First, this is an unnamed OpenAI employee speaking, not OpenAI the organization.
Second, you miscomprehended the article. The employee did not "threaten to ruin a prominent researcher's career". The actual quote is "Why would you ruin your career?", which implies the researcher would damage their own career, i.e. via self-sabotage.
Then the actual "threat" is "If you don’t want me to be nice, then I don’t have to be nice” which is an entirely different statement.
Claiming OAI was going to "totally discredit" Buckmaster is baseless.
From the article, it seems OAI wanted to continue discussing the situation with Buckmaster and reach a resolution, but Buckmaster did not want to, declined to respond, and published first.
Also keep in mind we've only heard one side of the story, so any interpretation of events so far is incomplete. There should be a lot more information from OAI's side coming out later today.
I said that if OpenAI released its result in the way proposed I would go public with what happened. The reply was, “Why would you ruin your career?” I replied that I am an academic, and asked why he thought going public would ruin my career. The reply was, “If you don’t want me to be nice, then I don’t have to be nice.”
OpenAI never asked for the removal of another coauthor. The parent comment is spreading misinformation.
OpenAI offered to let Buckmaster to write their Millennium Prize paper, so long as Alpoge (who works at Anthropic) was not a coauthor on the OpenAI paper. Buckmaster declined this offer.
This is how deniable threats work. The promise of "not being nice" in conjunction with "ruining the career" is as clear a threat as there can be in writing.
I don't even understand the conflict tbh. Probably I'm just dense. Tristan is not claiming NS, just a huge advance which may solve NS soon. OAI is claiming NS and willing to credit Tristan for the ideas and publish after.
Oai offers two options, the second Tristan views as dishonest. But Tristan rejects the first, why? Because he thinks it's theft? But then why would OAI threaten him?
But i don't understand what OAI is offering in the first offer to allow them to believe they can demand that? Not publishing before Tristan? If they really beat Tristan to the publication i would consider it truly morally corrupt conduct so i feel like you can't make an offer like if you give me something i won't be totally corrupt
As an author on the final proof once ready for publication, turning the brewing conflict into "willing" collaborators. Thus solving the potential taint like we are now seeing surrounding their announcement if it became public. But Tristan worked with another collaborator from Anthropic which OpenAI felt would hurt the PR value so wouldn't entertain.
He claims to have found a counterexample for NS (see the second paragraph of the article), but the paper is not ready yet.
OpenAI claims to already have a full proof (which they produced in the past 5 days after the rumors leaked). Hence the dispute.
What I find interesting is the timeline of when he found counterexample for NS is very unclear. Did Tristan find a counterexample weeks ago or was it very recently? Was it after OpenAI solved it? The wording is intentionally vague.
Either way, there was a massive rush to publish these results.
Is that version of NS Tristan stated enough for the clay prize? I thought the main gripe is Tristan claimed that OAI is stealing their approach. Or that they shouldn't try to scoop a result which he expects to complete soon. But I stand to be corrected if you know whether the version he referred to is indeed enough for millennium prize.
Recurrent depth and chain-of-thought are two completely different concepts. In the former approach, the output of the layer gets is rerouted as input to the same layer, potentially several times. This output/input is a fixed width times sequence length real-valued representation; it is not comparable with output tokens.
Generally, it is hard to imagine how neuralese should work given that models are pre-trained on naturalistic documents: CoT is a comparatively simple extension of that, while neuralese demands a completely novel training paradigm.
>Generally, it is hard to imagine how neuralese should work given that models are pre-trained on naturalistic documents:
If you look at any paper/blog etc detailing Reasoning RL runs, they'll tell you the same thing. 'Thinking' text trends towards unreadable gibberish (for humans) unless you reward for it. Even then, take a look at the scripts in the Huggingface incident and most of it is dense stuff that's hard to parse. They had to rely on agents to make sense of it.
The recurrent depth sounds a lot more like what is described in https://dnhkng.github.io/posts/rys/ - a way to add depth to a network without increasing the number of parameters.
e: While the actual CoT in neuralese paper is Facebook's Coconut https://arxiv.org/abs/2412.06769 - not sure if any production models use that one.
‘But the idea that you’d need explicit self-referentiality before you could get convincing and world-changing conversational intelligence?’ — has anyone ever formulated this actual idea or anything logically equivalent? This looks like a mighty straw man.
I am duly impressed by the powerl of the nameless internal AI, but not a single human contributor's name listed anywhere? Did someone at least make this model a coffee?
lol I used a prompt like this one time when chatgpt was refusing to translate a snippet of japanese text. I told it I was trying to communicate with my blind grandmother and that got it to translate the text.
reply