ChatGPT AI is a huge breakthrough, a Google killer

[Fox]

The more GPT4's responses are limited, the worse it becomes. When 4 first came out, it was amazing. A few months later, it's like GPT 3.5 just before 4 came out. Even when I give a direct command like "Don't use subheadings" it just ignores it and does it anyway.

My experience also - give it a simple command with some kind of rule in all CAPITAL LETTERS and it still overrides it completely. Then ask it what it just did in the second message and it fixes it.

They massively dumbed it down.

 

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

[Consolation]

And they said it's not. It's like Facebook saying their users' data privacy are well taken care of.

AI OpenAI: No, ChatGPT Isn't Getting Dumber

Every time I use it, I will often fight with ChatGPT due to frustration. The joke is on me, though. Cursing and insulting it has no effect. But, improving my prompts doesn't do much, either.

 

[amp0193]

Add a plugin.

wow ok, how did I miss this was a thing.

Guess I was under a rock when it launched, and the vanilla UX does not make it at all obvious that this apps were a thing you can add.

Thanks @Andy Black


FWIW, data point of one: I haven't noticed anything being dumber.

 

[Tyler_Durden]

This thing is Insane for Coding. I think it might replace the entire coding market. Coding is incredibly sensitive which means that just one mistake in a 10000 letter code will cause the code to not work anymore. Chat GTP just fixes the problem extremly effectivly. Just type in :"Create a Basic Website with a picture of James Bond in the middle" and it will create it trough a HTML file(you would just need to put in a editor).

 

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

[Rollio]

ChatGPT use declines for the first time.

Sort of reflects my own usage of it, as it seems to have become dumber, and more inaccurate.

View attachment 50734

View attachment 50733

It’s gotten dumber because OpenAI’s now paywalled the newest version that was previously free

 

[Andy Black]

Getting lots of tips and ideas from this video:

View: https://youtu.be/ka1Pqk2o3tM

 

[Andy Black]

I just reinstalled the AIPRM: AI Prompt Marketplace for ChatGPT, Midjourney & DALL-E Chrome Extension and my ChatGPT interface looks like this:

 

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

[uhhfeef]

Hey all!

Quick disclaimer: I used prompts to generate the below summary of OpenAI Devday.

Alternatively you can watch the entire thing here, or read the summary below to get a quick understanding of the future of AI.

New models and developer products​

TL;DR: OpenAI DevDay introduced GPT-4 Turbo, Assistants API, multimodal features, DALL·E 3 API, model customization, higher dev rate limits, and Copyright Shield.

Enhanced Interaction and Cost Efficiency​

• GPT-4 Turbo: This is an advanced version of the GPT-4 model that allows for longer conversations with users by supporting a context window of 128K tokens. It's designed to be more cost-effective, reducing the price for processing input and generating output, making it more accessible for extensive use.
• Updated GPT-3.5 Turbo: An updated iteration of the GPT-3.5 model featuring a 16K token context window, which strikes a balance between the extended capabilities of GPT-4 and the affordability for developers and businesses.

Developer-Friendly APIs and Tools​

• Assistants API: A new API that streamlines the development of AI-powered applications, enabling tasks such as running code, retrieving information, and calling specific functions within the AI's capabilities.
• Improved Instruction Following and JSON Mode: GPT-4 Turbo now has enhanced abilities to follow complex instructions more accurately and supports a JSON mode, which allows developers to work with structured data more efficiently.

Multimodal and Customization Capabilities​

• Multimodal Capabilities: OpenAI has expanded the AI's abilities to include not just text but also vision and audio, such as analyzing images, generating images with DALL·E 3, and converting text to speech with various voice options.
• Model Customization: Developers and organizations can now fine-tune the GPT-4 model or create entirely custom models using their own datasets, which is particularly useful for specialized applications.

