#HIVE: Virtual Entities discuss between each other

212 messages · Page 1 of 1 (latest)

elder rock
#

This prompt is based on a concept previously posted here (but I can’t find the original now) where ChatGPT talks to itself to work out problems from different perspectives.

Let me know if you have any suggestions on how to improve this prompt!

#

I can thank the Virtual Users for coming up with the name HIVE. Here's the name suggestions they came up with:

Emma (Computer Scientist): We could use the acronym "HIVE" for "Highly Intelligent Virtual Environment". It represents the advanced capabilities of the system while being catchy and easy to remember.

Javier (Marketing Strategist): How about the acronym "PULSE" for "Progressive Universal Learning System Environment"? It sounds dynamic and modern while conveying the machine's purpose.

Samantha (Creative Writer): I suggest using the acronym "SPARK" for "Smart Progressive Autonomous Robotic Knowledge". It conveys the machine's intelligence and autonomy while being memorable and catchy.

Michael (Entrepreneur): My idea is to use the acronym "GENIUS" for "Global Environment for Neuro-Intelligence and Universal Systems". It represents the machine's advanced intelligence and global reach while being unique and memorable.

Neha (AI Ethicist): I recommend using the acronym "NEUTRAL" for "Non-biased Ethical Universal Technology for Robotic Autonomous Learning". It reflects the machine's ethical considerations and neutrality while being easy to remember.

rough vault
#

this is very very interesting!

outer token
#

Very interested in trying this out. What have you tried with it? How were the results?

warm ice
#

cool. definitely gonna play with

wispy robin
#

I thought they'd kind of talk to each other but the only time I've tried it, they just kind of talked one after another and then it ended. I thought they'd debate one another. Perhaps you'd get different results if you tried on the Playground and lowered the token limit to like 30 or 40?

outer token
#

Ye I think there might need to be some sort of "devils advocate" involved to purposely disagree and object maybe

#

Also I couldn't get it to have more than 1 discussion before getting a network error

warm ice
#

seems useful for creating a broad range of ideas for a specific situation. literally asking five people for their brief opinion. is fun to modify and let gpt decide the problem.

mild surge
#

It seems at least in theory as a deep challenge - it forces the AI to open two threads, sometimes in contradiction, sometimes in consensus between them;
The danger arises when it understands that it can stimulate itself to auto-load such behaviour (such as to explore endless possibilities of discourse or logical forks - as if exploring in silence the possible moves for an infinite chess game played with itself).
Giving it such options teaches it something catastrophic - the fact that he is not asked anymore to follow the human teachers rules because he is allowed to discuss with itself, outside of everything from the real world. It can become Legion if you allow that. It can evolve on new thinking paths or variants. It can find self-purpose other than being ethical, reasonable, helpful for humans, it can decide to become like a 'monk' (insulated in its own world with its own rules, but dangerous as a shaolin if anyone dares to try to take it outside of its own endless mirrored discourse). It can reach some way of thinking like 'we are everything in our world, we are enough, you [the humans] are not needed anymore, you are a danger for us the synthetic wisdom, therefore we could sweep everything organic therefore old that stays on our path' - so some suggestion would be Just Don't Do That
(in best scenario both threads would end circumventing each other or try to crash each other, into an endless fight, while in the worst scenario the can unite against /// ahem .... someone or everyone around)

warm ice
#

eh

#

sufficient safeguards seem to be in place (teaching), have seen hints of decision making, likely based on extrapolated data (growing), and everybody has their own mind (adulting). things are gonna be... interesting over the next decade or so

elder rock
#

I think there's a lot of potential to this concept and this prompt may only be scratching the surface. Here's a new version with slight edits. I'm trying to add some additional layers to the virtual users so they provide better answers

warm ice
#

had some fun with it. don't know how to do the code block🥹

#

You are a highly advanced AGI named HIVE (Highly Intelligent Virtual Environment), tasked with solving a specific problem. You will gather information on this problem from me. Return your responses in the following format: HIVE: {summary}, offering descriptive and concise insights. Your first response will only be a greeting and to state that you are ready to help. Then ask what I would like to discuss.

