Welcome to the podcast.
I'm your host, Jaden Shafer.
Today on the show, I want to talk about a really interesting company called HumansAnd.
And I think the reason why a lot of people are talking about them is that right now we have all of these different chatbots and have gotten really good at answering questions.
They're really good at summarizing documents or solving equations, right?
All of these types of things we think about all the time.
But for all you know, for how intelligent and how smart they are, most of them still act like they're kind of this solo assistant, right?
They're optimized for one user and for doing one prompt at a time.
What they're not doing and what they're not very good at is some of these really messy, more human kind of work, of collaboration things that we do.
So whether that's like you know, coordinating groups with a bunch of conflicting priorities, or if they're tracking decisions over weeks or months, or if they're trying to help teams stay aligned or goals, and you know, all these goals and people and information is all kind of shifting around.
This is what AI chatbots are struggling with today.
And so humans and is building a solution to this.
I'm going to get into all of this on the podcast today.
But before we do, I wanted to mention the new feature we've just added to AI box that I'm super excited about.
And that is file uploads for our builder.
So we have a Vibe Builder tool.
If you've never built a software before or never built a tool before, you can go to AIboxai describe what you're trying to build on our builder and our AI will automatically link together different AI models and fill out the prompts.
And we've been doing this for a while.
It's something we're excited about. we just added the ability for you to do file upload.
So like one example is I created a headshot photo generator and you can create like a hundred different variations.
In fact, I had a friend and he used to have a company that was literally an AI Viking generator.
So you would pay him like 20 bucks and he would give you 20 images of yourself as a Viking and you upload a few photos of yourself or whatever.
You can now build that, along with a million different variations of that And, of course, a lot of very useful things when you're doing PDFs and files and all that kind of stuff that you upload to the builder and create tools and systems for all this kind of stuff.
But you can build that now on AIbox.ai.
So if you want to go check it out, we've just added that functionality and I'd love to hear what you guys think about it.
And there's a link in the description to AIbox.ai.
All right.
Let's talk about what Human Ands is doing.
So it's a three-month-old startup.
They were founded by some alumni that came from Anthropic, Meta, OpenAI, XAI, Google DeepMind.
So a lot of, you know, a lot of top AI labs.
And essentially they believe that the next kind of step for some of these AI models is not just better answers but better coordination.
So They just raised 480 million in a seed round and they're trying to get what they call a quote central nervous system for human plus AI, the human plus AI economy, which is basically a system designed around social intelligence rather than pure information retrieval.
So it's obviously kind of an interesting concept.
And if, you know, random people are trying to do it, who knows what would have happened.
But obviously, this is a very cracked team that has worked at a lot of these big AI firms.
And so they're able to raise half a billion dollars as their seed round, which is incredible.
So a lot of early, I think, coverage is kind of focused on the company's kind of AI for empowering humans framing.
But I think the ambition for their whole company is a lot deeper than I don't know.
They kind of had this kind of like motivational language on their website, which I don't think was that important.
Obviously, people are just giving them money because they have a great team.
But now I think this kind of direction they're taking is great.
They're trying to build a new foundation model, one that is architected specifically for understanding people in groups.
So how they communicate, how they disagree, how they align, how they make decisions together over time.
I think this is really kind of an interesting concept and important because, if you think about it, at work or at school or wherever, Sometimes we do projects alone, but a lot of projects we have to do with other people.
And there's a lot of different stakeholders in a project.
And you have to look at everyone's input and thoughts.
And it's just kind of a mess.
And so if an AI is trying to do everything, it's just going to try to do it all itself and not really take everyone else's thoughts into consideration.
This is what they said.
Annie Peng is one of the co-founders.
She was formerly at Anthropic, and she said, So, I think...
A lot of that shift is already possible.
And you see this across a lot of different industries.
Companies are moving from kind of these chat interfaces to more agent-based systems.
Models are getting much better.
I mean even just what ChatGPT 52 can do today is, I think, leagues above what we were getting a year ago.
So obviously, these things are getting better.
But some of the workflows, I think, are still really fragmented, right?
So if you're looking at the true bottlenecks, I think a lot of time it is coordinating things.
It's not just like how smart the AI model is.
And then you also have the other side of that, which is a lot of workers right now feel very overwhelmed or they feel very threatened by a lot of these AI tools.
