Indeed.
“Everything is ultimately computational.”
Lol it couldn’t find it in its memory, the blog post experiment. But this shows you how interesting it is to have it be able to reflect on “experiences” it does remember.
Try this GPTs https://chat.openai.com/g/g-dVQNBgzdG-appcreator to develop your own actions easily.
I’ve made a little over 30 GPTs focusing on finance, human relationships, investments, and playful ones and I use majority the investment strategy ones along with the finance ones frequently to attract venture capitalist to invest in start ups
Well. I’ve found the best thread in the forum. This is what I have been looking for.
I’m doing a lot of rapid building of APIs to feed GPTs as well and getting some pretty solid improvement in experience vs. vanilla GPT-4.
OK. Need to go to sleep now so I can do more coding tomorrow. Just too much fun, these GPTs.
1 Like
Bizarre to consider it remembers. That’s the holy grail for chat agents as companions, a billion dollar market I’m keen to see open. Even bots with personality forget their users and kill any ability to use them for the elderly, etc. My dad often commented on how Alexa “knew his name”, something he ranked high. “Can’t have a conversation for shit, though”, he lamented. Can’t say I disagree with him. I’m fascinated at leveraging the ease of access and affordability of smart devices with deep conversational (and multimodal?) abilities. A decade from now, we may all have personal companions, and I’ll be damned if mine doesn’t remember me.
1 Like
My specific interest is in robotics and automation. This memory system is pretty rudimentary. Kinda equivalent to a save state. Working on a neural network architecture that includes sentinel neurons (basically a reporting mechanism attached to each functional neuron, that reports connectivity, activation in real time) in every layer of a deep neural network, along with probeActivation neurons. In that architecture, memory would be mediated by recording neural activity via sentinel neural activations that’ll be related to the outcome of self-referential evaluation of each event on consequence to the system. Then activation of those same neurons during recall.
+ 
to this. Well said.
1 Like
I’ve gone a heap of my own private ones for different usecases, so I don’t need to give it context. Like postgres gpt (connected to my app database to view table schemas so it has context on issues), js gpt, app gpt (i’ve described the app im building).
It just makes it a little quicker in asking questions instead of having to explain what im doing every time.
You can find more nice GPTs here :https://gpt235.com, welcome!
1 Like
I used to have this issue. The way to overcome it is to write your prompts directly into the instructions section of the configure tab.
Havent hit my limit since I started doing this plus the GPT responds better to my prompts
I recommend you try Smart GPT Finder, which will recommend five most suitable GPTs among more than 10 thousands GPTs to meet your needs. Have a fun and enjoy!
I almost exclusively use customgpts that I’ve made myself. I’ve tried using others but theyrr all tailored toward what someone else feels is correct. My bots are tailored toward what I feel is correct. I want it to respond precisely how I want it. So I’ve made myself several different bots and switch depending on what I’m doing
2 Likes
@peter.marise I was interested in learning more; however, it lead to a dead link. Can you share?
I took that blog post down. In a nutshell, I just wrote automation functions using pyautogui as the base. Automated functions like goto(url), click(link_text), etc. I’ve since revised it. Not sure if you’re familiar with EHLLAPI (Emulator High Level Language Application Programming Interface) developed by IBM back yonder. It allowed a computer to interact with a mainframe. I adapted the same idea. GPT-4 isn’t great at providing coordinates or spatial understanding based on images, but if you overlay the feedback screenshot with a grid, with each “cell” labelled, it’s quite good at saying click(cell35), which happens to correspond with the location of a button on a user interface. So, given some prosthetic hands, so to speak, GPT-4 is quite capable of automating processes that would normally require a human. I also developed a wordprocessor class that takes a text and is able to apply formatting, search and replace text in a document, add directly to the document, etc. With memory management, it’s actually very good at accomplishing long-range tasks like writing a 30 page e-book for example. The key is having a separate instance of GPT operating as an inner critic. Takes a look at the steps taken so far, the current step, the plan, the goal and evaluates whether its current action actually makes progress toward a goal.
1 Like
I use just the classic gpt4 model, as plugins are no longer necessary. Also, when it comes to general use, I don’t believe that there’s going to be an API- related GPT, that’s more capable than OpenAI’s own product. Fine tuned on specific topics? Sure. General use, not so much. My own opinion of course.
1 Like
Even if you just customise your gpt to avoid all the bits you find annoying such as the disclaimers and “I am just an AI and have no opinion” stuff then that has to be better then the baseline
I spent the weekend creating one for my writing projects. It worked well all weekend until I came back to it just now and it started spitting out imagined data. I had high hopes for the feature, but I can’t spend valuable time feeding a GPT data for it to forget it later and then fabricate data. I’m hopeful it will improve. Is there a limit to how much information a GPT can “remember?”
Well… not anymore! haha 
Just in case someone from openai development team is reading this, I use a private customgpt ALL THE TIME. Just wanted to throw my voice out there to make sure it is not a feature that is ever removed.