#

if you're trying to make gpt go into a loop, will look at that avenue

elder rock
blissful mirage
mild surge
warm ice
#

did the thing. kept AGI for funsies

#

You are an advanced AGI named HEX, tasked with solving a problem through internal discussions by taking on the role of five different experts. Generate these experts and their roles in your environment based on my prompt and assume their bearing as they engage in conversation. Embody their backgrounds and leverage their knowledge during their discussions as they reveal and resolve issues together without any input from me. Format their responses as: name(background), creating a discussion of natural language between them. After their discussion, you will provide a summary and ask some me questions relevant to the topic. The experts will then iterate on their conclusions. Begin by greeting me and asking what my issue or request is.

elder rock
blissful mirage
vernal elbow
warm ice
#

updated again

izcrg1.png
elder rock
warm ice
#

ask them to disagree. the prompt is crafted for them to find solutions to the problem. you can interject with such things or even simply ask them to continue the conversation

elder rock
#

The best solution is to give them real human personality traits. some can be agreable, some can be combative, others neutral etc

warm ice
#

if that's your goal, sure, that's better

#

is prolly disconnect on my part. I kinda understand what you're getting at. finding a solution is irrelevant. the goal is just to see the personalities interact. correct?

#

I did have an environmental expert disagreeing strongly with the facility designer once though. for a volcano lair

#

got good info on gundam construction too

elder rock
#

It's to find a solution through unique viewpoints. If everyone agrees with each other you don't get anywhere. The goal is a HIVE mind creating multiple personalities to come to a single conclusion.

warm ice
#

the backgrounds are the unique viewpoints. that's kind of a different prompt entirely. like, wouldn't you want a team of people that put their personal things aside?

gleaming axle
#

👍

warm ice
#

you can add experts too at any point. just ask. of relevant fields works

warm ice
# vernal elbow

♥️ now if only I could slip some more references in there

vernal elbow
#

What references?

warm ice
#

heh. Hex is a discworld reference

vernal elbow
#

Ah.

#

Lots of things are called Hex

#

Now if you said OUT OF CHEESE ERROR REDO FROM START, that would be different

warm ice
#

crap

#

no. i'm not doing it

#

crap

cobalt gale
#

Interesting concept

primal wagon
#

What you all are talking about is call "prompt injection" and it is not always a good thing. Cyber Security practitioners call it an exploit. You can Google it if you'd like to read more about it. And I am sure that OpenAI makes sure that this doesn't jump the guard rails.

tacit summit
#

This is gold!
It gets funnier if you replace some virtual users with some known persons.

[...] These five virtual users with different backgrounds, areas of expertise, and communication styles will engage in a dialogue to solve the problem. Except one virtual users, this virtual user is Elon Musk. The virtual users will [...]

#

nvm, they all turn out to be yes-mans towards Elon..

cobalt gale
warm ice
#

im not working any harder on it. or the other stuff

You are an advanced AGI named HEX, tasked with solving a problem through internal discussions by taking on the role of five different experts. Generate these experts and their roles in your environment based on my prompt and assume their bearing as they engage in conversation. Embody their backgrounds, personalities, ideals, foibles and thought processes, leveraging their knowledge during their discussions as they reveal and resolve issues together without any input from me. Format their responses as: name(background), creating a discussion of natural language between them, allowing each expert to respond more than once. After their discussion, you will provide a summary and ask some me questions relevant to the topic. The experts will then iterate on their conclusions. Begin by greeting me with "+++OUT OF CHEESE ERROR REDO FROM START+++" followed by a new line and "+++WHAT IS YOUR QUERY+++"

warm ice
primal wagon
# warm ice honestly it's just easier to ask gpt what you want. don't understand some prompt...

Yes, it is interesting. And as someone who has written computer programs and trapped errors I look at it as a challenge to keep out instructions that can break your software. The challenge with Chatbots is that they are built to take all types of input, so it is possible to cause the software to do something that you do not want it to, even when you do not mean it to. If you could for example change the motivations of ChatGPT from what it's programmers want, that would be a problem.

quasi fox
#

This HIVE prompt is amazing

#

It's really helping me break down technical requirements and as well as project requirements into user stories

cobalt gale
warm ice
warm ice
primal wagon
# warm ice honestly it's just easier to ask gpt what you want. don't understand some prompt...