Advanced Functionality and Accessibility​

• Function Calling Updates: The AI's function-calling feature has been improved to allow multiple functions to be called in a single message, and the returns of function parameters are now more accurate.
• Reproducible Outputs and Log Probabilities: OpenAI has introduced features that enable the AI to produce consistent outputs for the same inputs and to provide log probabilities for each token it generates, aiding in the analysis of the AI's decision-making process.

Visual and Audio Integration​

• GPT-4 Turbo with Vision: This feature allows GPT-4 to process images as inputs, enabling it to perform tasks like detailed image analysis and reading text from documents.
• DALL·E 3 API: Developers can integrate the DALL·E 3 image generation model into their applications, allowing for the creation of unique images based on textual descriptions.
• Text-to-Speech API: This API provides the ability to generate human-like speech from text, offering a range of voice types for various applications.
• Whisper v3 and Consistency Decoder: An improved automatic speech recognition model and a new decoder that enhances the quality of generated images, improving the AI's ability to understand spoken language and create visual content.

Cost Management and Legal Protection​

• Lower Prices and Higher Rate Limits: OpenAI has reduced the prices for using its models and increased the rate at which tokens can be processed, making it more affordable and scalable for users.
• Copyright Shield: A new initiative to provide legal support and cover costs for customers who face copyright infringement claims related to the content generated by the AI, adding a layer of protection for businesses.

Introducing GPTs​

TL;DR: OpenAI's GPTs are customizable AIs for various tasks, monetizable via the GPT Store, with strong privacy and API integration. ChatGPT Plus is updated for simplicity and current information.

Customization and Community Innovation​

• Tailored AI Tools: GPTs (customizable ChatGPTs) are designed to allow individuals to create their own AI that can assist with specific tasks. This could range from helping to learn new games, assisting with educational activities, or aiding in creative processes like design. The process is made simple and accessible, requiring no programming skills, which opens up AI customization to a broader audience.
• Community Contributions: OpenAI is leveraging the creativity and expertise of the wider community by enabling users to contribute their own GPTs. This initiative is expected to result in a rich variety of GPTs that cater to different niches and interests, all built by users from diverse backgrounds.

Monetization and Marketplace​

• GPT Store Launch: OpenAI is introducing a GPT Store, a platform where individuals can share their custom GPTs with the public. This marketplace not only serves as a hub for discovering GPTs but also as a venue for creators to potentially earn revenue based on the popularity and usage of their GPTs.

Privacy, Safety, and Real-World Application​

• Data Privacy and User Safety: With the introduction of GPTs, OpenAI emphasizes the importance of user privacy and safety. Users have control over their data, and there are measures in place to ensure that interactions with GPTs are secure. Additionally, OpenAI has systems to review GPTs against usage policies to prevent the distribution of harmful content.
• API Integration and Actions: Developers can enhance the functionality of GPTs by connecting them with external data sources or services through APIs. This allows GPTs to perform tasks that have tangible impacts in the real world, such as managing databases, integrating with email systems, or supporting e-commerce activities.

Enterprise Solutions and Enhanced User Experience​

• Enterprise Custom GPTs: For businesses, GPTs offer a new level of customization. Companies can create GPTs for specific departments or functions, which can assist with tasks like drafting documents in a particular brand voice or providing support for customer service inquiries.
• Improved ChatGPT Plus: The updated ChatGPT Plus service now provides users with more current information, up to April 2023, and a more user-friendly experience by consolidating all model interactions into a single interface, eliminating the need to switch between different AI models.

 

[MJ DeMarco]

Hey all!

Quick disclaimer: I used prompts to generate the below summary of OpenAI Devday.

Alternatively you can watch the entire thing here, or read the summary below to get a quick understanding of the future of AI.

New models and developer products​

TL;DR: OpenAI DevDay introduced GPT-4 Turbo, Assistants API, multimodal features, DALL·E 3 API, model customization, higher dev rate limits, and Copyright Shield.