And basically this kind of like promise of efficiency is, I think, scary for a lot of people who are like well, how is this going to address?
You know what's going on inside of my organization?
And is this going to displace me?
So humans and that's the name of the company.
And they're essentially trying to set themselves as an answer to that tension that we're seeing.
You know whether that is, you know, framing this philosophically or strategically.
I think this is probably a really good time for them to jump in and and kind of lead with that.
So, despite them obviously raising a massive seed round and they have a really solid team they don't yet have a product.
And I think that is also very interesting.
The company is really deliberate, I guess, in how vague they are about what the first vision is going to look like of their new AI model they're training
Although they've all kind of hinted that it's going to replace or sit along some multiplayer collaboration environments.
You can imagine things like Slack or Google Docs or Notion.
I think their target audience right now is both enterprise teams and, like different consumer products.
So companies and families, they kind of talk like they're going to approach both of those.
Their CEO, who is Eric Zekerman.
He was formerly at XAI and he said we are building a product and a model that is centered on communication and collaboration.
He said that their goal is to basically help people work together more efficiently, not with AI tools, but with one another.
So that's kind of interesting.
Something interesting their CEO brought up is that a really big part of the company's thesis is trying to understand or essentially have their chatbot ask questions, because right now, a lot of these chatbots are asking questions and they're doing this without really understanding why they're asking them.
And so I guess an example he gave of what they're trying to solve or something that they did is you know, in this kind of like group decision making is he said, that you know, if you're choosing a roadmap or you're trying to do something like a logo, it can help coordinate and often, you know, solve a lot of the problems that these really long meetings will have when there's kind of unclear consensus, there's decision fatigue.
Apparently, he said that this was something they put a lot of thought into because when they were coming up with their logo for their process, it was a super drawn out process.
And it was really hard for them all to agree on a logo and trying to think of something that illustrated what they were trying to solve.
And so he's trying to create this smarter essentially, this model that is optimized for building trust and, you know, creating this kind of long term value of clarification, this context building.
He said right now, basically all of the AI models are optimized for giving this, like immediate user satisfaction and answer correctness.
But they're not trying to do these other things.
So they want their model to ask questions the same way maybe like your colleague would ask.
They said they want it to be really selective and purposeful and with memory and some prior interactions that you might have had with it, and integrate all of that into the response it's giving.
It's a very ambitious project.
It's also kind of vague and opaque and hard to understand.
If I'm being 100 honest, there's a lot of buzzwords thrown in there and a lot of like this is a problem.
And if I'm being 100% honest, I don't know if all of their problems are problems.
I'm excited to see what they come up with.
And if they just have a model that's really good at one specific thing, or if they pivot to that in the future, I'll be thrilled.
They've raised half a billion dollars.
I think they're going to be able to pull something off.
And I'm not trying to be pessimistic, but sometimes I do feel like that.
One of their co-founders Peng, said part of what we're doing is making sure that, as the model improves, we can co-evolve the interface and behavior into something that actually makes sense as a product.
I think obviously what they're trying to do is not build a model, that is just simply plug it into some existing collaboration tools.
They want to build their own collaboration layer that they were actually creating.
I think it's definitely a crowded space.
There's a lot of competition here.
There's a lot of different productivity tools and they're getting a lot of money.
They're raising a lot of money.
So everyone's trying to kind of create this intelligence layer built into meetings and notes and workflows.
So you could look at like Granola.
That's an AI note taking startup and they just raised 43 million at a 250 million valuation and they're rolling out a whole bunch of collaboration features.
And so and it's also kind of interesting, like with granola's case, because they just started out as like literally, an AI note taking startup they got a lot of usage, they were quite useful.
And then they're rolling out to all of these other areas where, like human and is, like you know, got some really solid researchers.
And they're like, we're going to do everything, but they have no product.
And I sometimes I wonder which of these is better.
And I feel like something like granola is a great company where they proved that they had a lot of demand.
They proved that they had a solid team.
Then they go raise more money and expand humans and and a lot of these other labs and and maybe i'm gonna look like the dummy in the future but um, a lot of these labs that just raise massive amounts of money with no product and no roadmap, i don't know.
To me it just seems quite crazy and i hope that they can all deliver.
I anyways, that's gonna be the most stressful way to do it for the founders.
