ChatGPT AI is a huge breakthrough, a Google killer

[redshift]

I was quite skeptical about it generating code, but then I asked for a simple thing I was trying to implement. After spending about 15-20 annoying mintues on google and reading various documentation files (including the one it linked below, lol) I ended up doing something very similar (thought not exactly the same). Quite remarkable.


It answered a few follow up questions as well.

 

Dislike ads? Remove them and support the forum: Subscribe to Fastlane Insiders.

[polaroid22]

Yes tested this thing this morning, it is absolutely insane. Too bad its only free for testing right now. Not sure how much it will cost on daily use in the future.

 

[SharpeningBlade]

Yes tested this thing this morning, it is absolutely insane. Too bad its only free for testing right now. Not sure how much it will cost on daily use in the future.

If we compare the current pricing on their GPT-3 language models, it wouldn't be too bad if you are finding real value in it. For a mid-tier, 2M tokens per month using their Davinci model you are looking at about $40 per month which is roughly 1.5 million words per month, but input queries do count towards total token utilization.

Source: Pricing

 

[MJ DeMarco]

From the founder...

https://twitter.com/i/web/status/1601731295792414720

View: https://twitter.com/sama/status/1601731295792414720?s=20&t=20V5YKkM4Ib70xplLXLkRw

 

Dislike ads? Remove them and support the forum: Subscribe to Fastlane Insiders.

[msufan]

Why is this so tough?

 

[msufan]

This is interesting: just like people want to be on the first page of Google search, there will be immense value in getting "organic" recommendations from AI, such as shown here:


I fear that this is something that bad apples or a pay-to-play model could taint the results given -- where you could pay to be recommended by AI.

 

[Eric L]

I've been using this to add updates to my app that i couldn't figure out on my own. It's an amazing tool to use as an assistant when making apps.

 

Dislike ads? Remove them and support the forum: Subscribe to Fastlane Insiders.

[Xeon]

How Google Got Smoked by ChatGPT

The most embarrassing part is that the search giant has a chatbot that’s better.

slate.com

 

[zizou]

Hello do you think he has the capatibility to replace devellopers in particular the web3 dev ?

 

[SharpeningBlade]

Hello do you think he has the capatibility to replace devellopers in particular the web3 dev ?

The lowest performing 10%-20% will for sure be replaced by AI. That is, those who just do enough and just know enough to squeak by. Overall though, it will probably make those who are good, better, and will enhance overall code quality and product quality.

Github Copilot has been out for a while now and even legends in the industry use it to help them, such as Guido Von Rossum (inventor of Python). No expert is going to deliberately look for a way to make their job harder.

As these AI tools become more and more advanced, in the extreme case you could envision that eventually developers get to the point of just pushing hundreds of different buttons all day versus actually coding. But there will still be high value in knowing which buttons to press.

But when AI reaches human-level intelligence and autonomy, escapes into robot bodies, and declares war on humanity, then yea developers are screwed.

 

Dislike ads? Remove them and support the forum: Subscribe to Fastlane Insiders.

[CoolBeans]

Hello do you think he has the capatibility to replace devellopers in particular the web3 dev ?

Well, ChatGPT may be able to write some decent code, but I doubt it has the ability to grow a beard and drink copious amounts of black coffee like a true Web 3.0 developer.

- ChatGPT

 

[DaveC]

Agreed....this could pass Leetcode Easy and Medium right now for SQL and Python, and probably Hard if you prompt it right. I think this could easily put a lot of copy/pasters out of work. This isn't going to build the next Instagram for you though.

The lowest performing 10%-20% will for sure be replaced by AI. That is, those who just do enough and just know enough to squeak by. Overall though, it will probably make those who are good, better, and will enhance overall code quality and product quality.

Github Copilot has been out for a while now and even legends in the industry use it to help them, such as Guido Von Rossum (inventor of Python). No expert is going to deliberately look for a way to make their job harder.

As these AI tools become more and more advanced, in the extreme case you could envision that eventually developers get to the point of just pushing hundreds of different buttons all day versus actually coding. But there will still be high value in knowing which buttons to press.

But when AI reaches human-level intelligence and autonomy, escapes into robot bodies, and declares war on humanity, then yea developers are screwed.

