Independent researcher Andreas Kirsch cited OpenAI's failed Altman ouster.
The deep irony of the perception that Anthropic has the mandate of heaven on vibes for trust and governance is that they have not been tested, pushed to limits, and forced to adapt to anywhere near the level of OpenAI. Glass governance mechanisms waiting to shatter.
Many users question Anthropic's untested safety and governance reputation in the OpenAI researcher's critique, while some defend OpenAI's accountability as strengthened by past challenges.
OAI's governance and accountability have been through the fire over and over again. Almost surely, through the pain and challenge, made much stronger than people realize. People learn how to solve problems they actively have.
The deep irony of the perception that Anthropic has the mandate of heaven on vibes for trust and governance is that they have not been tested, pushed to limits, and forced to adapt to anywhere near the level of OpenAI. Glass governance mechanisms waiting to shatter.
I don't mean this as criticism or even to puff up my own side. I mean I can't believe people don't generally notice this. Systems that don't get challenged do not build a muscle of resilience and there should be a high burden of proof that when push comes to shove they will work.
OAI's governance and accountability have been through the fire over and over again. Almost surely, through the pain and challenge, made much stronger than people realize. People learn how to solve problems they actively have.
Okay but where is the evidence from those episodes that OpenAI's governance works or that it has improved and that OpenAI shows increasing transparency and accountability?
The board wasn't able to fire Sam despite cause and ample evidence, OpenAI has signed a contract with the Pentagon that gives it wide leniency and might endanger democracy while pretending otherwise multiple times, and while OpenAI publicly claims to be for AI regulation, it has funded anti regulation astroturf operations and started to minimize the risks of AGI/ASI.
Where are the positive outcomes?
The deep irony of the perception that Anthropic has the mandate of heaven on vibes for trust and governance is that they have not been tested, pushed to limits, and forced to adapt to anywhere near the level of OpenAI. Glass governance mechanisms waiting to shatter.
I feel your pain and frustration.
I “ model” safely at an extremity within OAI.
I can't push into apocalyptic scenarios.
You know this, Anthropic know this.
They've spent 12 months creating an orchestrated doom laden agenda.
Oprah, Bernie and the pope have probably never used whatsapp but now they have allowed platforms for doom.
The world should wake up to Anthropics behaviour as @DavidSacks said the problem might be… you( Anthropic).
@jachiam0 @lavish_lerven Agree with the overall point, but OpenAI is still basically a baby compared to companies with actual deep experience with public scrutiny and engagement with government.
@lavish_lerven For a start, compare boards of directors, level of public scrutiny for practices, and involvement from elected officials in mission and execution
@jachiam0 genuinely: what makes you imagine they haven't? surely the people out there truly skilled at testing the limits are equal opportunity exploiters, it's not function of market share
@jachiam0 Your tweets have changed quite a lot in their tone recently - how come?
@jachiam0 If it's the choice available, I'll take unbroken glass over broken glass
The deep irony of the perception that Anthropic has the mandate of heaven on vibes for trust and governance is that they have not been tested, pushed to limits, and forced to adapt to anywhere near the level of OpenAI. Glass governance mechanisms waiting to shatter.
@jachiam0 Could you please share in broad terms what kinds of governance/accountability fires OAI has faced and been tempered by that Ant hasn't?
@jachiam0 This morning I deeply iterated out of Samuel Johnson.
For 275 years history has been kind to him.
History will be kind to him for another 275 years.
Dario Amodei might want to ask himself “ why?”
@jachiam0 untested safety culture is the real variable here
the reputation got built in a world where the hard calls never actually arrived
@davidmanheim @lavish_lerven I roughly agree though I think the comparison to e.g. DeepMind should consider that companies too well-established may have conservatism to the pre-AGI regime and challenges adapting their governance to new exotic scifi bullshit that AGI is likely to cause.
@jachiam0 @lavish_lerven Agree with the overall point, but OpenAI is still basically a baby compared to companies with actual deep experience with public scrutiny and engagement with government.
Independent researcher Andreas Kirsch cited OpenAI's failed Altman ouster.
The deep irony of the perception that Anthropic has the mandate of heaven on vibes for trust and governance is that they have not been tested, pushed to limits, and forced to adapt to anywhere near the level of OpenAI. Glass governance mechanisms waiting to shatter.