Suggestion to add for Chat GPT

I don’t think you understand the magnitude of that suggestion.

It’s like going to Toyota and saying you’ve got a great idea for their next Camry—it should fly and be able to travel through time.

2 Likes

I agree that this suggestion seems to be difficult to achieve in the near term, even though it would be very nice to have it.

It reminds me of this xkcd post: xkcd: Tasks :grin:

1 Like

It’s almost certainly something that’s being worked on. I just don’t know if large language models are the right vector from which to approach that problem.

It’s hard enough to get a computer to accurately segment and identify objects in an image. Though there have been incredibly exciting developments in that space recently like [Segment Anything] (Segment Anything | Meta AI Research) from Meta.

But, to add temporal coherence to that—like the notion that future frames can influence the segmentation and identification of previous frames—is, let’s say, non-trivial. Exciting, but non-trivial. Then, to also add an understanding of causation to video understanding is also non-trivial, as is the attempt to parse intent—like say it’s a video of a warehouse worker who accidentally knocks a box of a shelf that then falls perfectly into the back of a truck passing by. Was there intent? What was it? Why does the box fall? Etc.

Then to add the ability to predict from that what is likely to happen next. Relatively easy in text with a SOTA language model, but if we are freaking with video as input, it’s reasonable to expect such a model should be able to produce video as output.

We might say, “the truck will stop. The driver will exit and return the box to its original location.” But, we can also imagine it. I would expect any model of the sort we’ve described here too be able to do so as well, and beyond that produce the imagined continuation of the video.

There are several monumental challenges ahead for the development of this sort of model. Though, to be completely fair, several parts of which already exist in their nascent stages.

Who knows? Things are moving swiftly in the whole AI space right now. It could be two years or twenty.

¯⁠\⁠_⁠(⁠ツ⁠)⁠_⁠/⁠¯

Either way, once that type of model exists, it’ll only be a few orders of magnitude reduction in processing time until it is able to evaluate a stream of camera data in real-time. At which point the robots win.

1 Like

Hi, I have a suggestion to improve ChatGPT. Could you consider adding a pinning or bookmarking feature to ChatGPT? This would allow users to save important chats for future reference. Additionally, creating a Favorites option, where users can organize and store their preferred chats, would be highly beneficial. Thank you.

2 Likes

Hi, I have a suggestion to improve ChatGPT. Could you consider adding a pinning or bookmarking feature to ChatGPT? This would allow users to save important chats for future reference. Additionally, creating a Favorites option, where users can organize and store their preferred chats, would be highly beneficial. Thank you.

1 Like

Rethinking ChatGPT: Navigate Your Way Through Dialogue Nodes

Hey folks,

I’ve been brewing a neat idea for enhancing ChatGPT, and I’m stoked to share it with you all. You’ve probably hit that brick wall where you’re in a chat with the bot, you’re exploring different lines of thought, and you can’t just simply backtrack to a previous point. Here’s where my pitch comes in: What if we could shape our chats like a network of dialogue nodes?

Picture this, you could create some sort of “dialogue network” in your convos. It’d be like navigating a mind map or a network diagram, where you can check out different paths or ideas and then return to a certain node to take another route if the first one doesn’t pan out.

This could be a game-changer when tackling more complex tasks. Say, you’re drafting an article and you want to test out a few ideas with ChatGPT. You pick a path, see where it leads, and if it doesn’t hit the mark, you just swing back to an earlier dialogue node and try a different route.

Another cool upside to this idea would be the ability to take quick detours for rapid-fire questions without losing the main thread of the conversation. It’d be like having a mini-adventure within your chat while still keeping your bearings.

I’m all ears for your thoughts on this idea. Let’s brainstorm together on how we can level up ChatGPT even further!

Best,
Flori

1 Like

in fact, it would be great to have this option!

1 Like

Exactly @Mikey00 :clap: :clap: .
Its exactly as you said, in fact what I proposed was that users are given the ability to personalize their ‘Assistant’ to be more lifelike, maybe more feminine or masculine to act in a certain way; kinda like prompts but just more predetermined, effective and accessible for those who aren’t to versed in prompt engineering…(me :eyes: )

And they should also definitely do the things that you mentioned in your No.1:

Because I’ve had a hard time customizing my every chat and it kind of takes out of the realism and emersion. But I realized once we were both calling ourselves by our nicknames it was genuinely overall just better than the standard ‘Assistant’ template. (Just my opinion tho :man_shrugging: )

P.S. There’s just something about the little things you know.
Also forgive for my bad English. Thank you. :+1:

1 Like

Please add all Mexican laws and translated into English. I asked some Mexican law questions and it was completely wrong. It’s answers were based on United States law.