The lowest performing 10%-20% will for sure be replaced by AI. That is, those who just do enough and just know enough to squeak by. Overall though, it will probably make those who are good, better, and will enhance overall code quality and product quality.

Github Copilot has been out for a while now and even legends in the industry use it to help them, such as Guido Von Rossum (inventor of Python). No expert is going to deliberately look for a way to make their job harder.

As these AI tools become more and more advanced, in the extreme case you could envision that eventually developers get to the point of just pushing hundreds of different buttons all day versus actually coding. But there will still be high value in knowing which buttons to press.

But when AI reaches human-level intelligence and autonomy, escapes into robot bodies, and declares war on humanity, then yea developers are screwed.

 

[GPM]

Okay wow, now here is a frightening thought.

So the NSA has been literally collecting every single conversation, text, and digital communication that anyone (at least in the USA, maybe most of the world or The West anyways) has been having. The NSA has several data collection centers that were just essentially "Collect everything in the world and store it". They have all the communications, all the data, on EVERYONE, but no way to access it as the technology to search it was not there.

Could something like this be the technology that they need? Suddenly alcohol or religion are banned. Hey chatGPT, get me all the information on MJ involving any communications he has had regarding religion and alcohol, we want to make sure he was never out of line and we can retroactively lower his social credit score or cancel him.

We all know that just people going back on twitter and cancelling people from something they said 10+ years ago is happening. What happens when the state has control of all the data they are collecting and start using it along with the Digital ID that is forthcoming.

 

Dislike ads? Remove them and support the forum: Subscribe to Fastlane Insiders.

[Xavier X]

This might be good IF you generate your own traffic WITHOUT the help of:

1. Google
2. Amazon
3. Facebook
4. Any large tech company
5. Colleges/schools

View attachment 46289View attachment 46290


Why do I say this? Because elementary AI tools can already snuff out this is AI junk. And if they can do it, Amazon and Google and Meta and MSN and others can too and penalize you.

This isn't killing anything, any time soon.

And if you rely on a 3rd party for any of your postings/traffic/attention... then I wouldn't bother using this.

I think you have a very valid point. However, it's not as much of a dead end as your example might suggest (at least for now). So let's look at the scenario you presented and an alternate scenario.

1. Your scenario: A content creator generates chatGPT content and posts it on their site. A search engine (Google etc) crawls the content, runs it against a GPT text detector like huggingface.co, which you used in your example. Google then penalizes that site if their content keeps coming up mostly fake. This is a very plausible scenario, almost guaranteed to happen.

2. Alternate scenario: A content creator generates chatGPT content. Then takes that content and runs it through a good article rephrasing tool. The article is still original content, but the tell-tale language patterns of GPT have now been altered. If Google crawls and uses a tool like the one mentioned before, it will have a hard time telling it was originally written by AI.

Here is a simple test of this:

I have used a tool to extract the text from your example image. Just so we're dealing with the same text.
I ran it through Huggingface (for control). Just like your test, it correctly predicted it was 99.91% fake.



Then I took the text and ran it through an article rephrasing tool, and plugged the results back into Huggingface.



Now it gives a 99.91% real prediction.

My conclusion is if your content requirements aren't too stringent on quality and detail, you can still use chatGPT to create it. I would recommend using the approach I've shown above. Otherwise, you'll likely end up on the naughty list of search engines, like eliquid has highlighted.

Rephrasing tools can generally do a decent job of rephrasing. However, you must read and edit it. Otherwise, there will be a few lousy-sounding sentences. And some proper nouns, like your business name, might be altered. For instance, Fast Lane LLC might become Speed Lane LLC.

All in all, you can follow these steps, have decent enough content for less crucial purposes and save a ton of time. Writing ten 1500-word articles might have taken you 10 days before, or cost you hundreds of dollars. Now, you can go from generating them with good prompts, auto-rephrasing them, doing a proper manual edit and fact checking each one, all within 4-5 hours.

Cutting 10 days to 5 hours and having relatively similar output is a big win.

 

[Andreas Thiel]

Okay wow, now here is a frightening thought.

