Hacker News Reader

Comment by saliagato | Hacker News Reader

You can now [1] pay from $2 to $3 million to pretrain custom gpt-n model. This has gone unnoticed but seems really neat. Provided that a start-up has enough money spend on that, it would certainly give competitive advantage.

[1] https://openai.com/form/custom-models

Edit: forgot to put the link

View on HN

Well it won’t because they’ll use the model you paid for and take your customers.

Is it the same as avoiding AWS because they will take your software and run it themselves to steal your clients?

This hasn't happened often, but it has happened. Elastic search for example.

Also dynamo db.

This is misleading - intentionally using the incorrect definitions of the words in the parent post to construe a lie that plausibly addresses the concern when read by somebody unfamiliar with the situation.

AWS took an open source project (Elastic) and forked it. They did not take an AWS customer's code.

It’s more like running a PaaS product backed by AWS and then your customers realizing they can just use AWS directly, pay less, and have less complexity. And they have done this before for what it’s worth.

The assumption behind why you'd use the program is that you have access to a proprietary dataset of sufficient size to build a large model around (say, for example, call center transcripts). This almost certainly means OpenAI doesn't have access to train their other models off that data, and it almost certainly means your customers can't take the same data and go straight to OpenAI.

I'm assuming that the target customer of this is people whose moat is proprietary data. If their moat is a unique approach to building a model, then it would indeed be dangerous to engage OpenAI. But then I'd think OpenAI would be hesistent to engage as well.

How do you square this with OpenAI's assertion that they never use data from enterprise customers for their own training? Are you suggesting they're lying?

They don't have to be currently lying for this to be a valid concern.

Clauses in terms of service are routinely updated or removed.

> Clauses in terms of service are routinely updated or removed.

True, but that plays a bit differently in B2B land, because your customers also have legal teams and law firms on retainer.

No-one seems to have sued Unity over an even more egregious set of EULA changes - claiming that users retroactively owed money on games developed and published under totally different terms.

Afaik, they rolled the retroactive fee back, right?

And technically, that was a modification of the future EULA.

If you wanted to continue to use Unity, here was the pricing structure, which includes payments for previous installs.

You were welcome to walk away and refuse the new EULA.

Which is a big difference with historically collecting data, in violation of the then-EULA, and attempting to retroactively bless it via future EULA change.

OpenAI just slurped the entire internet to train their main model, and the world just looks on as they directly compete with and disrupt authors the globe over.

Whoever thinks they are not interested in your data and won't use any trick to get it, then double down on their classic "but your honor, it's not copyright theft, the algorithm learns just like an employee exposed to the data would", isn't paying attention.

This is exactly why I am personally intensely opposed to treating ML training as fair use. Practically speaking the argument justifies ignoring anyone or any group’s preference not to contribute to ML training, so it’s a massive loss of freedom to everyone else.

I agree with you. What come to my mind, is that GPT using private data to learn, if given back to (any) customer, you would have an indirect "open source everything".

Do you have proof or are you just throwing baseless accusations out there?

No. The downside is you spent a lot of money on a model and don’t own it.

I had multiple déjà vus, from the API golden days "build anything you want", social media telling you "build your communities with us", to app stores "we take care of the distribution"...it all works, until it doesn't...if they decide to change terms, pricing or make your product/service redundant because its 'strategic'.

It's their platform, their business, their rules.

Wow, this is directly going to affect my company in the near term. We had been trying to do it all internally but have found little success. Even at ~$3M it's going to be an attractive choice.

If your an OpenAI end customer, you'll be fine pre-training gpt-n models for your business; if you're an OpenAI middleman pre-training gpt-n models for other customers, what makes you think OpenAI won't eventually bypass you ? Lookup startups built around APIs and platforms, for every success, there's a graveyard filled due to APIs and platforms changing the rules.

Reply on news.ycombinator.com