ChatGPT Is Starting To Do What You (Actually) Want It To

ChatGPT Is Starting To Do What You (Actually) Want It To

Have you wrestled with ChatGPT's faulty and limited memory? Or wished you could customize it more for your life or work? It's happening now. ?? If you are a time-strapped worker or independent business, this 4-point update is for you.

(Point #3 is essential if you want to know what's about to affect your entire #techstack.)


Friends, for real, did you hear about OpenAI's big AI announcements yesterday?!?(This is not just impacting developers!!)

Let me break down what their announcements mean for us normal folks just trying to work 'smarter' and live better...


1?? Better memory (like, right now)

If you've ever wrestled with ChatGPT "forgetting" something, news for you! The new #GPT4Turbo model has a memory over 16x longer than the GPT-4 version you were using in ChatGPT or Playground before yesterday.

This new model can remember up to 100,000 words (128k "tokens") -- instead of just a short essay's worth before. 100k words is about 300 pages of a book.

.gif from a cartoon character saying "I've already forgotten what we were talking about." We know, GPT. We know.


?? What to know:

  • You can use this GPT-4 Turbo model NOW if you're a Plus user. It's not rolling out for everyone in ChatGPT yet (expect it this week), but you can use it on Playground already. Comment if you need help finding this.
  • The regular GPT-4 model is now trained on internet history through April 2023 (rather than September '21).
  • All existing ChatGPT Plus users just had their rate limits doubled. So if you could do 25 messages/hr, you can now do 50. You can also request to have your rate limits increased! Ask me how below.

??? And possibly the most exciting update to the regular ChatGPT Plus user...

  • Now, all those separate GPT-4 abilities (image reading, internet browsing, #DALLE3 image gen, etc.) will be combined into ONE tab - so they're now just built-in abilities of the new default GPT-4. So, you don't have to switch between 3 different ChatGPT tabs to scan an image, then search the internet, then create a new image, etc... but the plugins are going away. See point #3 below for what's replacing them.


(?? If you have ChatGPT Plus, you might not see all of these shiny new changes reflected in your ChatGPT interface yet - I still don't.

Expect to get the update this week though! And remember, you can already access the longer memory in Playground; comment with any Qs.)


2?? More Copyright Suit Protection (for some)

OpenAI is now joining other AI-forward, Big Tech companies when it comes to addressing copyright claims their customers might face.

Now if you face legal action on grounds of copyright infringement or abuse, their new "Copyright Shield" will pay the legal costs incurred...

??? ...IF you're using the "generally available" features of their developer platform or ChatGPT Enterprise. This does not seem to extend to the free or Plus tiers of OpenAI at this point.

(Is this an area you want to know more about? Let me know)


3?? The New App Store (but bigger)

You are about to be able to use, create, and even sell* custom GPTs.

Would your work be easier if you had an assistant that could take in your data, your questions, your files (etc) and give recommendations... then actually implement them too?? Then you should know:

OpenAI announced these "GPTs" - like customized mini AIs (aka "agents") that can help with specific tasks. And yes, they can be trained on YOUR documents (?).

That also means, your apps are getting smart assistants.?Or, they will soon be replaced BY / become smart assistants.

?? OpenAI previewed example use cases for these GPTs like lesson planning, graphic design, workflow automation, and more.

Sam Altman shows an example of a new GPT in action. The Canva GPT responds to a user who asks it to make a poster, giving a few key details. The Canva GPT creates 2 versions of the poster with different aesthetic approaches. Their info is accurate.

Even wilder, the average person will be able to build, use, and even sell their own custom GPTs just by using a text prompt and uploading some documents. That's right. YOU can build your own AI app/agent, called a GPT.

Think of these like apps on your smartphone -- do you use the App Store? Would you have benefited from knowing how big the App Store was going to be when it launched... especially if you could actually just describe the kind of app you want in text to create it?!

Get ready for the new GPT Store. Start dreaming now about what apps you want to use - and even make.

*You can get revenue share from OpenAI if your app is successful. This is about to become some people's - and companies' - fulltime endeavors.

.gif from Parks & Rec showing one of the most absolutely ridiculous characters (played by Jenny Slate) saying "Money, please!!" with her hand out. Same.

I'll let y'all know when the GPT Store actually launches but it is coming later THIS month, November 2023.

(?? Let's connect or follow me so I can keep you updated - this is getting too huge to not share about. Should I start this as an ongoing newsletter?)


4?? Assistants API

Helloooo #TechStartups ?? the new "Assistants API" makes it simple for developers (even beginners) to give apps conversational abilities, like chatbots or virtual assistants.

These now allow -- as of yesterday:

  • Persistent threads (memory can be accessed & used between chats... huge.)
  • Built-in retrieval (to give the assistant more knowledge beyond user messages)
  • Code Interpreter (all the abilities of Code Interpreter / "Advanced Data Analysis" are now IN the API. It can read/write code.)
  • Improved function calling (the API / AI model is better at working with users in a more interactive, context-aware way = more accurate and relevant responses)


There is a lot more to discuss here, but I want to keep this brief and focused on ppl who aren't super familiar with APIs. If you want to see it in action, skip straight to this part of the OpenAI #DevDay speech: https://www.youtube.com/live/U9mJuUkhUzk?si=jiQB487Xe3l49J9f&t=2050


?? Oh, and: costs for models like GPT-4 and GPT-3.5 just dropped 2-3x, making AI more attainable for organizations of all sizes. That's a big cost-saver for those creators/devs building on top of it. So if the AI apps you're using (or building) raise their prices, it's not because of the API cost ;)


5?? Fine-tuned Models for Enterprise

And last point -- I am not focused on enterprises, but FYI, y'all will be able to train custom models company data, for internal company use. At scale. You can apply for this experimental fine-tuning program now and OpenAI will help you set it up.

As in, work directly with you. Let's advocate for expanding access to smaller companies.

In the meantime, I'll be sharing practical alternatives here - the new GPTs are a huge step in that direction.


There's sooo much more to share about how these new updates can be used for greater productivity, creativity, and accessibility. More on that to come.


So now the question is...


What will you leverage AI for? Automating repetitive tasks? Building a custom advisor?

Or, are you hesitant due to the massive ethical implications, or the rapid stream of hype-heavy announcements? (You are NOT alone.)


Let me know how you think AI could upgrade your work and quality of life (or, if it already has)!


Mark Modesti

Client Success

11 个月

I must be missing something. I spent the morning creating a GPT and as I can tell, it’s drawing on the same knowledge base as regular ChatGPT 4. And when you’re configuring it, every session is a new session. Apart from the prompt starters, nothing you add is retained by the model. What’s the primary purpose of this new feature?

回复

要查看或添加评论,请登录

社区洞察

其他会员也浏览了