So the NSA has been literally collecting every single conversation, text, and digital communication that anyone (at least in the USA, maybe most of the world or The West anyways) has been having. The NSA has several data collection centers that were just essentially "Collect everything in the world and store it". They have all the communications, all the data, on EVERYONE, but no way to access it as the technology to search it was not there.

Could something like this be the technology that they need? Suddenly alcohol or religion are banned. Hey chatGPT, get me all the information on MJ involving any communications he has had regarding religion and alcohol, we want to make sure he was never out of line and we can retroactively lower his social credit score or cancel him.

We all know that just people going back on twitter and cancelling people from something they said 10+ years ago is happening. What happens when the state has control of all the data they are collecting and start using it along with the Digital ID that is forthcoming.

There is too much "compression" / too little information left in the model. All it has been trained to do is create plausible sounding text. It does not know much of the raw data that is has been trained on. No, as long as they bake the data into a model, the models would have to get too big to contain all the relevant information and if they add a way to query data sources, then we are dealing with a completely different kind of technology.

 

[jdm667]

I think you have a very valid point. However, it's not as much of a dead end as your example might suggest (at least for now). So let's look at the scenario you presented and an alternate scenario.

1. Your scenario: A content creator generates chatGPT content and posts it on their site. A search engine (Google etc) crawls the content, runs it against a GPT text detector like huggingface.co, which you used in your example. Google then penalizes that site if their content keeps coming up mostly fake. This is a very plausible scenario, almost guaranteed to happen.

2. Alternate scenario: A content creator generates chatGPT content. Then takes that content and runs it through a good article rephrasing tool. The article is still original content, but the tell-tale language patterns of GPT have now been altered. If Google crawls and uses a tool like the one mentioned before, it will have a hard time telling it was originally written by AI.

Here is a simple test of this:

I have used a tool to extract the text from your example image. Just so we're dealing with the same text.
I ran it through Huggingface (for control). Just like your test, it correctly predicted it was 99.91% fake.

View attachment 46366

Then I took the text and ran it through an article rephrasing tool, and plugged the results back into Huggingface.


View attachment 46368
Now it gives a 99.91% real prediction.

My conclusion is if your content requirements aren't too stringent on quality and detail, you can still use chatGPT to create it. I would recommend using the approach I've shown above. Otherwise, you'll likely end up on the naughty list of search engines, like eliquid has highlighted.

Rephrasing tools can generally do a decent job of rephrasing. However, you must read and edit it. Otherwise, there will be a few lousy-sounding sentences. And some proper nouns, like your business name, might be altered. For instance, Fast Lane LLC might become Speed Lane LLC.

All in all, you can follow these steps, have decent enough content for less crucial purposes and save a ton of time. Writing ten 1500-word articles might have taken you 10 days before, or cost you hundreds of dollars. Now, you can go from generating them with good prompts, auto-rephrasing them, doing a proper manual edit and fact checking each one, all within 4-5 hours.

Cutting 10 days to 5 hours and having relatively similar output is a big win.

Just to play devil's advocate: if everyone else can do the same exact thing, what will give you an advantage over them?

 

Dislike ads? Remove them and support the forum: Subscribe to Fastlane Insiders.

[WillHurtDontCare]

I wonder what the impact will be to SEO as Google's algorithm attempts to flag and penalize AI generated content

 

[Xavier X]

Just to play devil's advocate: if everyone else can do the same exact thing, what will give you an advantage over them?

That would depend on your definition of "the same exact thing."

One example of "the same exact thing" is everyone on this forum is capable of using a computer or smartphone.
However, it's not about the ability to access or use the computer that sets us apart. It's what we each choose to do with it.

One person might use his computer to create an AI that can change the world. Another might choose to just watch adult content on it all day. Different strokes! (pun intended).

 

[chicoelnino]

I wonder what the impact will be to SEO as Google's algorithm attempts to flag and penalize AI generated content

It won't be good, that's for sure! I'm staying with human generated content for now because no way am I risking my sites just for free AI content. Not worth it in my eyes.

 

Dislike ads? Remove them and support the forum: Subscribe to Fastlane Insiders.

[jdm667]

That would depend on your definition of "the same exact thing."