Enhanced Interaction and Cost Efficiency​

• GPT-4 Turbo: This is an advanced version of the GPT-4 model that allows for longer conversations with users by supporting a context window of 128K tokens. It's designed to be more cost-effective, reducing the price for processing input and generating output, making it more accessible for extensive use.
• Updated GPT-3.5 Turbo: An updated iteration of the GPT-3.5 model featuring a 16K token context window, which strikes a balance between the extended capabilities of GPT-4 and the affordability for developers and businesses.

Developer-Friendly APIs and Tools​

• Assistants API: A new API that streamlines the development of AI-powered applications, enabling tasks such as running code, retrieving information, and calling specific functions within the AI's capabilities.
• Improved Instruction Following and JSON Mode: GPT-4 Turbo now has enhanced abilities to follow complex instructions more accurately and supports a JSON mode, which allows developers to work with structured data more efficiently.

Multimodal and Customization Capabilities​

• Multimodal Capabilities: OpenAI has expanded the AI's abilities to include not just text but also vision and audio, such as analyzing images, generating images with DALL·E 3, and converting text to speech with various voice options.
• Model Customization: Developers and organizations can now fine-tune the GPT-4 model or create entirely custom models using their own datasets, which is particularly useful for specialized applications.

Advanced Functionality and Accessibility​

• Function Calling Updates: The AI's function-calling feature has been improved to allow multiple functions to be called in a single message, and the returns of function parameters are now more accurate.
• Reproducible Outputs and Log Probabilities: OpenAI has introduced features that enable the AI to produce consistent outputs for the same inputs and to provide log probabilities for each token it generates, aiding in the analysis of the AI's decision-making process.

Visual and Audio Integration​

• GPT-4 Turbo with Vision: This feature allows GPT-4 to process images as inputs, enabling it to perform tasks like detailed image analysis and reading text from documents.
• DALL·E 3 API: Developers can integrate the DALL·E 3 image generation model into their applications, allowing for the creation of unique images based on textual descriptions.
• Text-to-Speech API: This API provides the ability to generate human-like speech from text, offering a range of voice types for various applications.
• Whisper v3 and Consistency Decoder: An improved automatic speech recognition model and a new decoder that enhances the quality of generated images, improving the AI's ability to understand spoken language and create visual content.

Cost Management and Legal Protection​

• Lower Prices and Higher Rate Limits: OpenAI has reduced the prices for using its models and increased the rate at which tokens can be processed, making it more affordable and scalable for users.
• Copyright Shield: A new initiative to provide legal support and cover costs for customers who face copyright infringement claims related to the content generated by the AI, adding a layer of protection for businesses.

Introducing GPTs​

TL;DR: OpenAI's GPTs are customizable AIs for various tasks, monetizable via the GPT Store, with strong privacy and API integration. ChatGPT Plus is updated for simplicity and current information.

Customization and Community Innovation​

• Tailored AI Tools: GPTs (customizable ChatGPTs) are designed to allow individuals to create their own AI that can assist with specific tasks. This could range from helping to learn new games, assisting with educational activities, or aiding in creative processes like design. The process is made simple and accessible, requiring no programming skills, which opens up AI customization to a broader audience.
• Community Contributions: OpenAI is leveraging the creativity and expertise of the wider community by enabling users to contribute their own GPTs. This initiative is expected to result in a rich variety of GPTs that cater to different niches and interests, all built by users from diverse backgrounds.

Monetization and Marketplace​

• GPT Store Launch: OpenAI is introducing a GPT Store, a platform where individuals can share their custom GPTs with the public. This marketplace not only serves as a hub for discovering GPTs but also as a venue for creators to potentially earn revenue based on the popularity and usage of their GPTs.

Privacy, Safety, and Real-World Application​

• Data Privacy and User Safety: With the introduction of GPTs, OpenAI emphasizes the importance of user privacy and safety. Users have control over their data, and there are measures in place to ensure that interactions with GPTs are secure. Additionally, OpenAI has systems to review GPTs against usage policies to prevent the distribution of harmful content.
• API Integration and Actions: Developers can enhance the functionality of GPTs by connecting them with external data sources or services through APIs. This allows GPTs to perform tasks that have tangible impacts in the real world, such as managing databases, integrating with email systems, or supporting e-commerce activities.