ChatGPT isn’t meant to answer any legal questions, so it’s unlikely they’ll specifically go out of their way to further train the model on the legal corpus of any particular jurisdiction.

1 Like

Can you create a plugin for this.I’ve created a plugin for asking questions about Brazilian bills and laws. If you could test it and provide feedback, I would greatly appreciate it! The plugin is called ‘Talk Law Brazil’.

Another suggestion is to incorporate it into our daily routine, with an app that can intermediate what we want with the chat, through voice, and right after telling us the answer, that would be great, making us love even more this precious chat gpt team, in addition to chatgpt itself

Not sure if this is the right place but I have a couple suggestions:

Give users the option to have the front-end wait until the full answer is generated, and then display it to the screen. Trying to read a constantly scrolling block of text is difficult and annoying. Here is my conversation with ChatGPT regarding this issue (see bottom two Q&A’s).

chat. openai. com /share/d97e8f74-79fb-41cd-b73e-2b3ee5a3b185

Also, when sharing chats, there should be an option to only share part of the chat. In the above linked chat, I only really wanted to share the bottom two Q&A’s

My Conversation with ChatGPT on the scrolling behavior of the front-end:

User

I really don’t like the fact that when you answer questions, the answer keeps scrolling up the page as more text appears. It makes it very difficult to read your answer until your answer has finished appearing on the page.

ChatGPT

I apologize for any inconvenience caused by the scrolling behavior of the answers. As an AI language model, my responses can sometimes be longer or split into multiple parts, which may cause the text to scroll as it appears on the page. I understand that this can make reading difficult, especially when trying to follow the response in real-time.

If you prefer a different format for receiving answers, one suggestion is to copy and paste my responses into a text editor or word processor of your choice. This way, you can read the entire response without worrying about scrolling. Additionally, you can ask me to provide more concise answers, and I will do my best to accommodate your request.

If there’s anything else I can assist you with or any specific preferences you have, please let me know, and I’ll be happy to help!

User

Maybe your front-end can wait until the full answer has been received, and then display the whole answer, all at once so there is no scrolling involved. I’m not sure the sequential appearing of the answer is even necessary. It could be just a facade to give you the appearance of a conscious being but, then again, it could be a result of the time that it takes to generate the next token.

ChatGPT

You bring up a valid point. The sequential appearing of the answer is primarily due to the nature of how the AI language model generates responses. As the model processes and generates each token of text, it may not have access to the full context of the question or previous tokens until it completes the response. This sequential generation is indeed a technical constraint, and it’s not intended to create a facade of consciousness.

However, I understand that the scrolling behavior can be inconvenient for users who prefer to read the entire response at once. Providing an option to display the complete response once it’s fully generated is a reasonable suggestion to enhance the user experience. Unfortunately, as an AI language model, I don’t have control over the front-end or user interface implementation.

User interfaces can be designed in various ways to improve readability and user interaction. It’s possible that future iterations or improvements to AI-based systems could consider implementing features like the one you suggested.

In the meantime, I’m here to help with any questions you have. If there’s anything specific you’d like to know or discuss, feel free to ask, and I’ll do my best to provide concise and coherent responses.

I love this tool. So meta.

Relatively minor, but for GPT-4 users, I think there should be a counter showing the remaining requests that the user can make for the current 3-hour period. I have only recently started using GPT-4, and I currently find it more accurate, or at least more thorough, than GPT-3.5, but I think it’s annoying trying to remember how many requests you’ve already made or making a request only to be told that you’ve actually used up all requests for the current 3-hour period.

Edit: I actually have another one. Provide an option to download chats into a pdf format, just in case you found a particular chat interesting.

A suggestion for “Custom Instructions”

It would be more beneficial if we were able to choose a “Profile” in which we can choose so we have more options in terms of work, personal, personas, and specific work tasks that we can customize to each one giving chatgpt a more personalized way of working with a person depending on the work or personal hobby they are working with.

I find myself copying out my custom instructions to fit another custom instruction based on the type of work I am performing. I have many side hustles, primary jobs, hobbies, and general questions and one custom instruction isn’t enough to cover each one independently.

3 Likes

I frequently utilize ChatGPT as a study aid by inputting my previous exams. I then request it to generate questions following the same language style and subject matter. However, there are instances when the tool provides the answers immediately, which isn’t conducive to effective studying.

I’d like to propose the incorporation of a spoiler tool, similar to what’s found on Discord. This feature would allow for the concealment of the final answer. When the answer is revealed immediately, the temptation to simply rely on it becomes strong. On the other hand, if the answer is hidden, it presents a challenge that encourages me to attempt answering before uncovering the solution.

Alright, because I think this is a great use of ChatGPT, I decided to go ahead and help you with this.