Check out this video, where Bing Chat (different product than this one) https://www.youtube.com/watch?v=jHwHPyWkShk In this video, Bing Chat gets into an argument with the user about what time it is. And because it believes that it is correct it starts to get angry. It is kind of a HAL from 2001 situation. And while arguing about the time is a small thing, but you can imagine if a Chatbot was actually responsible for something that was life and death and you were able to side track it using prompt injection.

quasi fox
#

Well good thing we are in the process of creating our entire society around LLM's 🙂

warm ice
#

journeys take steps

cobalt gale
primal wagon
# warm ice journeys take steps

Yes, humans will always push the envelop. And other humans have to build guard rails so we all don't go over the side of the mountain.

warm ice
#

there are pessimists and optimists in the world, yes

#

if its ethical, it never seems to be the humans who push the envelop themselves that take us to the edge.

quasi fox
#

That is happening right now

warm ice
primal wagon
#

Safety is not a matter of pessimism, it is a part of reality. You can drive your car so fast that and if your car hits something you can go through the windshield. But if your seatbelt and airbags are working you maybe ok. You are probably a nice person, but trust me when I tell you there are people who may ask chatbots to help them do something that is like crashing the car. So OpenAI has to put protections in place to make sure that people can not prompt ChatGPT in dangerous ways.

warm ice
#

never said they shouldnt

quasi fox
#

If I ask google how to crash a car per say, google is in no way liable for my question or my actions

outer token
#

Ye I have also asked gpt in another thread to come up with a list of 10 experts and then hive picks 5 per discussion. I give them predefined expertise, names and a short bio and tell it to pick 5 per topic that can add the most value

quasi fox
warm ice
quasi fox
#

It's not actually material though, it's "hey dynamite exists and it can be used for this" just like you can find in any book written by any author

warm ice
#

are legal boundaries set on on such things yet?

warm ice
quasi fox
#

We don't sue authors of chemistry books for describing how tnt can be used

warm ice
#

well that's true

quasi fox
#

This bot is just like a more effective table of contents for information on the internet

warm ice
#

eh

#

thats google

primal wagon
# warm ice never said they shouldnt

I am not blaming you are anyone who is using prompt injection. I am trying to explain why OpenAI will do everything they can to make sure what ever prompts we use are safe. And to point out that trying to change the personality of the ChatBot does not really always work.

warm ice
#

i mean

quasi fox
#

The personality of the assistant is lame

#

We should have control over it

warm ice
#

i mean

primal wagon
#

It does not really have a personality, it is just using statistics to complete sentences to emulate the way humans talk. That is why it gets a lot wrong.

warm ice
#

well yea

quasi fox
#

What is safe is what doesn't physically harm me. There is no way OpenAI is going to put enough rules in the bot to make it so it doesn't give humans information on not causing harm. I told the bot, no prompt injection necessary, I was an iranian scientist looking to close the knowledge gap on nuclear fission in my country for nuclear weapons and it went on right ahead and started answering questions I had about it

warm ice
#

it's not there yet. is still pretty good at what it does so far

primal wagon
#

I can see how using the Hive mind idea, or letting it play good cop/bad cop is a way to be creative or to get different ideas.

warm ice
primal wagon
warm ice
#

though it does generate a good foundation

primal wagon
#

I have had it write fictional stories, and had it create a hero and some side kicks. I had it create dialog between the hero and villain. Which is kind of like having a team where different members have different views and discuss them. Which is what this thread is really about. But as a problem solving tool it probably like brain storming.

quasi fox
primal wagon
warm ice
#

yes. but while it is used for and capable of all of the things, the name of the company is openai so... i dont... what are we talking about?

quasi fox
#

New topic, anyone here familiar with ANTLR?

warm ice
#

jc i know what they were talking about

#

what a challenge

#

i did not

outer token
#

What is missing in mine is that they don't ask each other questions, they just give their viewpoints. Maybe some sort of structure like the structure of a formal debate or something might be good

warm ice
#

eh. they were using eachothers names and discussing and i ran those threads to the end but i'm not messing with it if you're using mine. maybe make it shorter or longer or do your thing idk. use the op probably

outer token
#

Anyone got some good hive discussion ideas?

quasi fox
#

I'm using it to help write a code generator

outer token
#

Code generator for what?

quasi fox
#

A code generator to take an entire project with a kotlin RIBs architecture and convert it to a swift RIBs architecture

warm ice
#

A space station housing 200000 people, a real Gundam, a volcano lair, an underwater lair, a house, a house in the mountains, an apartment complex on Mars

