President Ronald Reagan memorably quoted the Russian proverb “trust but verify” when cautioning he’d trust the former Soviet Union’s adherence to a nuclear arms treaty, but verify it complied.
Based on that same admonition as it relates to the often questionable accuracy of artificial intelligence search results, a new startup called mixus.ai has just launched adding actual human intelligence to the mix for the verification end of the epigram.
Co-founded by entrepreneurs Elliot Katz and Shai Magzimof in the San Francisco Bay area, mixus.ai users can conduct typical AI searches on any platform of their choosing such as ChatGPT, Gemini, Perplexity or Claude, but can then solicit input from humans with expertise or knowledge of the search topic.
Katz describes mixus.ai as the “first collaborative platform,” borne out of the knowledge the results of AI searches may not be completely accurate.
“AI responses, even from top models, they’re incorrect over 50% of the time, and that figure is from an OpenAI study from just last month,” said Katz in an interview conducted by the author on the podcast Tales From the Beat. “We think that you really need to combine it with a human expert or a human brain. In that way, you’re going to have a lot more faith in the data points that you’re gleaning from that conversation.”
On mixus.ai, when a user makes an AI query, in addition to a AI-generated response mixus also recommends people who have expertise, experience or knowledge on the specific topic. The user can then add those recommended people into their chat and converse with them and AI together, Katz explained.
To help ensure the humans flagged are credible, mixus requires would-be experts to be verified by filling out a form on the website then presenting identification such as a drivers license. Experts can also post links to their professional or social media accounts as further documentation of their experience.
To further bolster, or question, an expert’s credibility, mixus allows for a crowdsourced rating system, Katz explained.
“Assuming that this person says, I’m a top expert on X, Y, Z, well, if they’re a top expert, they probably answered a lot of questions on this topic,” said Katz. “So you can go to their profile with the system that we’re implementing and saying, you know, this person is 100 out of 100 or 90 out of 100 or 80 out of 100 and you can get a lot of comfort that way.”
The site is now open to anyone, following a beta phase on Linkedin, and its launch a few weeks ago at Stanford University where Katz says hundreds of students and faculty have signed on.
Jeff Epstein, a professor at the Stanford Graduate School of Business, plans to use mixus.ai as a teaching tool in his popular course Lean Launchpad starting in January.
“Mixus lets our students and faculty work side by side with AI,” said Epstein in emailed remarks. “We plan to have one mixus thread for all students and faculty, one for each of our eight student teams, along with their mentors and one for the faculty and teaching assistants. For the five students on each of our teams, mixus is their sixth team member.”
Stanford student, and member of its volleyball team Emmy Sharp, said she used mixus to help develop a reconditioning plan after an injury.
“After tearing my ACL, I turned to mixus for quad-strengthening techniques,” said Sharp. “While the AI provided a helpful high-level response, one of mixus’ people recommendations was a certified strength and conditioning specialist I added him to my AI chat, and he developed a tailored quad-strengthening program for me.”
Student Katharine Sorensen used mixus to help plan for some leisure time activities, explaining, “I have found mixus to be most engaging and helpful when I have prompted the AI for restaurant and activity recommendations for a potential travel destination. I can then add friends who might be interested into the chat to hear their input and/or loop in other users who might have local expertise about the places we want to investigate.”
Access to mixus.ai ranges from free membership to several levels of paid subscriptions starting at $20 a month for the Premium plan and upwards to $150 per month for the Team plan and then the Enterprise plan, which is priced to each company’s needs.
The various levels are based on the number of users, AI chat messages per month, number of members in private groups and several other considerations.
Since its launch mixus has attracted “thousands of users” including several large companies, according to Katz.
A lot has happened in a short time for the slim team of four behind mixus.ai, but Katz is convinced adding actual human intelligence to the artificial variety is a game changer with plenty of more work to be done.
“We’re still very early. If this is a nine inning baseball game, we haven’t even gotten into our uniform and left to the ballpark yet,” Katz noted. “We’re learning a ton. We’re acting upon those learnings. We’re iterating on the product in reaction to those learnings, and where we want to go with this.”
