5 ms·

How do we know the mission got thrown out a window? The board still, after days of intense controversy, have yet to clearly explain how Altman was betraying the

How do we know the mission got thrown out a window? The board still, after days of intense controversy, have yet to clearly explain how Altman was betraying the mission. Did he ignore safety? Did he defund important research? Did he push forward on projects against direct objections from the board? If there’s a good reason, then let everybody know what that is. If there isn’t, then what was the point of all this?

paulddraper 3y ago
Exactly. All this "AI safety" stuff is at this point pure innuendo.
iteratethis 3y ago
Because the mission is visibly abandoned. There's nothing "open" about OpenAI. We may not know how the mission was abandoned but we know Sam was CEO, hence responsible.
- Wytwwww 3y ago
  What as "open" about it before that?
  - iteratethis 3y ago
    The first word in their company name.
- thrwmoz 3y ago
  There was never anything open about open ai. If there were I should have access to their training data, training infra setup and weights. The only thing that changed is the reason why the unwashed masses aren't allowed to see the secret sauce: from alignment to profit. A plague on both their houses.
  - marricks 3y ago
    They don't publish papers now, they actually published papers and code before. No doubt OpenAI was never a glass house... but it seems extremely disingenuous to say their behavior hasn't changed.
- wvenable 3y ago
  What was "open" before ChatGPT?
  - Zambyte 3y ago
    https://github.com/openai/gpt-2 https://github.com/openai/gpt-2
- nikcub 3y ago
  In terms of the LLM's, it was abandoned after GPT-2 when they realised the dangers of what was coming with GPT 3/3.5. Better to paywall access to and monitor it than open-source it and let it loose on the world. ie. the original mission was never viable long-term.
- gsuuon 3y ago
  Isn't Ilya even more against opening up models? OpenAI is more open in one way - it's easier to get API access (compared to say Anthropic)
shrimpx 3y ago
He went full-bore on commercialization, scale, and growth. He started to ignore the 'non-profit mission'. He forced out shoddy, underprovisioned product to be first to market. While talking about safety out one side of his mouth, he was pushing "move fast and break things", "build a moat and become a monopoly asap" typical profit-driven hypergrowth mindset on the other. Not to mention that he was aggressively fundraising for two companies that would be either be OpenAI's customer or sell products to OpenAI. If OpenAI wants commercial hypergrowth pushing out untested stuff as quickly as possible in typical SV style they should get Altman back. But that does seem to contradict their mission. Why are they even a nonprofit? They should just restructure into a full for-profit juggernaut and stop living in contradiction.
- jmull 3y ago
  chatgpt was under provisioned relative to demand, but demand was unprecedented, so it's not really fair to criticize much on that. (It would have been a much bigger blunder to, say, build out 10x the capacity before launch, without knowing there was a level of demand is known to support it.) Also, chatgpt's capabilities are what drove the huge demand, so I'm not sure how you can argue it is "shoddy".
  - shrimpx 3y ago
    Shipping broken product is a typical strategy to gain first mover advantage and try to build a moat. Even if it's mostly broken, if it's high value, people will do sign up and try to use it. Alternatively, you can restrict signups and do gradual rollout, smoothing out kinks in the product and increasing provisioning as you go. In 2016/17 Coinbase was totally broken. Constantly going offline, fucking up orders, taking 10 minutes to load the UI, UI full of bugs, etc. They could have restricted signups but they didn't want to. They wanted as many signups as possible, and decided to live with a busted product and "fix the airplane while it's taking off". This is all fine, you just need to know your identity. A company that keeps talking about safety, being careful what they build, being careful what they put out in the wild and its potential externalities, acting recklessly Coinbase-style does not fit the rhetoric. It's the exact opposite of it.
Zambyte 3y ago
> How do we know the mission got thrown out a window? When was the last time OpenAI openly released any AI?
- msikora 3y ago
  Whisper maybe?
- shwaj 3y ago
  Maybe whisper? https://en.m.wikipedia.org/wiki/Whisper_(speech_recognition_system) https://en.m.wikipedia.org/wiki/Whisper_(speech_recognition_...
- darknoon 3y ago
  Whisper v3, just a couple weeks ago https://huggingface.co/openai/whisper-large-v3 https://huggingface.co/openai/whisper-large-v3

paulddraper 3y ago

Exactly. All this "AI safety" stuff is at this point pure innuendo.

iteratethis 3y ago

Because the mission is visibly abandoned. There's nothing "open" about OpenAI. We may not know how the mission was abandoned but we know Sam was CEO, hence responsible.