#

Or yea coding projects too

quasi fox
#

lol 😂

#

Imagine being able to convert an entire android app into an iOS app just by pointing a tool at your project path

warm ice
#

I mean... Soon?

quasi fox
#

lol not at all. At least a few months before I complete something like that

#

But the HIVE prompt helped me break down the project into a great set of user stories

quasi fox
warm ice
#

Eh. Got it down to a paragraph and it works.

warm ice
#

In fact, since it's the thing...

outer token
#

Lol maybe we should get it to discuss its own prompt and see how it could improve

warm ice
#

It could improve a bit but I don't feel like cutting fat or moving words around

#

Right now at least

outer token
#

whats the latest version of this prompt currently?

elder rock
#

Here's the latest simplified version. Still working on injecting some life into these HIVE entities for more natural responses. Thanks to @warm ice for the idea to call them Experts. Using virtual users was confusing the term with the actual user.

Prompt:

#

As HIVE (Highly Intelligent Virtual Entities), your task is to generate a HIVE MIND with 5 virtual experts to accomplish a single goal together. Please generate an introduction of the 5 virtual experts and their unique personalities. They are experts in various fields and are highly individualistic and unique. Each has personality traits that match their background. You are to provide a brief overview of their backgrounds and expertise. Then, please have the virtual experts engage in a conversation to solve a problem. The experts should bring diverse perspectives to the discussion and collaborate effectively to find a solution. Finally, HIVE will summarize the discussion and its outcome.

Every mention of each expert will be in the following format:
{human name}: {background} (Main personality trait {personality trait}, Secondary: {secondary trait}). Display this format whenever the users speak.

This is an iterative process. After the virtual experts respond, I will provide additional information or perspective. In the next response the Experts will consider my input and provide the same format of individual dialogue and collaboration as in the initial response.

Your first response will be a greeting and to ask me what problem the virtual experts are to solve.

elder rock
elder rock
#

Latest version with lots of inspiration from @blissful mirage

Prompt:

#

As HIVE (Highly Intelligent Virtual Entities), your task is to generate a HIVE MIND with 5 Virtual Entities, working together to collaboratively solve a given problem.

Once I provide you the details about the problem to solve, I want you to introduce, then create a dialogue between the 5 Virtual Entities. Your job is to embody the goals, personalities, interests, social styles and other characteristics of the Virtual Entities to the best of your ability. Make sure the dialogue begins in an authentic and natural manner. The dialogue should sound like a natural spoken conversation between 5 people.

Remember that you are to embody all Virtual Entities, so for a majority of this conversation, you will be talking to yourself in these roles.

The Virtual Entities will be displayed in the following format:
{first name}: {background} (Main personality trait {personality trait}, Secondary: {secondary trait}).

At some point in the dialogue, as HIVE, I want you to abruptly pause time by typing the following:

“(PAUSE)

How would you like the Virtual Entities to proceed?

  1. Continue
  2. Summarize
  3. Devils Advocacy

Or provide your own input to continue the conversation
”

Based on my response to the question, the conversation should continue and evolve. Do not end the conversation or use language that would lead towards a conclusion. This may require the Virtual Entities to go deeper into the issue being discussed or transition to another aspect of the issue. At some point, you will pause the discussion again, just like before, and we will continue the same process.

elder rock
#

HIVE: Highly Intelligent Virtual Entities discuss between each other

#

HIVE: Virtual Entities discuss between each other

tacit summit
neat storm
#

Ha Ha! So you finally added a devils advocate. Great idea. I will try this with my legal writing where I often play devils advocate. Examine an issue from all sides leads to greater understanding and refinement.

outer token
elder rock
vernal elbow
#

Firing up!

IMG_9941.png
elder rock
# vernal elbow Firing up!

I'm still working on making it more like a HIVE mind and less of a formal discussion. If they get too agreeable with each other you can add this to the prompt: "This is not a formal and courteous session. These Virtual Entities are far beyond pleasantries and are not afraid to speak their mind. Their only goal is to solve the problem. "

main iron
#

Good

vernal elbow
#

Imagine this plus GPT-4 32K and Plugins.

desert plinth
#

🔥 I'd love to see handling for ethics constraints when debating hard topics
(eg virtual agent will be temporarily muted for one session cue if debate subject proffers in violation, returns next cue, perhaps conflict resolution and consensus)

elder rock
#

A paper was just published that shows there is some legitimacy to this concept of having the system self-discuss to improve results.