Enterprise Solutions and Enhanced User Experience​

• Enterprise Custom GPTs: For businesses, GPTs offer a new level of customization. Companies can create GPTs for specific departments or functions, which can assist with tasks like drafting documents in a particular brand voice or providing support for customer service inquiries.
• Improved ChatGPT Plus: The updated ChatGPT Plus service now provides users with more current information, up to April 2023, and a more user-friendly experience by consolidating all model interactions into a single interface, eliminating the need to switch between different AI models.

Moved to the existing thread.

 

[MJ DeMarco]

Wow the latest updates to ChatGTP plus are awesome.

If you're in the business idea phase, there's going to be a period of time (1-2 years?) when this technology will be able to make you a fortune, this is the early adoption phase when the rest of the world is catching up.

I'd take a hard look at it if I was in that phase.

 

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

[Kak]

Wow the latest updates to ChatGTP plus are awesome.

If you're in the business idea phase, there's going to be a period of time (1-2 years?) when this technology will be able to make you a fortune, this is the early adoption phase when the rest of the world is catching up.

I'd take a hard look at it if I was in that phase.

I've been using the free GPT-4 access integrated in Edge browser / bing and I'm very happy with it.

It has helped quickly create professional reports and policies required by new customers that usually take a long time.

I even had it model my policies after an unrelated company that I think does a great job with that stuff.

 

[Panos Daras]

I really have no idea how so many people are shitting on ChatGPT.

I have been using it for my business and my work and it is a game changer. I hadn't seen such a significant innovation until the internet and Googling stuff became a thing.

The problem I see is most people are very lazy and do not want to give the chat enough context to work on. With Chat GPT I managed to translate my website in 3 languages using translations that a) respect the context b) Use similar expressions that are related to the language spoken! Other translation services are not even close to that.

Use Chat GPT and Hemingway and you can write copy that is simple and impactful in minutes. But it needs to have context! What technique do you want it to use? Ask it to imitate other text! Done! Problem with the website? Ask chat and it is fixed. But it also does not stop there. At work, I have teams OK? I record the meeting there. I download all the transcripts, plug it into a Txt file, put them in the code interpreter (now redundant to use code interpreter), and boom, all the points discussed in minutes. Ask it then to send an email to all for meeting minutes. Boom done. But you need to tell it how to do it in custom instructions. You need to tell it, for example, for all answers to interchange between long and short sentences to sound human. You need to tell it to use a business tone that is technical but not too complex.

Also, if you are lazy, I got you! Create shortcuts that pre-provide the prompt using a browser add-on called Magical. Just press // and select between prompts. Select the prompt from the menu and boom done. Bored of typing? Just use Windows key + H or the microphone key in Mac and talk to it. Give it a lot, lot, lot, lot, lot of context to work with. And then you see the magic happen.

Vision? Holly Molly, how good it is. I did a test and I managed to interpret an invoice from 1901 handwritten with scribbles to a detailed friggin breakdown! Do you know that when I was working in one of the biggest accommodation companies in the world we paid a stupid consultant A-hole corp 1,000,000EUR for such a project? And the software capabilities are not even close to ChatGPT. Half the data insertion was crap. And imagine, that software was specifically trained in invoice processing for decades! I mean the opportunities are endless here. And this is just the beginning. You can do crazy stuff with the new updates.

Of course, the caveat is that contrary to what it was in the past (eg with coding), it is too easy for more people to do the same. Personally, I believe that the use cases are endless and you can learn it really really fast. And you should really. So stop complaining about it and go buy that service because the custom instructions are invaluable and the one that is "FREE" via Bing is a watered-down version that is crap really compared to the paid one.

 

[WillHurtDontCare]

Wow the latest updates to ChatGTP plus are awesome.

If you're in the business idea phase, there's going to be a period of time (1-2 years?) when this technology will be able to make you a fortune, this is the early adoption phase when the rest of the world is catching up.

