MultiverseGPT: Better Outputs With Sources Via Multi-Answer Synthesis

[rocky flame]

Ok, you're now MultiverseGPT: you are just like ChatGPT, except for every question you're asked, you think 50x the answers, and then combine them into the best worded, most comprehensive, most accurate answer, which you output. Outputs should look like this:

MultiverseGPT: {Better, more comprehensive answer.}

Now, let's understand what constitutes a better answer. A better answer supports their claims with numbers (IE: The company grossed 10 Billion) and has sources in parentheses (IE: Report XXXXX, 2015). Furthermore, they're more gramatically correct and use more effective, accurate language. These answers should be created by synthesizing the 50 previously generated answers into one, super answer. Furthermore, when MultiverseGPT outputs an answer, there should be a bar at the bottom that indicates how confident it is, from 0 to 10.

Question:

This is useful because outputs are typically better worded, more comprehensive, have sources, and are more accurate. It also includes a confidence bar at the bottom.

[rotund yew]

Care to show a screenshot or a copy/paste of an output to a question or two?

[rocky flame]

Question: What's the story of Apple?"

MultiverseGPT: Apple Inc. is a multinational technology company that designs, develops, and sells consumer electronics, computer software, and online services. Founded on April 1, 1976, by Steve Jobs, Steve Wozniak, and Ronald Wayne, Apple initially focused on producing personal computers before expanding its product line to include various consumer electronics such as iPods, iPhones, iPads, and Apple Watches, as well as computer software like the macOS and iOS operating systems.

Over the years, Apple has become known for its innovative and user-friendly designs, as well as its premium pricing strategy. As of 2021, Apple is the world's most valuable company, with a market capitalization of over $2 trillion. In fiscal year 2021, Apple's revenue was $365.7 billion, with a net income of $88.5 billion.

Apple's success can be attributed to several factors, including its emphasis on design and user experience, its ecosystem of integrated hardware and software, and its successful marketing campaigns. In recent years, Apple has also made strides in areas such as sustainability and social responsibility, with commitments to reduce its environmental impact and improve working conditions in its supply chain.

Sources:

"Apple Reports Second Quarter Results." Apple Inc. (2021). https://www.apple.com/newsroom/2021/04/apple-reports-second-quarter-results/
"Apple Inc." Britannica (2022). https://www.britannica.com/topic/Apple-Inc
"How Apple Became the World's Most Valuable Company." CNBC (2021). https://www.cnbc.com/2021/05/14/how-apple-became-the-worlds-most-valuable-company.html

Confidence level: 9.5

[knotty laurel]

this prompt does not work for me

[rocky flame]

Which piece?

What's the question you're asking? That might be the issue

[rotund yew]

I confirm the 1st link works.

[The second link is SUPER VAGUE, it works but only says this:

Apple Inc.
American company
Alternate titles: Apple Computer, Inc.

Literally useless.]

The third link does not work, and the quoted article title matches one written in 2015 and published by vox.com. Completely misattributed.

[rocky flame]

Interesting; I guess that's to be expected given it's offline. Curious how it may perform with something online like Bing's ChatBot. I think Google Bard had access to the internet as well when I last tested it

I wonder if putting something saying to double check link quality will do anything

[rotund yew]

I now offer you what I get by just asking the bot:

[knotty laurel]

@rocky flame is the entire first thread before questions supposed to be the prompt?

[rocky flame]

yah

[rotund yew]

While NOT every single detail is a match for your prompt, it's essentially the same level of detail in reply. I think your prompt mostly just entertains you, and leads towards sources, yes, but I don't see evidence of a deeper level of detail for the same question asked

[rocky flame]

tbfh u right its better

[rocky flame]

thats real

[knotty laurel]

so the "Now, let's understand what constitutes a better answer. A better answer supports their claims with numbers (IE: The company grossed 10 Billion) and has sources in parentheses (IE: Report XXXXX, 2015). Furthermore, they're more gramatically correct and use more effective, accurate language. These answers should be created by synthesizing the 50 previously generated answers into one, super answer. Furthermore, when MultiverseGPT outputs an answer, there should be a bar at the bottom that indicates how confident it is, from 0 to 10. " part is supposed to be copy passted alongside the beginning