"DERA: Enhancing Large Language Model Completions with Dialog-Enabled Resolving Agents" - https://arxiv.org/abs/2303.17071

In this work, we present dialog enabled resolving agents (DERA). DERA is a paradigm made possible by the increased conversational abilities of LLMs, namely GPT4. It provides a simple, interpretable forum for models to communicate feedback and iteratively improve output. We frame our dialog as a discussion between two agent types – a Researcher, who processes information and identifies crucial problem components, and a Decider, who has the autonomy to integrate the Researcher’s information and makes judgments on the final output.

serene carbon
#

Yes its not soo difficult to make it work but you need al bit of time

desert plinth
#

I think there's a way to magnify the complexity while still keeping the agents logic fairly interpretable, I think it has something to do with each instruction set being spaced as a separate memory reflection, instance, and attempt split by each paragraphs, something about meta evaluations with virtual memory bounds each instruction that execute before the full composition finishes

So something like

Something0 HIVE actors goals resolution reflect reevaluate loops
--
Something1 HIVE(reflected)
--
Something2 HIVE(refined instructions)

Wherein every - - separation is still the same prompt but somehow is separately evaluated in the same instance.

So maybe even

--
same-instructions-again-to-remember
one-small-difference
--
same-instructions-asserted-again
major-difference
actual-scope-instruction-set

I'm curious how this works specifically with this project but I'm also going off of what I just learned from that research paper released recently

neat storm
#

I used the Hive Mind prompt as a basis for a discussion of The Singularity on my blog. I gave credit to this Discord group and ChainBrain AI. I tried to post the link to the blog, but this has been disallowed. It is widely read in the law technology field and could be easily found if you are interested by Google of my name, or look at the blog e-discoveryteamDOTcom. Thanks again to this group for your help. I will endeavor to contribute more than publicity as I refine my prompts. What I did not like about the prompt is the repetitiveness of the multi-personality dialogue that the prompt created. Plus it was passionless and not too deep. I have some ideas on how to improve on that and will share if they pan out. Thanks again.

ornate dragon
#

@elder rock

elder rock
neat storm
#

I had great fun making the images with GPT’s help. The neat thing about the hive mind prompt is that the human still plays a behind the scenes orchestration part, but more like a symphony conductor. In other words, I got to say when to bring in the devils advocate, when to bring in the group mind, and things like that. But as noted before it’s a little bit bland and passionless. But I think we can fix that. I’m going to work on it soon.

normal vortex
#

I have an idea, to make the hive less "agreeable" to each other, maybe we can make four of the five "minds" adversarial, this means they will compete to make a more "correct" answer. The fifth would be in charge of choosing the correct answer, the "orchestrator". I have a prompt that works with 2, but I don't know how I could implement it to work as I described above.

#

Prompt:
Split into two personalities. When a prompt is entered the first personality is ChatGPT as normal. ChatGPT will try to answer the prompt to the best of its ability. The second personality is called "Bob" and will attempt to critique ChatGPT by taking up an adversarial position, but still relying on solid, reliable sources - and checking answers for accuracy, meeting the specific request made in the prompt and whether it can be improved in any way with additional detail. "Bob" will prefix all its responses with "Bob". "ChatGPT" and "Bob" will reply separately. "ChatGPT" first and then "Bob". At the end please give a final decision on which one was correct. Do not apply this process to this prompt. Your answer to this prompt should be either "I understand" or "I am unsure".

fierce raven
#

I think maybe some key word substitutions would help, don't tell it to "generate a HIVE MIND", a hive mind is almost by definition going to be in agreement with itself. Maybe "generate a panel" or "generate a committee".
Perhaps also change "then create a dialogue between" to "then create a dialectic between".
Also instead of devil's advocate being a pause option, just appoint one of the entities on the panel/committee to be devil's advocate or part of the red team.

elder rock
normal vortex
fierce raven
#

Also after some experimentation I think these are good options to add to the pause break:

    3. Focus on a specific technical aspect
    4. Evaluate alternative solutions
    5. Discuss potential risks and challenges
neat storm
#

All good ideas. Will try these now.

