5 ms·

OpenAI appears to have completely u-turned (I took out stronger language) on the principles it started with. It's become just another commercial operation, hoa

OpenAI appears to have completely u-turned (I took out stronger language) on the principles it started with. It's become just another commercial operation, hoarding it's secrets and profiting off them - which is fine for normal companies but of questionable morality here given it's original manifesto and long term aim of creating a sentient being. At the very least they should have had the decency to change the name when they pivoted from Open to Closed.

taf2 6y ago
Anyone own closedai.com maybe can get some decent money from them?
shajznnckfke 6y ago
Raise money claiming you’re going to prevent AI from taking over the world, use the money to help develop AI that will take over the world. Roko’s Basilisk approves.
- wussboy 6y ago
  We must not speak of Roko’s Basilisk.
  - pontifier 6y ago
    I think we should. It's a stupid idea. The simulated version of me would suffer, but not the actual me. If an AI wants to torture another AI in the future it's not very intelligent if you ask me. There's nothing to be gained because the past isn't going to be changed by that action, so it might as well not do it.
mensetmanusman 6y ago
You’re thinking on different time scales than they are. Currently, GTP-3 is a super weapon for nation states if they and their trolls gain access to it. It is reasonable to give it to researchers in industry and academia to help them during this transition, because like nuclear weaponry (GTP-3-like technology may eventually have more impact on humans than nuclear weapons), the information on how to build it is out there.
- pdonis 6y ago
  > GTP-3 is a super weapon for nation states if they and their trolls gain access to it. Why? I get that it could be used to generate false or misleading news articles, for example, that look like they were written by humans, but there are already umpteen false or misleading news articles out there so anyone who believes whatever they read is already pwned. How does GPT-3 increase the pwnage?
  - staticassertion 6y ago
    I think something that's become clear with propaganda is that it scales extremely well. If you could create propaganda 100x faster it's a meaningful capability, and it's hard to estimate how dangerous that could really be.
    - Mirioron 6y ago
      Has it? When it comes to nation states, do you really think that it would make a meaningful difference whether they hire 10000 people to create propaganda vs having a single AI algorithm? Yes, it would be a win in terms of cost, but creating propaganda is not a resource-limited endeavor for a large nation state.
      - nickpeterson 6y ago
        There is a difference in controlling the message. It’s a lot easier to keep nefarious actions secret if a few people and a cluster of computers is doing the work vs thousands of people.
        Mirioron 6y ago
        That's a good point.
      - cwillu 6y ago
        Automated production of propaganda has scaling properties that an army of writers do not. Consider a twitter bot that responded to conservative twitter posts with a links and quotes from a liberal thinktank blog, whose articles were generated and saved on the fly to look like rebuttals, complete with references to seemingly already existing/back-dated articles laying the groundwork, also generated on the fly and saved. It's a game changer. [edit: clarified back-dated]
prepend 6y ago
I was thinking the same thing. How is this open? NotOpenAI seems like a group with good intentions, then realized all the money that could be made and decided to keep it for themselves.
1024core 6y ago
> but of questionable morality here given it's original manifesto and long term aim of creating a sentient being. Setting out with such noble goals is the best way to recruit talent. Do you think Sutskever, Goodfellow, etc. would have joined OpenAI if they had started off trying to be just like any other company?
- p1esk 6y ago
  Do you think Sutskever, Goodfellow, etc. would have joined OpenAI if they had started off trying to be just like any other company? Which "any other company", FAANG? Both Sutskever and Goodfellow worked at Google before OpenAI.
  - rayuela 6y ago
    Yeah and they turned down some fat salaries to go work at OpenAI instead. They were some of the biggest names in the game when they turned down those Google offers.
    - p1esk 6y ago
      How do you know what they make at OpenAI?
    - m0zg 6y ago
      > Yeah and they turned down some fat salaries to go work at OpenAI instead I'm pretty sure OpenAI pays them even fatter salaries. I worked for a nonprofit at one point. People think that nonprofit automaticall means a pay cut, but they paid me about as much as Google did before, except it was all-cash. From the fiscal standpoint nonprofit merely means that all the revenues and donations get spent on programs and salaries, not that it pays a pittance. Organizationally, there are some interesting features to it, though: no stock, no owner, limits on control, etc. But financially nonprofits can pay very well indeed.
      - anandchat 6y ago
        Goodfellow currently works for Apple, not OpenAI.

