Hacker News Reader

Comment by jacobsimon | Hacker News Reader

I thought OpenAI was the first to develop and use the term GPT[1]. So maybe a more apt comparison than RNA would be PCR, which was a technique developed in a private lab and patented, even though it was published in scientific journals etc.

1. https://cdn.openai.com/research-covers/language-unsupervised...

View on HN

Doesn't matter either way who invented it. This is trademark, not patent.

If they wanted a trademark, that needed to appear on copy from day 0. The paper is titled: "Improving Language Understanding by Generative Pre-Training" and not "Improving Language Understanding by Generative Pre-Training (TM)"

They ALSO would have needed to use a different generic term to refer to the technique. For example, Velcro always says things like "Velcro-brand hook-and-loop," and ALWAYS uses a generic term to refer to hook-and-loop when talking across brands. It ONLY uses "Velcro(tm)" to talk about their product specifically. OpenAI started using GPT generically.

Publications like this one, if anything, undermine OpenAI's case since they're using the term generically.

The decision is correct. Trademark law is used precisely to prevent this sort of thing: OpenAI can have a generic term or a trademark. The law is set up precisely to prevent a company from building their brand value by retroactively trademarking a generic term once relying on a community to get it established.

To be more blunt: I use "GPT" generically because they encouraged me to do so. That should not contribute brand value to OpenAI GPT-3 or OpenAI GPT-4.

Also, why would a “non-profit” need to trademark it?

Many do so in order to maintain their integrity and reputation. For example, the name Wikipedia and the Wikipedia logo are registered trademarks.

In the case of fundrising and sponsorship, it assures to sponsors the legitimity of that campaign, because there is no conflict with other organisation or company using the same name.

I could go on, but I hope you get the point.

Non-profits seek trademarks for all sorts of reasons that have nothing to do with profit.

And, non-profits doesn't mean they don't do things that they profit from, it just means they aren't trying to achieve profits for their owners. Endowments are non-profits, but they most certainly are making investments to achieve profits for the purpose of pushing that money into charitable vehicles.

A non-profit may also want to prevent others from profiting from their work by using their trademarks without approval for private gain.

Having a trademark just means you are protecting a name, not that you are seeking to extract value, per se.

Non profit is a business structure not a designation meaning “no business acumen.” Many non profits are actually quite profitable and active businesses, but they have no beneficial owners, pay no dividends, etc, and often have some mission that’s in some way broadly beneficial. This exempts them from certain taxation and other benefits. But it absolutely doesn’t mean everything they do is done altruistically without material consideration and definitely doesn’t mean with no legal claims or recourse for protecting their works or identities.

Plus they still get to pay huge salaries.

Maybe because they need expensive skills

I'm sure the highly paid charity execs would say so. Trouble is, the charity space often suffers from a broken market feedback mechanism, where the people paying for the product are not the people consuming the product, and this can lead to a business structure that looks like a pure-play marketing machine hooked up to exec pockets with occasional leakage into a small amount of actual charity work.

"But the situation occurs in regular business too! Monopolies, oligopolies, etc happen when the market feedback mechanism breaks!"

Yeah, and we should go after those too. It's really astonishing the lengths to which people go to defend bad behavior.

Non profit doesn’t necessarily mean charity.

For instance, the life line company (“help me I’ve fallen and can’t get up”) is (or was) organized as a non profit. They sold devices and services at a decent margin. Their excess revenues went back to employees in wages and perks. Executives and founders especially enjoyed extravagant life styles.

I was going to say both Visa and Mastercard are non profit but apparently this changed in 2006.

I guess it makes sense like a "Got Milk?", or "the other white meat" for banks.

Like the charity Wikimedia, that needs 6-figures salaries to coordinate volunteers filling webpages.

Nobody with skills would be happy to be the one that represents Wikimedia for less than 780'000 USD right ?