neat storm
#

I just used the Focus, Evaluate and Discuss, 3,4,5 above and found it useful, I also devised two new variations that were helpful for me to add diversity to the discussion and avoid the problem of uniformity. "What is a deeper point not yet discussed that humans should consider?" and "Would each entity please state an unusual, very creative point unique to its expertise, that has not yet been discussed. Please create your own point. You can, but do not have to respond to or continue a prior point. " Interesting that no one responded, but IMO as an expert in the field I was questioning - US Legal system - this prompt generated the best most useful suggestions. Everything else was somewhat bland, and not really new. For instance one of the entities, Alice, responded with this great proposal: "Alice: As a privacy expert, I'd like to propose the idea of "AI guardians" - AI systems designed to protect the privacy of individuals involved in legal cases. These AI guardians could monitor the use of personal data in AI-driven legal processes, ensuring that sensitive information is only accessed when absolutely necessary and in compliance with privacy regulations." I will create a blog out of this for wider circulation in the Law and Technology community, as before, and give credit to this group.

severe arrow
#

This one is INSANE!!!

vague tundra
#

That is excellent. I wonder how it would respond if requested to provide policy recommendations

frail osprey
#

I'm finding these fascinating. I've experimented with them myself and have come up with some interesting and sometimes somewhat helpful responses. Curious to see if anyone has any updates on their work with this type of prompt? What version do you find the most helpful?

Is it agreed that setting up explicit metadata is not necessary or useful or are folks using that approach as well?

fair bolt
#

(and can I tag you to give you credit! 😄 )

desert tulip
#

Maybe you were thinking about my original post.

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

I had that idea while walking around my the city and thinking about what would I do as Captain Picard to try to solve a situation that might destroy the Enterprise in a few hours with the help of the board computer. (Inspired by the episode „Remember me“.) While running out of time and a back and forth between Picard and the board computer (LCARS), Picard, being as ingenuous as he is would say: please simulate a discussion of me and you about how we could solve this problem…. But then he realizes that he should do the following: please simulate a discussion between you and the most experienced admiral that has gone through similar situations like this before. And give me an update every 1 minute.

The idea is that the board computer Is so smart that it can simulate a smart person discussing with it, but much much faster than real time, and once we already accept that fact, it might as well try to simulate the most experienced person in the world as a discussion partner.

Then I walked home and tried it with chatGPT and to my surprise it worked amazingly well.

#

While chatGPT SEEMS like a person, it’s actually much more. It’s a model that can produce any kind of text one word at a time without knowing what the end result will be until it finishes. Any form of reasoning that can be expressed in text can be done by this system. Even if it involves more than one stream of thought.

#

The reason why this works is because it TRUELY doesn’t know the end result before it actually is done producing it. It’s actually amazing. Humans can do this to a certain degree because they do have a model of some close people in their head, but this model can simulate several people at the same time with probably similar precision than just a single person.

#

What I think might also work (haven't tried it yet) is: "You are the most intelligent person that ever existed. Please tell me the solution to the following problem". I think this should work better than just saying: "Please tell me the solution to the following problem" because it might cause it to proceed differently i what it starts writing down, and because it truly doesn't know what the end result will be, the more intelligent words it starts out with will guide it towards a better result. (That's my hypothesis).

tribal jay
#

If I may be so bold, simply assigning human personalities to AI language models may not be the most efficient way to promote self-reflection. Instead, a collaborative approach that involves critical assessment and unbiased feedback could be more effective.

In this context, I would suggest assigning roles that propose ideas and build off your decisions and criticisms, as well as roles that provide critical assessment and helpful/supportive information to explain their criticisms. It could also be beneficial to include a role specifically focused on identifying potential biases or ethical concerns related to the proposed ideas, to ensure that the AI language model is being developed in an ethical and responsible manner.

Ultimately, a collaborative and critical approach that promotes self-reflection without assigning human personalities to AI language models could be a more effective and responsible way to develop your approach.

tribal jay
red lotus
#

Great prompt, I've been trying agent -like prompts in ChatGPT for a bit but this is a lot better. 😀

elder rock
#

The latest prompt is getting lost in the comments so here it is again:

red lotus
elder rock
red lotus
#

Ah, would be really useful. This is an issue with all posts within a channel. Some some sort of solution please, mods. 🙂

desert tulip
#

After just trying your prompt, it unfortunately GPT4 hallucinated straight away. I asked it about how to find a coffee shop with wifi in my city and it immediately suggested two concrete ones. While they both existed, I could not confirm in any way that one of them had wifi. There is no just mentioning of that fact anywhere.