Wytwwww 3y ago
What as "open" about it before that?
- iteratethis 3y ago
  The first word in their company name.
thrwmoz 3y ago
There was never anything open about open ai. If there were I should have access to their training data, training infra setup and weights. The only thing that changed is the reason why the unwashed masses aren't allowed to see the secret sauce: from alignment to profit. A plague on both their houses.
- marricks 3y ago
  They don't publish papers now, they actually published papers and code before. No doubt OpenAI was never a glass house... but it seems extremely disingenuous to say their behavior hasn't changed.
wvenable 3y ago
What was "open" before ChatGPT?
- Zambyte 3y ago
  https://github.com/openai/gpt-2 https://github.com/openai/gpt-2
nikcub 3y ago
In terms of the LLM's, it was abandoned after GPT-2 when they realised the dangers of what was coming with GPT 3/3.5. Better to paywall access to and monitor it than open-source it and let it loose on the world. ie. the original mission was never viable long-term.
gsuuon 3y ago
Isn't Ilya even more against opening up models? OpenAI is more open in one way - it's easier to get API access (compared to say Anthropic)

shrimpx 3y ago

He went full-bore on commercialization, scale, and growth. He started to ignore the 'non-profit mission'. He forced out shoddy, underprovisioned product to be first to market. While talking about safety out one side of his mouth, he was pushing "move fast and break things", "build a moat and become a monopoly asap" typical profit-driven hypergrowth mindset on the other. Not to mention that he was aggressively fundraising for two companies that would be either be OpenAI's customer or sell products to OpenAI. If OpenAI wants commercial hypergrowth pushing out untested stuff as quickly as possible in typical SV style they should get Altman back. But that does seem to contradict their mission. Why are they even a nonprofit? They should just restructure into a full for-profit juggernaut and stop living in contradiction.

jmull 3y ago
chatgpt was under provisioned relative to demand, but demand was unprecedented, so it's not really fair to criticize much on that. (It would have been a much bigger blunder to, say, build out 10x the capacity before launch, without knowing there was a level of demand is known to support it.) Also, chatgpt's capabilities are what drove the huge demand, so I'm not sure how you can argue it is "shoddy".
- shrimpx 3y ago
  Shipping broken product is a typical strategy to gain first mover advantage and try to build a moat. Even if it's mostly broken, if it's high value, people will do sign up and try to use it. Alternatively, you can restrict signups and do gradual rollout, smoothing out kinks in the product and increasing provisioning as you go. In 2016/17 Coinbase was totally broken. Constantly going offline, fucking up orders, taking 10 minutes to load the UI, UI full of bugs, etc. They could have restricted signups but they didn't want to. They wanted as many signups as possible, and decided to live with a busted product and "fix the airplane while it's taking off". This is all fine, you just need to know your identity. A company that keeps talking about safety, being careful what they build, being careful what they put out in the wild and its potential externalities, acting recklessly Coinbase-style does not fit the rhetoric. It's the exact opposite of it.

Zambyte 3y ago

> How do we know the mission got thrown out a window? When was the last time OpenAI openly released any AI?

msikora 3y ago
Whisper maybe?
shwaj 3y ago
Maybe whisper? https://en.m.wikipedia.org/wiki/Whisper_(speech_recognition_system) https://en.m.wikipedia.org/wiki/Whisper_(speech_recognition_...
darknoon 3y ago
Whisper v3, just a couple weeks ago https://huggingface.co/openai/whisper-large-v3 https://huggingface.co/openai/whisper-large-v3