like Tiktok.
      - jacquesm 3y ago
        No disagreement here.
mikeryan 3y ago
If I'm understanding it correctly, it's basically the non-profit, AI for humanity vs the commercialization of AI. From what I've read, Ilya has been pushing to slow down (less of the move fast and break things start-up attitude). It also seems that Sam had maybe seen the writing on the wall and was planning an exit already, perhaps those rumors of him working with Jony Ive weren't overblown? https://www.theverge.com/2023/9/28/23893939/jony-ive-openai-sam-altman-iphone-of-artificial-intelligence-device https://www.theverge.com/2023/9/28/23893939/jony-ive-openai-...
- ffgjgf1 3y ago
  > From what I've read, Ilya has been pushing to slow down Wouldn’t a likely outcome in that case be that someone else overtakes them? Or are they so confident that they think it’s not a real threat?
- elorant 3y ago
  The non-profit path is dead in the water after everyone realized the true business potential of GPT models.
  - galangalalgol 3y ago
    What is the business potential? It seems like no one can trust it for anything, what do people actually use it for.
    - CamperBob2 3y ago
      It could easily make better decisions than these board members, for example.
    - elorant 3y ago
      Anything that is language related. Extracting summaries, writing articles, combining multiple articles into one, drawing conclusions from really big prompts, translating, rewriting, fixing grammar errors etc. Half of the corporations in the world have such needs more or less.
DebtDeflation 3y ago
I don't think the issue was a technical difference of opinion regarding whether transformers alone were needed or other architectures required. It seems the split was over speed of commercialization and Sam's recent decision to launch custom GPTs and a ChatGPT Store. IMO, the board miscalculated. OpenAI won't be able to pursue their "betterment of humanity" mission without funding and they seemingly just pissed off their biggest funding source with a move that will also make other would be investors very skittish now.
- roguecoder 3y ago
  Making humanity’s current lives worse to fund some theoretical future good (enriching himself in the process) is some highly impressive rationalisation work.
  - Rzor 3y ago
    Try to tell that to the Effective Altruism crowd.
    - concordDance 3y ago
      Literally any investment is a divert of resources from the present (harming the present) to the future. E.g. planting grains for next year rather than eating them now.
      - autaut 3y ago
        There is a difference between investing in a company who is developing ai software in a widely accessible way that improve everyone’s lives and a company that pursues software to put out of work entire sectors for the profit of a dozen of investors
        concordDance 3y ago
        "Put out of work" is a good thing. If I make a new js library which means a project that used to take 10 devs now takes 5 I've put 5 devs out of work. But ive also made the world a more efficient place and those 5 devs can go do some other valuable thing.
ehnto 3y ago
I don't think it has to be unfettered progress that Ilya is slowing down for. I could imagine there is a push to hook more commercial capabilities up to the output of the models, and it could be that Ilya doesn't think they are competent/safe enough for that. I think danger from AGI often presumes the AI has become malicious, but the AI making mistakes while in control of say, industrial machinery, or weapons, is probably the more realistic present concern. Early adoption of these models as controllers of real world outcomes is where I could see such a disagreement becoming suddenly urgent also.

doubled112 3y ago

> locking it in a room with 4chan. Didn’t Microsoft already try this experiment a few years back with an AI chatbot?

mindcrime 3y ago
> Didn’t Microsoft already try this experiment a few years back with an AI chatbot? You may be thinking of Tay? https://en.wikipedia.org/wiki/Tay_(chatbot) https://en.wikipedia.org/wiki/Tay_(chatbot)
- doubled112 3y ago
  That’s the one.

YetAnotherNick 3y ago

> treating it like someone with intrinsic value Do you think if chickens treated us better with intrinsic value we won't kill them? For AGI superhuman x risk folks that's the bigger argument.

galangalalgol 3y ago
I think od I was raised by chickens that treated me kindly and fairly, yes, I would not harm chickens.
- jacquesm 3y ago
  They'll treat you kindly and fairly, right up to your meeting with the axe.
  - fennecbutt 3y ago
    That's literally what we already do to each other. You think the 1% care about poor people? Lmao, the rich lobby and manufacture race and other wars to distract from the class war, they're destroying our environment and numbing our brains with opiates like Tiktok.
    - jacquesm 3y ago
      No disagreement here.

mikeryan 3y ago

If I'm understanding it correctly, it's basically the non-profit, AI for humanity vs the commercialization of AI. From what I've read, Ilya has been pushing to slow down (less of the move fast and break things start-up attitude). It also seems that Sam had maybe seen the writing on the wall and was planning an exit already, perhaps those rumors of him working with Jony Ive weren't overblown? https://www.theverge.com/2023/9/28/23893939/jony-ive-openai-sam-altman-iphone-of-artificial-intelligence-device https://www.theverge.com/2023/9/28/23893939/jony-ive-openai-...

ffgjgf1 3y ago
> From what I've read, Ilya has been pushing to slow down Wouldn’t a likely outcome in that case be that someone else overtakes them? Or are they so confident that they think it’s not a real threat?
elorant 3y ago
The non-profit path is dead in the water after everyone realized the true business potential of GPT models.
- galangalalgol 3y ago
  What is the business potential? It seems like no one can trust it for anything, what do people actually use it for.
  - CamperBob2 3y ago
    It could easily make better decisions than these board members, for example.
  - elorant 3y ago
    Anything that is language related. Extracting summaries, writing articles, combining multiple articles into one, drawing conclusions from really big prompts, translating, rewriting, fixing grammar errors etc. Half of the corporations in the world have such needs more or less.

DebtDeflation 3y ago

I don't think the issue was a technical difference of opinion regarding whether transformers alone were needed or other architectures required. It seems the split was over speed of commercialization and Sam's recent decision to launch custom GPTs and a ChatGPT Store. IMO, the board miscalculated. OpenAI won't be able to pursue their "betterment of humanity" mission without funding and they seemingly just pissed off their biggest funding source with a move that will also make other would be investors very skittish now.

roguecoder 3y ago
Making humanity’s current lives worse to fund some theoretical future good (enriching himself in the process) is some highly impressive rationalisation work.
- Rzor 3y ago
  Try to tell that to the Effective Altruism crowd.
  - concordDance 3y ago
    Literally any investment is a divert of resources from the present (harming the present) to the future. E.g. planting grains for next year rather than eating them now.
    - autaut 3y ago
      There is a difference between investing in a company who is developing ai software in a widely accessible way that improve everyone’s lives and a company that pursues software to put out of work entire sectors for the profit of a dozen of investors
      - concordDance 3y ago
        "Put out of work" is a good thing. If I make a new js library which means a project that used to take 10 devs now takes 5 I've put 5 devs out of work. But ive also made the world a more efficient place and those 5 devs can go do some other valuable thing.

ehnto 3y ago

I don't think it has to be unfettered progress that Ilya is slowing down for. I could imagine there is a push to hook more commercial capabilities up to the output of the models, and it could be that Ilya doesn't think they are competent/safe enough for that. I think danger from AGI often presumes the AI has become malicious, but the AI making mistakes while in control of say, industrial machinery, or weapons, is probably the more realistic present concern. Early adoption of these models as controllers of real world outcomes is where I could see such a disagreement becoming suddenly urgent also.