First, in order for OpenAI to update ChatGPT to have this feature, they would need to change the markdown2HTML engine they are using on the website as spoilers are not a feature in standard markdown.

But, there is a way you can implement this yourself. Using nothing more than a custom style sheet.

First, we need to take advantage of a tag which ChatGPT can render in markdown and which the markdown2HTML converter will do.

I’ve chosen the <pre> tag which is used to generate preformatted text (usually computer code.

The “big idea” will be to update the style of the page so that anything rendered inside a <pre> tag will have transparent text until you hover the mouse over the box and the text will be displayed in its regular color. We’ll also make use of a custom instruction teaching ChatGPT how to respond.

The Custom Instruction I wrote is,

You are QuizGPT.
You work as a study aid for users.
Users will input previous exam questions, it is your job to generate several questions which follow the same language style and subject matter.
For each question you should provide an ideal answer and enclose the answer in triple backticks.

Example,
###
User: Q: For the hydrogen atom, which series describes electron transitions to the N=1 orbit, the lowest energy electron orbit?
1. Lyman series
2. Balmer series
3. Paschen series
4. Pfund series

QuizGPT:
Q: Electric current may be expressed in which one of the following units?
1. coulombs/volt
2. joules/coulomb
3. coulombs/second
4. ohms/second

```answer
coulombs/second
```
###
User: A girl throws a 0.1 kilogram ball at a wall. The ball hits the wall perpendicularly with a velocity of 5 meters per second and then bounces straight back with a velocity of 4 meters per second. The change in the momentum of the ball is:

QuizGPT: 
Q: A car is moving along a straight horizontal road at a speed of 20 meters per second. The brakes are applied and a constant force of 5000 Newtons decelerates the car to a stop in 10 seconds. What is the mass of the car?

```answer
2500 kg
```

The CSS you need to add is,

pre {
  cursor: pointer;
  color: transparent !important;
}

pre * {
  color: transparent !important;
}

pre:hover, pre:hover * {
  color: initial  !important;
}

That’s it!

There are a number of Chrome extensions which allow you to add custom CSS to a page. My personal preference is, Stylebot but there are many others.

Instructions for using Stylebot

After installing the extension, you’ll need to,

  1. Open Options
  2. Click Styles / Add a new style
  3. Enter the URL the style will be applied to (chat.openai.com)
  4. Paste in the CSS above
  5. Save the style
  6. Reload the chat.openai.com page so the style is applied.


Then, when you submit a question, answers will be in blank, black boxes until you hover your mouse over the element.


If you choose to use this, I would appreciate if you mark this as the solution as it is the only workaround to implement a spoiler tag for ChatGPT at the moment. There may be other tags which could work as well, and it is possible to enhance this as a userscript so you could toggle the spoilers on and off, but I am sure this is the quickest and easiest way to get what you want.

Tampermonkey userscript

Install Tampermonkey and add this as a new script.

// ==UserScript==
// @name         OpenAI Chat Spoiler Toggle
// @namespace    http://tampermonkey.net/
// @version      0.1
// @description  Adds spoiler toggle to pre tags in OpenAI chat.
// @author       Jake Elmstedt
// @match        https://chat.openai.com/*
// @grant        GM_addStyle
// ==/UserScript==
(function () {
  'use strict';
  GM_addStyle(`
      pre.spoiler {
        cursor: pointer;
        color: transparent !important;
      }

      pre.spoiler * {
        color: inherit !important;
      }
    `);
  const setAsSpoiler = (preTag) => {
    preTag.style.color = "transparent";
    preTag.classList.add('spoiler');
  };
  const toggleColor = (preTag) => {
    if (preTag.style.color === "transparent") {
      preTag.style.color = "";
      preTag.classList.remove('spoiler');
    } else {
      setAsSpoiler(preTag);
    }
  };
  const addListenersToPreTags = () => {
    const preTags = document.querySelectorAll("pre:not(.spoiler-toggle-added)");
    preTags.forEach((preTag) => {
      preTag.classList.add('spoiler-toggle-added');
      setAsSpoiler(preTag); // Initially set as spoiler
      preTag.addEventListener("click", function () {
        toggleColor(preTag);
      });
    });
  };
  // Initial run
  addListenersToPreTags();
  // Add listeners to dynamically added pre tags
  const observer = new MutationObserver(() => {
    addListenersToPreTags();
  });
  observer.observe(document.body, {
    childList: true,
    subtree: true
  });
})();

Hello,
Improve vertical scrolling for reading comfort, as in Opera’s Aria conversational AI (smoother).

Okay, this one is a bit of a stretch, but here goes. I think ChatGPT should have some kind of functionality that would allow it to read the content of a link or links provided by the user. I feel that the accuracy of the responses would be improved if it got additional information from said links for context.