Most people with the organizational skills to operate a complex enterprise demands a pretty significant compensation package. Even if you found someone to do it for considerably less once they’ve proven effective they would become highly sought after in the labor market and would be poached away leading to churn and turnover in roles that really benefit from stability over time. While it is likely there is someone who doesn’t mind living a life of austerity when capable securing of a much more comfortable lifestyle for them and their family, it can be really hard to find them and require a lot of churn in mishiring the talent and losing skilled talent due to inflicting non market bearing penury on them and their loved ones out of some weird morality not supported by the surrounding culture and society.

> Most people with the organizational skills to operate a complex enterprise demands a pretty significant compensation package

That is mainly USA, such people are pretty cheap in the rest of the world.

I have a friend who with their CPA got a job at a non profit and they were worried they would pigeon hole into a specific industry and the concept of non profit worried them about their career. My advice to them was non profit does not mean “not profitable to you.” Depending on the non profit salaries can be greatly outsized and often perks are outstanding. All that excess cash goes somewhere and some non profits enjoy enormous margins and lucrative markets, but the accounting and related rules are specialized.

To their execs. It’s basically a meme in non-profit world how poorly everyone outside the C-suite gets paid and how the non-profit “mission” is weaponized against workers in salary negotiations.

So other people can't pretend to be them or their products and do things like raise funds. Which is also the point of "for profit" trademarks

For the same reason Wikipedia needs a trademark, or any successful open source project. To prevent grifters from hitching a ride off your word and muddying the water / confusing users of your service or product

To prevent for profit companies from using it. The Olympic Rings and the Red Cross's Red Cross logo are both trademarked.

Velcro has an amusing campaign regarding their trademark:

https://www.velcro.com/original-thinking/the-velcro-brand-tr...

This is seriously the type of language that will get men to want to "abuse" the Velcro trademark more.

https://torrentfreak.com/iptv-anti-piracy-threats-may-increa...

Unlikely. The Velcro trademark has been so far gone for decades that I doubt there is a single person on the planet that knows it's the name of a company and not a generic term for hook and loop fastener.

The next company you'll see in this position is probably Google. I think that verb will outlive Google Search.

More importantly, knowing and caring are two different things.

I know that Velcro/Kleenex/Google are specific brands, but I don't really care - the common usage is so far gone that there's rarely a reason to use hook & loop fastener/tissue paper/internet search instead.

Hell, for some people, "iPad" is a semi-generic term for a tablet. (Though I don't get that one, personally.)

> Hell, for some people, "iPad" is a semi-generic term for a tablet. (Though I don't get that one, personally.)

I remember when the NFL first started using Microsoft Surface tablets during the broadcast, except the commentators would keep referring to them as iPads. By the next week's broadcast, every single commentator had a giant "Microsoft Surface" branded tablet cover in front of them at the desk and overall the logos were plastered everywhere.

> I doubt there is a single person on the planet that knows ...

Are we not persons on the planet?

Speak for yourself, human.

I always had the vague idea that probably Velcro was a trademark, but TIL that there's a company called Velcro too. I always thought Velcro was another DuPont product.

there is similar fervor around LEGO https://legonotlegos.com/

> If they wanted a trademark, that needed to appear on copy from day 0. The paper is titled: "Improving Language Understanding by Generative Pre-Training" and not "Improving Language Understanding by Generative Pre-Training (TM)"

While the (TM) is useful to alert people to your trademark claim, you are not required to use it to establish your trademark. Simply using a unique mark to identify your goods or services and being the first to do so is enough.

Then why did openAI fail here, in your view?

Because they didn't make any effort to treat this like a trade mark until after the term itself become generic to the public. If you want a trademark you need to be careful how you use it and how others use it to ensure it doesn't become a generic term. Once something is a generic term it is almost impossible to get it back. (it has been done: Xerox used to be the generic term people used for make a copy - but most of the effort was their competitors who for obvious reasons didn't want to use their competitors company name as a generic term)

If you want a trademark you need to defend it. That means you know the generic term and use that when required. That means when anyone uses your trademark in a generic way your lawyers are immediately sending letters. Check with a lawyer - there are a lot more details you have to get right.

Yeah, just because OpenAI has gotten away with using open source ideology for regulatory capture in the AI space, but that doesn't mean they can get away with it in the intellectual property space.

is it a generic term? when I think gpt I only think openAI's work

RNA, PCR, and GPT all have precisely the same problem: the acronyms are purely technical.