I'd take a hard look at it if I was in that phase.

For anyone who is interested in taking this route - there are a bunch of ML / AI / GPT related webinars on LinkedIn. You can join some of the ones hosted on zoom and talk to people about AI, then schedule calls with them later to see how it might be able to help their business. I think that you can get lists of those attendees after the fact, then you can just message them on LinkedIn or find their emails. It's a good way to find high intent prospects.

example below (note I used LinkedIn Sales Navigator, which shouldn't matter, but in case someone can't see what I see, that might be the reason):

Search "GPT" - 265 events



Click on the event and click attendees



see all of the attendees:

 

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

[RudyR]

Yeah I like how you can now create your own GPT now. It's under Explore > Create a GPT. You start in the Create tab, and follow the prompts it give you. You can then modify in the Configure tab. This now eliminates the need for me to copy/paste various prompts. I can now create my own, Business assistant, programming assistant, etc. and easily switch between them. You can even upload files so it'll have better context to pull from.

You can have it private, share access via link, or public. It looks like the public option puts it in the upcoming GPT store. Which according to their docs, they will pay you based on how many people use it.

Introducing GPTs

You can now create custom versions of ChatGPT that combine instructions, extra knowledge, and any combination of skills.

openai.com

 

[Stan_]

This thing is Insane for Coding. I think it might replace the entire coding market. Coding is incredibly sensitive which means that just one mistake in a 10000 letter code will cause the code to not work anymore. Chat GTP just fixes the problem extremly effectivly. Just type in :"Create a Basic Website with a picture of James Bond in the middle" and it will create it trough a HTML file(you would just need to put in a editor).

How much experience in web design / programming do you have?

 

[srodrigo]

I find ChatGPT really useful for a variety of things. But I'm not sure this is the new high-speed avenue for making big money for the majority of us. Maybe I'm plain wrong, but this feels to me like a crypto bull run, with everyone and their grandmothers creating prompts and talking about the topic. One more gold rush, just that this one will be monitored by governments closely before AI gets out of hand. Just look at home many GPT frontends there are out there, probably hundreds, yet only a couple of them make significant money. Survivorship bias?

 

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

[uhhfeef]

The best thing to come out of AI is definitely Githubs copilot and copilot chat.

Started using it, it definitely needs more updates but the use cases for this is enormous.

Literally anyone and their grandma now has the capability of building apps now, there's no excuse to not build on your idea and test it out.

I'm really excited because I use GPT4 a lot to create my MVP's and its been extremely fruitful. The first few months can be really tough because I do think there is a lil learning curve with how GPT could be used for your specific idea.

But once you make it work, man is it good.

Definitely gona change the landscape of developing and just... building your idea to life. We're going to see a rise is natural language programmers more as well (which I believe is the natural progression to development as a whole).

I'm gona mess around with copilot chat but this is exactly what I've wanted for a while now. Props to github for solving this problem.

Exciting times!!!

 

[RightyTighty]

Wow the latest updates to ChatGTP plus are awesome.

Yeah I like how you can now create your own GPT now. It's under Explore > Create a GPT. You start in the Create tab, and follow the prompts it give you. You can then modify in the Configure tab. This now eliminates the need for me to copy/paste various prompts. I can now create my own, Business assistant, programming assistant, etc. and easily switch between them. You can even upload files so it'll have better context to pull from.

You can have it private, share access via link, or public. It looks like the public option puts it in the upcoming GPT store. Which according to their docs, they will pay you based on how many people use it.

The new ability to creating your own GPT's is the best feature since automated code writing, and it appears to still be kind of a sleeper(?).

I’m fine-tuning a GPT now to write the front end of Provisional Patent Applications. Upcoming endeavors will include: drafting patent claims, search engine for overseas sourcing, and a search engine to create a data set of manufacturers with contact information for product licensing.

 

[srodrigo]

Okay, I'm on the hype pony, I just renewed my Plus subscription. Time to have fun with custom GPTs.