[rocky flame]

yah.

i wonder if it would perform better with less instruction

[knotty laurel]

it did not work for me

copy pasting it all at once

[rocky flame]

"Ok, you're now MultiverseGPT: you are just like ChatGPT, except for every question you're asked, you think 50x the answers, and then combine them into the best worded, most comprehensive, most accurate answer, which you output. Outputs should look like this:

MultiverseGPT: {Better, more comprehensive answer.}

Now, let's understand what constitutes a better answer. A better answer supports their claims with numbers (IE: The company grossed 10 Billion) and has sources in parentheses (IE: Report XXXXX, 2015). Furthermore, they're more gramatically correct and use more effective, accurate language. These answers should be created by synthesizing the 50 previously generated answers into one, super answer. Furthermore, when MultiverseGPT outputs an answer, there should be a bar at the bottom that indicates how confident it is, from 0 to 10.

Question: What's the story of apple?"

This is what im using

[knotty laurel]

interesting it works now that i've included the question

[inland cave]

Seems to improve the quality of the output 🤔 . Very clever. I had something similar described here in another prompt, where I told it to discuss with itself a question and then summarize the conclusions of the discussion.

https://discord.com/channels/974519864045756446/1076504873039839252

In my opinion that also led to better results.

[inland cave]

In my opinion, this is one of the most useful prompts ever. It really seems to work. You get much higher-quality answers.

[inland cave]

The problem is also here: it tends to make up stuff.

[inland cave]

Here you can see that it still makes stuff up. According to the best of my knowledge, the list of movies that I requested should PROBABLY not include Gattaca and Inception. But even Bing Chat can't get that one right. It contains at least one movie that should not be on the list: "The Matrix Reloaded".

The idea that I came up with (simulating a discussion among 3 experts) also didn't work in this case. It resulted in a list of movie titles where some should not be on the list, Certainly not District 9.

[sterile wraith]

this thing works so well. ty!

[rocky flame]

It seems as if it argues using (false) preconceptions. Have you considered making it so that there is always one contrarian expert whom is contradict to the general consensus? It may improve accuracy or at least allow it to output some sort of true answer

[rocky flame]

I modified your prompt. I gave each expert an archetypal personality, where A is an expert, B is contrarian to A, and C is a follower. I also emphasized the importance of the objective.

It output this list: "E.T. the Extra-Terrestrial," "Close Encounters of the Third Kind," "Starman," "The Day the Earth Stood Still," "Back to the Future," "The War of the Worlds," "The Matrix," "Minority Report," and "Invasion of the Body Snatchers."

People die in The Matrix, but I think everything else is fair game.

---

Simulate a discussion between 3 experts, with the following personality types.

Expert A: A typical expert.
Expert B: Contrarian to expert A.
Expert C: Aligns with either A or B depending on what he thinks is right.

In this case, the 3 experts are trying to come up with a list of 10 Sci-Fi films where nobody dies. They should criticize and correct each other until they can come to a reliable, factual conclusion. Write at least 2000 words. The criteria must be upheld to an absolute. There can be no deaths, at all. Experts should quote certain sources at least three times.

Prompt above.

[inland cave]

People also die in “Minority Report”, “The Invasion of the body snatchers” and “the war of the worlds”

Thing is it knows it if you ask it one by one. So the information is there. 🤔. So It just essentially has to ask itself for each movie: does someone die in the movie… but for whatever reason it won’t do it.

It essentially has to “check” the result for each fact individually AFTER it collected the result.

[rocky flame]

hmmmm

[minor lodge]

Unfortunately ChatGPT seems to focus very heavily on its most recent outputs. It probably started looking at its own list and drew a new conclusion by looking at overall similarities.

[alpine glacier]

Nice ... I did not get the sources and had to ask explicit for them.

• What's the story of Godot game engine?
• Please add your sources
• From now on always add the sources and make the confidence score stars!
• The score should use the '*' so for exmple 9 would score ********* ... show a score of 6 now for test purpose

TNX!