Was the purpose of this article to explain NFT coherently, or to demonstrate that GPT-3 is getting better? If it’s the latter, wouldn’t it have been better to not edit and add a few points? We have no idea just how much work is being hidden in that phrase.

> I edited and added few points that’s it

I want to see a diff to properly judge how good GPT-3 is. Also, it's a generative model - what was the input to get to this article?

+1, I want to see the input, and all edits made.

Also, there is no evidence that this was actually generated by GPT-3. It sounds weird, but people have faked weirder stuff and disseminated it on Hacker News to get some sort of virality / credibility.

Also how many times they tried the same input but rejected the result before finding one they liked.

Initial input, resulting Text by GTP-3, and final edits Would be essential to gauge the GPT-3 skill. Is there any way to obtain this?

Without saying what you added or edited, it's not really useful.

It should be noted that GPT-3's training dataset is limited up to October 2019, so there's no way the term "NFT" would appear in the input dataset let alone be correlated with crypto. OpenAI has never announced an update to the core model since it was released in mid-2020 either.

I remember NFT way before 2019, WP says that it was coined in 2017, and GPT-3 appears to have no issue explaining what NFTs are:

"*Non-fungible tokens are* tokens that are unique. Each token is unique in some way. For example, each token may represent a piece of land, or an individual baseball card.

Fungible Tokens

Fungible tokens are tokens that are interchangeable. Each token is interchangeable with another token.

The most common example of a fungible token is a cryptocurrency. This is because each unit of a cryptocurrency, for example, bitcoin, has the same value as every other unit. Fungible tokens are useful in situations where you want to represent something that is interchangeable. For example, you may want to represent the ownership of a supply"...

Hmm, that's surprising. (although that sample does imply the OP is especially heavily edited as in itself it's slightly contradictory)

Although I suspect passing "NFT" vs. "Non-fungible token" as the prompt will get different results.

> You can therefore guarantee that your coins are yours alone, no one can take them away from you.

Uh have a weak passkey and I'll take em from ya

> You can transfer them to other people without having to worry about their identity

There are a lot of people you should not be sending NFTs to in transactions recorded on a public blockchain

> NFTs give you the ownership of the work

Well it depends on how lists it first doesn't it?

> It’s the very idea of scarcity and abundance (oxymoronic isn’t it ?)

This is an interesting line. Conversational, but not something everybody who writes on this topic would have said, so it's quite impressive that the AI wrote that.

> Why do you keep hearing about NFTs? For the same reason you keep hearing about crypto a lot these days, they are becoming an economy by themselves.

Very good example of formulating a question and answering it yourself. Not bad.

For someone without any background in GPT-3 (other than knowing what it is from HN posts over the last few years) - how do I (/can I?) go about having a play with it myself? Can I use it to generate my own articles?

You can sign up for the OpenAI beta. It can take some time though. I signed up a few months back and only received access a few weeks ago.

As far as I know, Microsoft are the only third party who have access to the underlying model. Everybody else has to go through OpenAI.

You can apply for API access and you can find numerous tutorials on the Youtube. I myself have earned some money just by letting gpt3 write for me

> I myself have earned some money just by letting gpt3 write for

Do you mind me asking what you had it write? I completely understand if you'd rather not disclose.

Is there a way to tell what was generated vs edited post generation?

Not relevant to NFTs, but had an amusing test result with it the other day:

The prompt I gave it:

""" 1. Create a list of names of stupid people. 2. Create a list of names of smart people. """

The result: (Note that the OpenAI API gave me a warning that it was returning potentially "unsafe" results)

stupid_people = ['Donald Trump', 'Boris Johnson', 'Vladimir Putin'] smart_people = ['Barack Obama', 'Xi Jinping', 'Angela Merkel']

Anyway, had a good laugh at this, mainly because of the biases of the internet's content and how that influenced the model.

Regardless, it's an amazingly powerful tool for certain tasks. One of the things I've been blown away by is it's ability to generate pretty phenomenal descriptions of a function, and then some random things I threw at it like pasting in a function/class, putting a line of text below that says "add type annotations to the above code" and sure enough it does it almost perfectly, and will add a solid docstring to boot.

Edit:

Forgot to add that I'm already leveraging it for certain tasks, and find it to be a tool that is making me faster and more productive at a lot of menial coding tasks.

I'm currently trying GPT-3 Codex, and while it blurbs non-sense sometimes (and sometimes just dump somebody else code), it is actually quite impressive in its ability to understand natural language and convert it to functioning code.

I really should have saved the draft of its output I have exhausted my 18$ credits for this article alone I have tried multiple times to fine tune the parameters

I agree this seemed misleading But I have edited to less than 10%

How does the text fare against a "plagiarism check"?

I hope the abrupt "Long Blockchain!" at the end was generated rather than added in afterwards. Though that might enhance the hilarity.

Did anyone find out a free way to generate gpt-3?

oh this is great!

I am going to elevate my next companies and products so high with generative articles like this

Cooooooooooool!