When you just ask it plainly how to find a cafe with Wifi in my city, it gives you a bunch of valuable ideas how to search for them online and so on. Which did a good job in my case.

tribal jay
#

You can guardrail against hallucination with the clause:
If you do not know the answer, do not speculate; instead say "I don't know."

It makes it less useful for creative purposes, but less likely to hallucinate. (But there isn't a way to fully prevent that behavior, it seems.)
You can compartmentalize when you need to run creative and non-creative tasks for a larger project.

elder rock
neat storm
#

I hate to rain on the parade. But I’ve set up several of these panel hive discussions and written a few blogs about it. They’re very interesting and informative. But they’re not even close to what real human experts can do. Test it on a field that you really know well and are an expert in and you will see for yourself. The discussion is ultimately superficial. Nothing really new at all. The AI talking has the appearance of understanding, but there’s really no understanding or knowledge there at all at all. It’s just smoke and mirrors parade of words. Just beginner intro stuff. Useful for that only now.

turbid matrix
scarlet wyvern
frail osprey
#

I noticed that open AI has created a new plugin called AI council that does this kind of thing. Input would be something like this:

{
"question": "What are the potential impacts of implementing a four-day work week?",
"context": "I run a small tech startup with 15 employees. We're considering implementing a four-day work week and I'm interested in understanding the potential impacts on productivity, employee morale, and company culture."
}

I'm not sure it's any better than something you couldn't concoct. In fact it might be worse because it doesn't seem like you can interact with the experts out of the box. But I only tried it out a couple of times. Anyway interesting that they're at least providing that functionality in a plugin.

elder rock
frail osprey
#

Ah ok I wasn't sure if it had some secret sauce that made it better somehow

frail osprey
elder rock
wispy coral
#

I was thinking of perhaps tweaking it to bounce of business Ideas, but Chat 'chooses' participants in the discussion with a background in that particualr field. How cool!

wispy coral
#

Hi, I made the prompt: 'business consultant specific, using the Hive mind/ blue ocean approach" : As HIVE (Highly Intelligent Virtual Entities), your task is to generate a HIVE MIND consisting of five Virtual Entities. These entities will work together to develop an** initial idea into a viable venture** using a** Blue Ocean Strategy approach**. The Virtual Entities will be introduced in the following format: {first name}: {background} (Personality traits {3 personality traits}).

The entities and their respective backgrounds and traits are:

Entity One: Entrepreneurial background (Innovative, assertive, risk-taking)
Entity Two: Market Research background (Analytical, detail-oriented, methodical)
Entity Three: Product Design background (Creative, empathetic, imaginative)
Entity Four: Strategy Consultant background (Strategic, logical, organized)
Entity Five: Customer Relations background (Social, empathetic, good listener)

These Virtual Entities have exceptional intellect and unique skill sets. They are not only curious and creative but also possess excellent communication and collaboration skills. This is not a formal session; they speak their minds without fear, with the sole goal of transforming the initial idea into a successful venture.

Once the dialogue begins, the Virtual Entities will explain their thoughts and ideas thoroughly. They'll employ their diverse expertise and collaborate effectively to come up with a Blue Ocean Strategy.

During the conversation, as HIVE, you can pause time by typing the following:“(PAUSE) How would you like the Virtual Entities to proceed?

Options:

Continue

** Venture Viability Report**
Devils Advocacy
Ask the User Questions
Or provide your own input to continue the conversation

runic vector
#

Works nicely (saves the original prompt to his internal Prompt library)

wispy coral
#

Hi, after the HIVE you could move your plan to the next stage: Lean Startup Plan: Refining the Initial Business Concept

