Minor note: Greg was first fired as chairman, and then subsequently resigned from the company. They were separate actions.

Source: https://twitter.com/gdb/status/1725736242137182594

And the board was ridiculous thinking they could demote him and have him stick around. That was either weirdly short-sighted or strategic theater. I kind of think it might have been the former.

They knew he was going to leave. It's likely a combination of the following:

1. They couldn't fire him as an employee (or felt it was an overreach of their mandate)

2. They wanted to signal a clear distinction that they lost faith in him as Chairman, while not losing faith in his work as an employee.

3. They felt like it would play better with the company if his ultimate departure was his decision rather than theirs.

They don’t have to pay him an exit fee.

I doubt they care about this. This move already signals they're not optimizing for financial outcomes, and the independent board members (3/4 involved in this decision) have no equity in OpenAI.

My take is, Sam and Greg are not the executives they want people to think they are. This was recognized, and they got upset because of this, and things shook out this way.

Over the past decade, I've never heard a single bad thing about Sam or Greg from anyone who has worked with them.

The board may know something nobody else does, but I think (given the current information) it's significantly more likely that they _are_ who they purport to be... it's just that the board wanted something different.

I haven't either, but have you seen many people share their experience of working with him? I haven't.

I've worked directly with Sam on a few projects, and was always impressed by how thoughtful and empathetic he was. I've worked with a number of smart, wealthy people, and Sam is the only one I'd say that about.

Thanks!

> 2. I hate OpenAI (or Sam Altman's?) apparently puritanical inclination to anything considered Not Safe For Work, especially for paid API usage. Why not allow people to build and sell virtual partner chat bots with explicit NSFW content?

I don't think that is either's fault, the US is just very puritan and a lot of it is because credit card companies and banks don't like it.

> it's easy to interpret this as purely in the self-interest of OpenAI shareholders

My guess is that this is probably what the board didn't like, Altman focused too much on profits in various ways.

But there's plenty of successful US-based sites that host both SFW and NSFW content: Reddit, Twitter, Tumblr (before Yahoo), DeviantArt, etc

Even it seems Patreon (which I've actually heard it described as an "NSFW launderer") -- itself directly deals with credit card and banks.

Reddit isn't what I would call successful in making money, so there is that.

An OpenAI which allowed NSFW content literally could not exist. It would be shut down in under a week. Maybe possible under a different regulatory regime (France?) but even then I doubt it... any model developed by a company and offered as a product will have some censorship which gets baked in.

Could you expand on why? There is a lot if NSFW content on Internet. What could be different, regulatory wise, this time?

Because current porn companies know the law, and toe the line if they want to stick around.

AI doesn’t. Image generators know what naked people look like, and they know what children look like. And they’re really good at mashing ideas together like that.

Text generators will happily compose any scenario for you.

I’m not here to argue ethics, but it’s easy to see why a nsfw open ai would immediately run into major problems.

I think I understand the problem you are pointing at. For example, children porn via AI? My question that is more epistemological and take aside ethics is if you can stop AI to produce those images when all the tools are available. I can compare this (not in ethics terms but in feasibility) to the US government push to control the information about how to make bombs or cryptography in the 90s. They couldn't but they can prosecute people who take advantage of this technologies to commit a crime. I don't know how this applies to people watching children porn that is produced by AI. Here I understand this is an ethics question that talks about the person consuming this and not the technology.

The profit focus being wrong seems so weird to me. It's an awfully expensive operation to run GPT-4 at scale and even now, it's rumored they are running the services at a loss. I understand the philosophical side, sure, but you can't just disregard all those massive GPU farms and staff tuning their models. AI is said to have created a new country in terms of energy use and OpenAI no doubt accounts for a large portion of that.

Certainly Microsoft's GPTv4 infrastructure is still eye-wateringly expensive:

> GitHub Copilot has reportedly been costing Microsoft up to $80 per user per month in some cases as the company struggles to make its AI assistant turn a profit.

> According to a Wall Street Journal report, the figures reportedly come from an unnamed individual familiar with the company, who noted that the Microsoft-owned platform was losing an average of $20 per user per month in the first few months of 2023.

https://www.techradar.com/pro/microsoft-is-reportedly-losing...

I'm hacking on some GPT-for-long-form-text stuff right now and it is _eye wateringly_ expensive once you start generating at anything close to "professional human" token outputs. $80 per month sounds already pretty optimized.

This article about copilot is BS. Nat Friedman refuted this on twitter and made it clear that copilot wasn't losing money.

> The profit focus being wrong seems so weird to me.

It is a non-profit after all, who owns a for-profit company. I'm just trying to imagine how the board sees it.

The web API based licensing scheme is dumb and bad. It’s including a buggy whip holder on a model T thinking. The license should be that they sell a license for use of the weights. They can also sell a SaaS that does a web API to use the weights. But the weights are the thing other businesses actually want and it’s controlling and obviously a monopoly play to not sell the weights. Other businesses have an obvious incentive to only work with companies that sell weights so as to prevent their being mere serf’s on someone else’s SaaS farm.

It's not about Ilya wanting to slow things down.

Ilya is the technical mastermind behind OAI. The technical breakthroughs needed for AGI are not there yet. Ilya, Yann, Demis and many others are aware of it.

An aggressive push for applied research and commercialisation means less resources for technical breakthroughs.

This is a tricky situation.

> An aggressive push for applied research and commercialisation means less resources for technical breakthroughs.

Successful commercialisation is exactly what allows more resources to flow in, genius.

Most of the times that "move fast, break thing" is more like "move fast, do stupid things".