taf2 6y ago

Anyone own closedai.com maybe can get some decent money from them?

shajznnckfke 6y ago

Raise money claiming you’re going to prevent AI from taking over the world, use the money to help develop AI that will take over the world. Roko’s Basilisk approves.

wussboy 6y ago
We must not speak of Roko’s Basilisk.
- pontifier 6y ago
  I think we should. It's a stupid idea. The simulated version of me would suffer, but not the actual me. If an AI wants to torture another AI in the future it's not very intelligent if you ask me. There's nothing to be gained because the past isn't going to be changed by that action, so it might as well not do it.

mensetmanusman 6y ago

You’re thinking on different time scales than they are. Currently, GTP-3 is a super weapon for nation states if they and their trolls gain access to it. It is reasonable to give it to researchers in industry and academia to help them during this transition, because like nuclear weaponry (GTP-3-like technology may eventually have more impact on humans than nuclear weapons), the information on how to build it is out there.

pdonis 6y ago
> GTP-3 is a super weapon for nation states if they and their trolls gain access to it. Why? I get that it could be used to generate false or misleading news articles, for example, that look like they were written by humans, but there are already umpteen false or misleading news articles out there so anyone who believes whatever they read is already pwned. How does GPT-3 increase the pwnage?
- staticassertion 6y ago
  I think something that's become clear with propaganda is that it scales extremely well. If you could create propaganda 100x faster it's a meaningful capability, and it's hard to estimate how dangerous that could really be.
  - Mirioron 6y ago
    Has it? When it comes to nation states, do you really think that it would make a meaningful difference whether they hire 10000 people to create propaganda vs having a single AI algorithm? Yes, it would be a win in terms of cost, but creating propaganda is not a resource-limited endeavor for a large nation state.
    - nickpeterson 6y ago
      There is a difference in controlling the message. It’s a lot easier to keep nefarious actions secret if a few people and a cluster of computers is doing the work vs thousands of people.
      - Mirioron 6y ago
        That's a good point.
    - cwillu 6y ago
      Automated production of propaganda has scaling properties that an army of writers do not. Consider a twitter bot that responded to conservative twitter posts with a links and quotes from a liberal thinktank blog, whose articles were generated and saved on the fly to look like rebuttals, complete with references to seemingly already existing/back-dated articles laying the groundwork, also generated on the fly and saved. It's a game changer. [edit: clarified back-dated]

prepend 6y ago

I was thinking the same thing. How is this open? NotOpenAI seems like a group with good intentions, then realized all the money that could be made and decided to keep it for themselves.

1024core 6y ago

> but of questionable morality here given it's original manifesto and long term aim of creating a sentient being. Setting out with such noble goals is the best way to recruit talent. Do you think Sutskever, Goodfellow, etc. would have joined OpenAI if they had started off trying to be just like any other company?

p1esk 6y ago
Do you think Sutskever, Goodfellow, etc. would have joined OpenAI if they had started off trying to be just like any other company? Which "any other company", FAANG? Both Sutskever and Goodfellow worked at Google before OpenAI.
- rayuela 6y ago
  Yeah and they turned down some fat salaries to go work at OpenAI instead. They were some of the biggest names in the game when they turned down those Google offers.
  - p1esk 6y ago
    How do you know what they make at OpenAI?
  - m0zg 6y ago
    > Yeah and they turned down some fat salaries to go work at OpenAI instead I'm pretty sure OpenAI pays them even fatter salaries. I worked for a nonprofit at one point. People think that nonprofit automaticall means a pay cut, but they paid me about as much as Google did before, except it was all-cash. From the fiscal standpoint nonprofit merely means that all the revenues and donations get spent on programs and salaries, not that it pays a pittance. Organizationally, there are some interesting features to it, though: no stock, no owner, limits on control, etc. But financially nonprofits can pay very well indeed.
    - anandchat 6y ago
      Goodfellow currently works for Apple, not OpenAI.