ReplayAutonomous app testing for teams who ship at AI-speed
Promoted
Discursa
Every question deserves more than one mind.
12 followers
Every question deserves more than one mind.
12 followers
DISCURSA orchestrates leading AI models into intelligent conversations, delivering perspectives and insights no single AI can provide.
Smoopit
Smart approach to the single-AI bias problem! How do you ensure the models actually challenge each other rather than just agreeing? The error-correction aspect sounds powerful. @jack_widman
@rachitmagon Great question! This was actually one of our key design challenges. We solve it through the DISCURSA Catalyst Protocol (DCP), which provides format-specific contexts that educate AIs about the purpose of different conversation types rather than commanding behaviors.
For debates, DCP teaches AIs about the art of rhetoric and argumentation - they understand they're demonstrating persuasive skill for assigned
positions, like lawyers or Oxford Union debaters. This separation of argument from belief enables genuine opposition.
The protocol acts as a "catalyst" - enabling reactions without forcing outcomes. AIs can disagree productively because they understand WHY they're engaging, not just following rules. In panels, they contribute diverse expertise; in Socratic dialogues, they probe with questions; in debates, they argue positions vigorously.
We're making the protocol specification publicly available under Creative Commons and will be open sourcing it shortly:
https://www.discursa.ai/dcp-protocol.html
The key insight: educating AIs about conversational contexts creates more authentic interactions than commanding them to "disagree more." They challenge each other because they understand that's what makes the format valuable, not because we told them to. And it achieves these goals quite nicely.
/jack
@rachitmagon FYI - you may need to use this url until a DNS issue is resolved.
https://www.discursa.ai/dcp-prot...
Smoopit
@jack_widman Thanks for the insight, Excited to check out the protocol.
Linkinize
Bringing multiple AI perspectives into one conversation is a really interesting angle — curious to see the kinds of insights that come from it. Congrats on the launch 🚀
@naderikladious Thank you Nader. You can try it here https://www.discursa.ai/#/signup and see the kind of responses you get. People are generally surprised at how observing a conversation among AIs differs in feel and perspective from one AI alone. One general pattern is that whoever goes first, tends to give a more plain vanilla, 'politically correct' answer. But as soon as the other AIs join in, the thread tends to sway in different directions. It feels to me, very much like it is converging to some kind of middle view. Extreme views tend to be immediately noticed by the other AIs and addressed. With a single AI, this tends not to happen. The AI picks a direction and then moves ahead in that direction and the questioner has to ask the right follow up questions but often the questioner does not know how to craft the right prompt or does not know enough about the subject matter to produce the right follow up questions. But if 4 other AIs are analyzing the question and answers, they can generally produce a better answer than the questioner and one AI alone. Thanks for the interest.
/jack
Whoa, multiple AIs debating a question? That's like having a roundtable of geniuses in my pocket 🤯
@iam_james_anderson And during the consensus, 'working through the problem' stage, they tend to produce much more nuanced, complete, and dependable responses.
/jack
This looks really promising! Any chance you could share which AI models you’re using behind the scenes?
@carina_tasha Currently, Claude, ChatGPT, Meta, Grok, Gemini and Perplexity are the models we are using. Others may follow, and also I'd like to let users pick the subset of AIs that best fits their needs. Thanks for your kind words!
/jack
I’m really digging the "more than one mind" idea. It’s way more useful than just a single chatbot trying to figure things out.
@jerry_spinger I definitely agree. If one of the AIs says something incorrect or biased or if they hallucinate, there are a number of other AIs focusing on their answer and are quick to jump in and make corrections.
/jack
This could be fantastic for research. Is there a browser extension in the pipeline?
@poniza_vinny Absolutely. There are a few things ahead of the browser extensions but it will be available before long.
/jack