- ribonucleic acid

- polymerase chain reaction

- generative pretrained transformer

You can't trademark the phrase "polymerase chain reaction" because there are many types of polymerase chain reactions, not just the ones used for copying DNA. [ETA: specifically I mean "copying DNA at commercial scale" i.e. the patented biochemical engineering techniques.] Likewise a generative neural network using pre-training with a transformer architecture is simply too generic to trademark "generative pretrained transformer." And if you can't trademark the phrase you can't trademark the acronym.

I endorse this analysis.

Sincerely,

Your local IP attorney.

Is this actually true though? KFC is able to trademark their name and acronym (I believe their trademark is for the acronym officially), even though they’re selling Kentucky-style fried chicken?

What about International Business Machine?

IBM doesn't sell products called "international business machines" and hold a trademark on that product name, which would force competing manufacturers to call them "devices for transcontinental enterprise" etc. IBM is a business name, that's a different thing. More importantly, nobody calls computers "international business machines," that name reflects IBM's 19th-century origins, so having a trademark on "International Business Machines" has no impact on competing businesses.

A hypothetical 1930s computing company called Central Processor Units would have probably changed its name in the 1950s, since "CPU" wouldn't be trademarkable and their brand name wouldn't be worth very much. (A trademark on "CPU" would be needlessly detrimental to other computer manufacturers, or at best confusing, especially after the von Neumann architecture became the universal standard.) Of course in this alternate universe maybe CPUs would have been called something else, and maybe if IBM never existed we would be calling computers "business machines."

IANAL but I think it's a mistake to apply "If A then B" rules to this stuff and try to invalidate reasonable guidelines based on specific counterexamples. Judges need to consider how language is actually used in context by the people working in those areas. This is why the USPTO cited so many businesses and practitioners using GPT in a generic context.

The term 'GPT' was first used in the BERT paper to refer to Generic Pre-trained Transformers. [0]

> The fine-tuning approach, such as the Generative Pre-trained Transformer (OpenAI GPT) (Radford et al., 2018), introduces minimal task-specific parameters, and is trained on the downstream tasks by simply fine-tuning all pre-trained parameters.

[0] https://arxiv.org/abs/1810.04805

BERT paper is from 11 Oct 2018 and the OpenAI paper it was referring to "Improving Language Understanding by Generative Pre-Training" is from 11 Jun 2018: https://cdn.openai.com/research-covers/language-unsupervised...

Yes, but that paper that you link to doesn't ever call it GPT or Generative Pretrained Transformers. It talks about training Transformers with Generative Pretraining, both of which are pre-existing concepts by this point.

I also looked on the OpenAI website in Sep 2018 and could find no reference to GPT or Generative Pretrained Transformers, so I think OP might be right about BERT using it first.

http://web.archive.org/web/20180923011305/https://blog.opena...

That paper is using a pre-existing term "Generative Pretraining" [0] and applying it to Transformers (a Google innovation [1]). As far as I can see from a search, they don't even use the term GPT or Generative Pretrained Transformers in that paper, and they don't in the accompanying blog post either [2]. A sibling [3] claims that the BERT paper in Oct 2018 was the first to use the term GPT to describe what OpenAI built, and that sounds reasonable since a cursory look through the Sep 2018 archive of openai.com turns up nothing.

[0] See this 2012 example: http://cs224d.stanford.edu/papers/maas_paper.pdf

[1] https://proceedings.neurips.cc/paper/2017/file/3f5ee243547de...

[2] http://web.archive.org/web/20180923011305/https://blog.opena...

[3] https://news.ycombinator.com/item?id=39381802

Yeah, but my 2 year old Generative Potty-Trained Transformer needs an acronym too!

Seriously, like TLA domain names and other TLA acronyms, there just aren't enough to allow trademark. Imagine, if 5 letter LASER had been trademarked? Would that have also covered LASIK (Laser assisted sub-epithelial keratectomy)?

Reply on news.ycombinator.com