Yeah the basics of prompting are still super valuable and a great starting point…
The more meta question in this thread is can prompts or prompting techniques be considered IP and therefore protectable. As a startup founder who might be considered a “wrapper” on top of OpenAI this is a very real question. I get asked by investors in every single pitch “what prevents OpenAI from just figuring out what you’ve figured out and doing it themselves?” Nothing… other than the fact they haven’t done it yet. They could figure out AGI tomorrow but my guess is they won’t.
We’ve filed patents on our distributed reasoning techniques so I (and my patent lawyers) obviously feel those techniques are protectable.
FWIW… I have granted patents for the search input box in a file open dialog and the facebook news feed among other things… I figure someone somewhere owes me a few million dollars but I’m not holding my breath:)
If steps in making food is IP why not steps in building logic? Code can be IP so can plain language. So it makes sense plain language code can be IPed…
Source code can not be patented. Algorithms can be patented.
You can copyright source code which gives us the reasons for various software licenses such as copyleft.
Correct… the patent attorney advising us represents the NYT in a high profile lawsuit related to LLMs and he said our algorithm is totally protectable and easily defendable.
The OP said it was OK. Also I am no longer a moderator as I stepped away and now enjoy a few extra hours a day. The only real thing I gave up not being a moderator is not being able to resolve flags and not being able to delete topics and post. The only real reason to need to delete a post is when the triage bot has a false positive and that post needs to be deleted so I now have to flag them. However I can move selected post to a new topic which is why I asked.
Time stamps, e note books , prototypes etc and trade secret lets you demonstrate it. Unique makes it defendable. Keeping a solid path of development helps.
@EricGT incredible conversation . I can’t talk to most folks about anything like AI lol. My wife is sick of science she’s been married to me for 26 years
I’m generally in the defensive camp when it comes to patents. When I worked for Microsoft we patented everything just so we couldn’t be easily sued for something we created that was similar to someone else’s creation.
I don’t think I’d ever want to use them offensively but if you’re a small startup in a sea of small startups having a solid patent portfolio can be the difference between getting acquired and getting crushed.
Interesting thing about Trade Secrets in the U.S. is that it is not a federal law. There are state laws for Trade Secrets and state laws are often so different from one state to the next it could drive you nuts.
When I checked a few months ago that was what I found. Now I find
That’s an interesting distinction… I wasn’t aware of that but it makes sense. With a patent you can disclose the underlying algorithm without voiding the patent you just have to file a pre-disclosure before it becomes common knowledge.
We could technically disclose our algorithm and still be protected but we’ve elected not to because we want to preserve our first movers advantage as long as we can.
It is the secret part. You don’t actually register it. It is like the secret sauce on a hamburger. Think of the browser wars they all had niche. Same now under the hood secrets in everything.
Exactly… and if we did disclose it there would be a research paper documenting it in full detail within a week or two. It’s really interesting how different the motivations are for academia versus private sector… we’re starting to see the labs wrestle with these conflicting tensions
Indeed, that is all that solid paper or E trail. I posted my work public for months on social media I can defend it well.
Time stamps.
Just my dumb FB page I have around a 1000 followers too. I also run social media fan groups for old school tabletop RPGs and old school nerd culture in general near 160k total members spanning over my junk… … I’m very public .
The biggest issue imo facing this technology is the misconception of it. Some see it as a “magic box”. Having people who talk about it with the folks who don’t use it or understand it, teaching is what’s needed. It is a fine line between protecting one’s own hard and earned work vs stagnation of said work. that’s why I try to talk to anyone that will let me talk at um
We need a version of Bill Nye the Science Guy. Cartoons maybe imo kids will get this tech near intuitively.
IMHO educating the public could lead to a more informed perspective, which indirectly supports IP protection by fostering a community that understands and respects the technology and its creators
Oh to add to original post. You could use a version of non disclosure.
Like contract law. @kieran4
Add to that… It’s in OpenAIs best interest to make it seem more magical then it is… it’s not until you get deep into it that you realize while it’s absolutely amazing it’s not magic and worse there are serious flaws which aren’t easily fixed…
On the prompt engineering side I will agree with Sam that with each turn of the crank on the model there whole classes of prompt engineering “work arounds” that are no longer needed. We will get to a world to where these models have gigantic context windows and are rock solid at following instructions. I disagree with Sam’s timeline for that because these things need a lot of improvement but we will get there.
What’s not going anywhere though is how you chain all of your prompts together to do useful things. That’s just engineering and not going away anytime soon.