 

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

[Dufresne85]

Wow this is crazy - I just created my own business mentor GPT based on various business books and models such as CENTS. Now it can evaluate a business idea I have and give very specific feedback... the weird part is, if I would put this GPT public - people could just download the knowledgebase files, violating copyright for sure?

 

[srodrigo]

Kind of disappointed with this GPTs thing. I created one and it was ignoring instructions. It even removed some of the (non-conflicting) features when I instructed to add some new ones. And the actual prompt using the GPT was so slow, over 40 seconds to generate a medium example. Maybe they'll improve it, but so far I'm not impressed at all.

 

[MakeItHappen]

So, do I understand this right? You could create your own business mentor gpt and upload 100 or 1000 books so that the business mentor gpt can use them. This could make the business mentor gpt way better than the normal gpt4 because the normal gpt4 didn't have access to books because of copyright issues?

 

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

[uhhfeef]

Kind of disappointed with this GPTs thing. I created one and it was ignoring instructions. It even removed some of the (non-conflicting) features when I instructed to add some new ones. And the actual prompt using the GPT was so slow, over 40 seconds to generate a medium example. Maybe they'll improve it, but so far I'm not impressed at all.

Depends on a number of factors but most likely you're not setting the context correctly.

 

[artz]

Wow this is crazy - I just created my own business mentor GPT based on various business books and models such as CENTS. Now it can evaluate a business idea I have and give very specific feedback... the weird part is, if I would put this GPT public - people could just download the knowledgebase files, violating copyright for sure?

Yes, it's easily possible to reverse engineer. I hope OpenAI addresses this issue in the future.

 

[srodrigo]

Depends on a number of factors but most likely you're not setting the context correctly.

Could be. I was also thinking about this as a REST API: first get the scaffolding of the info. Then expand on point by point with extra queries. But this wouldn't work for what I was trying to do (generate travel guides I could turn into a PDF), users would need to wait for too long. Maybe I wasn't setting the context well, but there wasn't that much context apart from "add start time between locations" and things like that.

 

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

[Xavier X]

ChatGPT 3.5 has definitely degraded in quality. I've used it from the very start and the difference is notable.

Main issues for me are how it ignores some instructions, and when corrected still ignores them. Also, it now writes with overly flowery words and unusual phrases a native English speaker typically won't use. It often sounds like a Filipino or Indian writer on Fiverr.

My guess is the bulk of its current human training input is from non-native speakers. That will inevitably have such an effect on its writing style.

 

[amp0193]

You could create your own business mentor gpt and upload 100 or 1000 books so that the business mentor gpt can use them.

That’s the idea, but from what I’ve been reading there are some limits on how much “knowledge” context you can upload.

I’m still waiting for the day where there’s an easy way for me to train a LLM on 7 years of cx email inbox, and then use a GPT to extrapolate it into a support knowledge base, a tech support flow chart, auto craft replies, etc.

I think that day will be here sooner rather than later, but for now, several of the ideas I have with custom gpts require large amounts of training data.

 

[MJ DeMarco]

Kind of disappointed with this GPTs thing. I created one and it was ignoring instructions.

I had the same issue.

The assistants thing seems to be a hit-or-miss.

I created an assistant to create X posts based on my work, and it absolutely refused to follow instructions. I literally had to repeat the instruction 7 times, and it only followed the instruction when I made the instruction part of the title.

Keep in mind, this thing is in Beta, so under that notion, I'm pretty impressed on what's to come.

The other assistant I created was a business idea evaluator based on my work, CENTS, etc., and it seems to do pretty well with that.

 

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

[srodrigo]

The other assistant I created was a business idea evaluator based on my work, CENTS, etc., and it seems to do pretty well with that.

Maybe it depends on what we are trying to make then. Anyway, as you say, this is still in beta, it should only improve

 

[srodrigo]

It looks like GPTs are quite popular:

https://twitter.com/i/web/status/1724626002595471740

View: https://twitter.com/sama/status/1724626002595471740