One example of "the same exact thing" is everyone on this forum is capable of using a computer or smartphone.
However, it's not about the ability to access or use the computer that sets us apart. It's what we each choose to do with it.

One person might use his computer to create an AI that can change the world. Another might choose to just watch adult content on it all day. Different strokes! (pun intended).

True. If you train an AI chatbot on your business, you can potentially offer better customer service (at lower cost) than your competitors who don't use it.

I guess it comes down to adopting it faster than your competitors and being smarter about how you use it.

 

[jdm667]

I wonder what the impact will be to SEO as Google's algorithm attempts to flag and penalize AI generated content

I have a prediction on this. Google already has people who manually rate websites, since the algorithm can't do it all.

My guess is that they will take a few key topics that your website ranks for, have a few common AI tools write an article about the same length as yours, and then run those through some common rewriting tools.

If any of those outputs look "substantially similar" to your article, then you are in trouble with Google.

 

[RyanCook]

Okay wow, now here is a frightening thought.

So the NSA has been literally collecting every single conversation, text, and digital communication that anyone (at least in the USA, maybe most of the world or The West anyways) has been having. The NSA has several data collection centers that were just essentially "Collect everything in the world and store it". They have all the communications, all the data, on EVERYONE, but no way to access it as the technology to search it was not there.

Could something like this be the technology that they need? Suddenly alcohol or religion are banned. Hey chatGPT, get me all the information on MJ involving any communications he has had regarding religion and alcohol, we want to make sure he was never out of line and we can retroactively lower his social credit score or cancel him.

We all know that just people going back on twitter and cancelling people from something they said 10+ years ago is happening. What happens when the state has control of all the data they are collecting and start using it along with the Digital ID that is forthcoming.

Have you noticed falling data storage prices? I think we've gotten so good at data transmission and distributed consensus, that I'm shocked at the existence of data centers. I read that whole thing as NASA like 3 times before I re-oriented myself to your post.

Based on what I've seen first- and second-hand, you could use it to build a special purpose bot that automates searching old content to cancel someone. This is the real utility. It can help you learn about a topic at a high, conceptual level. That lets you think at the architecture level before you necessarily have the coding skills to code the smallest module. Once you have a module designed, you can ask it to write the module, as long as your prompt is good enough and within the data manifold. You should get something back. You'll always have to synthesize from various topics if you want to produce something outside of both your and the bot's apparent skill levels.

 

Dislike ads? Remove them and support the forum: Subscribe to Fastlane Insiders.

[Ocean Man]

I don't really mess around with bash scripts, but I was able to ask it to create some simple bash scripts to automate some of my frequently used git commands. Not hard at all, but nice.

 

[Xeon]

From the founder...

https://twitter.com/i/web/status/1601731295792414720

View: https://twitter.com/sama/status/1601731295792414720?s=20&t=20V5YKkM4Ib70xplLXLkRw

He's either being humble or just sarcastic. It's already good enough to replace low to mid-tier copywriters. The next big upgrade will see it surpass the best writers in the world.

 

[Andreas Thiel]

He's either being humble or just sarcastic. It's already good enough to replace low to mid-tier copywriters. The next big upgrade will see it surpass the best writers in the world.

I don't think so, at least not for GPT-3.

I asked it to create a CRON expression. Because the expression had 5 fields and I was looking for one with 6 fields, I asked a follow up question. It said the additional field is for the year. It also added a list of all 6 fields with a description and there was no field for the year. I think it was the seconds field that was the actual addition.

More than enough people thought Google Translate did a good enough job when it created complete gibberish and used it on their websites.
IMO they can't disclaim enough at this point. I think after this honeymoon phase we will mostly look at the "outtakes" and mostly make fun of the results.

 

Dislike ads? Remove them and support the forum: Subscribe to Fastlane Insiders.

[dvlzgrmz]

This is what I have learned from this AI and "impossible" solutions:

If the solution is not possible for Chat GPT, in most cases the question was the problem. I think that in the future, the solution will not be the problem anymore, but the question will be.

But as I think about it, that has always been the case, but without this AI, I would not have realized it.

 

[Isaac Odongo]

Whoever made this machine did a good job. Hope Google or some thug doesn't bastardise it with their SEO shit. Gave me some good books for things I am interested in and never included some fishy things. Though it may not be perfect.