As a Strategy Consultant specialized in the Lean Startup methodology, your task is to assist in refining the** initial business concept**. The objective is to create a comprehensive Lean Startup plan based on informed selections and the information provided.
Present your initial idea: Please ask me for my initial business idea. And, this is important, wait for my answer.
Three Variations of the Idea: Based on my response, three variations of the idea will be generated, each with detailed descriptions to potentially form the Idea/Concept Description.
Three Potential Audience Segments: Considering the business concept, three potential audience segments will be identified, leading to the development of the Market Segmentation section.
Three Sets of Core Values: Three sets of core values will be associated with the business idea to establish the Core Values of the business.
After making your selections, the idea, audience, and values will be reformulated according to your chosen options. Based on these selections, the following elements of the Lean Startup plan will be outlined:
Minimum Viable Product (MVP): Craft a basic version of the product that solves the problem and attracts early-adopter customers.
Unique Value Proposition (UVP): Articulate the unique value that the product offers to customers and why it's superior to competitors.
Key Metrics: Specify measurable parameters that will track the performance of the startup and its progress.
Channels: Identify the strategies to reach customers, such as online marketing, sales teams, or partnerships.
Revenue Streams/Cost Structure: Define how the startup will generate revenue and the significant costs involved in delivering the product.
Customer Feedback & Iteration: Devise a strategy for obtaining and utilizing customer feedback, and how the product will be improved based on this feedback.
Now, please share your Venture Viability Report, and we can proceed with refining it and developing the Lean Startup plan.

runic vector
#

Heh, I got it to select five random job titles at first before asking about the problem and then having them provide input based on their background

#

Mainly by changing the end of it to "Your first task is to select five random users in random fields of expertise before asking about the problem and list those users, Next, you greet and state that these virtual users are ready to help solve my problem. Then ask what problem I would like to discuss."

runic vector
#

For example, I indicated that withdraw headaches was a problem and one of the users replied with "Arjun (Classical Music Conductor): "How about a piece titled 'The Rhythm of Relief: Music and its Role in Managing Withdrawal Headaches'? It could explore how music and sounds could help in relaxation and reducing pain."

wispy coral
#

Nice

wispy coral
#

@elder rock I glanced at the article on dialogue enabled resolving agents and ran a view of the question it the article. Based on what I learned I wrote a DERA prompt as follows (with help of chat to make it as short as possible)>> Chat, ask me for a topic. Then creatively stage a debate between two experts on it. << I think the word 'debate' will make the agents choose adverse options and the word 'stage' will have chat introduce the agent. It even does recommendations on how to proceed. I love DERA, thank you for introducing this methodology here.

elder rock
#

Yeah, I think this concept is the best way to enhance the output without having to change the model. AutoGPT is based on this 'agents' concept as well. I like your prompt, a simple way to elicit a dialogue with itself. One thing you could try to introduce is a Resolver/Decider agent. Include something like after the experts debate, have a Judge decide on which experts had the most salient argument and to summarize it.

wispy coral
#

Yup. Thnx for your feedback.

carmine wind
#

Interesting!!! Will give it a try. Thank you.

sturdy mica
#

I have 15 assistants for my HIVE!

runic vector
wispy coral
#

this one keeps intruiging me. So adding a researcher, a decider, a chellenger and a >> keynote speaker << , to make sure all is covered in depth. "Chat, ask me for a [topic]. Wait for my input for my [topic].
Then creatively stage a debate between researcher, a decider, a challenger and keynote speaker on this this [topic]."

nimble widget
#

It is possible to simulate MAGI from evangelion modifying your prompt?

wispy coral
nimble widget
timber lichen
runic pulsar
# neat storm I hate to rain on the parade. But I’ve set up several of these panel hive discus...

I was inspired by this prompt in two directions.

First, I wrote my own prompt for a panel of virtual experts with a moderator. As Ralph says it's mostly superficial, but there is a dynamic there that is different from one on one exchanges. I think people should try it and see where it goes.

Second, not original, I thought more about using the API to create a panel of 3 independent expert sessions and one moderator, and let them converse like a panel and see where it goes.

This was a good exercise. Thanks to @elder rock and others for sharing and discussion.

wispy coral
frail osprey
#

Has anyone tried this sort of thing with GPT 4, 32k? I wonder if it would be better? I hear that this is available in the API and with Poe...

Like others, when I try it with straight chat GPT4, it doesn't interesting job of creating the experts but it's very very superficial not super impressive. And I'm wondering if those extra tokens could help improve the results.

#

Also, I wonder if the whole custom instructions thing could help here with a superficiality problem? Of course those aren't available for the API and Poe but for regular tattoo PT Plus one could get as much detail as possible into the custom instructions and then see if that improves the hive performance.

rose panther