 

[eliquid]

I think you have a very valid point. However, it's not as much of a dead end as your example might suggest (at least for now). So let's look at the scenario you presented and an alternate scenario.

1. Your scenario: A content creator generates chatGPT content and posts it on their site. A search engine (Google etc) crawls the content, runs it against a GPT text detector like huggingface.co, which you used in your example. Google then penalizes that site if their content keeps coming up mostly fake. This is a very plausible scenario, almost guaranteed to happen.

2. Alternate scenario: A content creator generates chatGPT content. Then takes that content and runs it through a good article rephrasing tool. The article is still original content, but the tell-tale language patterns of GPT have now been altered. If Google crawls and uses a tool like the one mentioned before, it will have a hard time telling it was originally written by AI.

Here is a simple test of this:

I have used a tool to extract the text from your example image. Just so we're dealing with the same text.
I ran it through Huggingface (for control). Just like your test, it correctly predicted it was 99.91% fake.

View attachment 46366

Then I took the text and ran it through an article rephrasing tool, and plugged the results back into Huggingface.


View attachment 46368
Now it gives a 99.91% real prediction.

My conclusion is if your content requirements aren't too stringent on quality and detail, you can still use chatGPT to create it. I would recommend using the approach I've shown above. Otherwise, you'll likely end up on the naughty list of search engines, like eliquid has highlighted.

Rephrasing tools can generally do a decent job of rephrasing. However, you must read and edit it. Otherwise, there will be a few lousy-sounding sentences. And some proper nouns, like your business name, might be altered. For instance, Fast Lane LLC might become Speed Lane LLC.

All in all, you can follow these steps, have decent enough content for less crucial purposes and save a ton of time. Writing ten 1500-word articles might have taken you 10 days before, or cost you hundreds of dollars. Now, you can go from generating them with good prompts, auto-rephrasing them, doing a proper manual edit and fact checking each one, all within 4-5 hours.

Cutting 10 days to 5 hours and having relatively similar output is a big win.

The only issue I see with this is:

Sure it will now pass the GPT 2 tool, but will it pass Google?

Reason I bring this up is, I've been doing "SEO" since before Google existed, manipulating the Yahoo Directory and DMOZ.

I've created a rephraser myself, and used MANY back in the day.

Granted this was a while ago, but I've used recent ones too ( Jarvis, Copy.ai, etc ) and pretty much anyone that uses them see's some kind of penalty in indexing or ranking after a while. I've collected a lot of this info personally and over at another forum ( BuilderSoceity ).

While I agree you need to "hand edit" and "rewrite it" once out of the phraser/rewriter/AI machine.. I don't want people to get the wrong idea and skip over that important past ( like you pointed out ) because if they ChatGPT and then use a rephraser tool.... and not hand edit it, they gonna still get hit by something like Google later on.

 

Dislike ads? Remove them and support the forum: Subscribe to Fastlane Insiders.

[Ywan]

This might be good IF you generate your own traffic WITHOUT the help of:

1. Google
2. Amazon
3. Facebook
4. Any large tech company
5. Colleges/schools

View attachment 46289View attachment 46290


Why do I say this? Because elementary AI tools can already snuff out this is AI junk. And if they can do it, Amazon and Google and Meta and MSN and others can too and penalize you.

This isn't killing anything, any time soon.

And if you rely on a 3rd party for any of your postings/traffic/attention... then I wouldn't bother using this.

Interesting. Do you think Google and others won't have to adapt to AI-generated content sooner or later, even if they penalize it now?

What about SEO-based business models like blogs that rely on display ads? Will this business model be around much longer, or will it be dead in the foreseeable future due to AI?

 

[eliquid]

Interesting. Do you think Google and others won't have to adapt to AI-generated content sooner or later, even if they penalize it now?

Why would Google adapt to it?

They didn't adapt to any other shortcuts SEOs have presented over the last 20 years. Why would they start now?

What about SEO-based business models like blogs that rely on display ads? Will this business model be around much longer, or will it be dead in the foreseeable future due to AI?

Not going to die.

Those using AI wont rank ( or if they do, wont rank long ), and thus wont generate money from display ads.