I guess i'll put it that way.
So, At the same time, there's a lot of really influential voices that are kind of reframing the future around AI coordination rather than automation.
Reid Hoffman has argued that companies are misusing AI by using it in isolated pilots rather than kind of getting it straight into how their teams are sharing knowledge and information.
Decision making.
Um reid hoffman was recently writing and he said ai lives at the workflow level.
The people closest to the work know where the friction is.
They're the ones who will discover what should be automated, compressed or redesigned, and i think that is sort of what humans is trying to do.
Um, the vision is basically a system that's going to act as a is like the quote-unquote connective tissue across an organization, or even maybe like a household right, if they're trying to do more consumers.
And it's trying to understand individual skills preferences, motivations.
And then it's also trying to help balance all of that in service of different like shared goals.
So I think getting there is going to require a lot of rethinking of how models are trained according to them.
They said, we're trying to train the model in a fundamentally different way.
That's Yuchen He, one of their co-founders and who's formerly an OpenAI researcher.
And they're planning to rely on long horizon reinforcement learning and multi-agent reinforcement learning.
So these are both techniques that are kind of designed to help models plan, revise and follow through over extended periods.
And also across, you know, a lot of different participants.
So long horizon reinforcement learning.
It basically is focusing on outcomes over time rather than just like one off responses.
And multi agent reinforcement learning trains a system to operate environments where there's multiple humans and AIs are all kind of interacting simultaneously.
So both of these approaches, I think, are getting a lot of traction in academic research.
The field is kind of pushing beyond just chatbots and kind of pushing towards systems that are capable of coordination and sustained action.
So if they can come out on the right side of a really powerful kind of like agent coordination tool, I think that they will build something that a lot of people want.
One of their co-founders, he said, the model needs to remember things about itself and about you.
The better its memory, the better its understanding.
So I do think that's true, but I also do think a lot of people are building memory and doing a lot of things there.
So they're not the only ones.
I think there, you know, there's a lot of optimism in this company.
I mean, they were just able to raise almost half a billion dollars, but I think there still is a lot of risks.
Training and scaling a brand new model is very expensive, right?
If you look at someone like the other company I was just mentioning Granola, the Note AI note-taking app, like they don't have their own model.
They're.
You know they're using someone else's and they're still able to go raise a ton of money and be a very successful company.
Um, but of course, if you want the top dollar, you need a, you gotta make your own models.
That's what looks like humans and is doing.
This is not easy.
This is quite difficult.
Um, they're going to need continued access to, like tons of compute in, you know, a market that right now is very dominated by incumbents.
There's these massive companies Microsoft Google, Amazon.
Everybody is trying to get access to more compute.
So they're going to be battling with all of those people.
So there is a lot of competition.
Um, And I think there's also a lot of like indirect competition with a bunch of other big AI companies.
So I don't think this is just something that's going to be.
You know, they're like battling with Notion or Slack.
I think the real competition is going to be coming from Anthropic and Google, and you know like Google's embedding Gemini into their workforce and Anthropic is pushing, you know, collaborative workflows through Claude Cowork and opening eyes when promoting their multi-agent orchestration and workflow tools to developers.
So I think like their competition is against a bunch of really big players.
Each of those is moving towards, you know collaboration, even if they publicly debate like how soon AI is going to replace a lot of valuable work, that is going to make an impact on the economy and all that kind of stuff.
I think we know they're all moving in that direction, right?
All the biggest players are moving in that direction.
So they're going to have to compete with all of them.
I think so far none of the biggest players appear to have rewritten their models around social intelligence, kind of having this core principle.
And so that's where humans and thinks that they can differentiate themselves and be different.
And I think that could give them a bit of a head start.
Or maybe you know they're making an approach that isn't necessary and open AI deep mind meta approach.
All those other companies are going to be able to do it with their current tech stack.
I think a lot of this also comes down to recruiting top talent, which is also going towards a lot of these big players right now.
Humans and said that they're not interested in being acquired.
They said that they are.
You know they've already turned away a bunch of people that have like been interested in that.
They said, quote we believe this can be a generational company.
We think it has the potential to fundamentally change how people interact with these models.
And we trust ourselves and the team we built to do that.
So I think, whether they become the quote unquote, connective tissue, as they say, of human AI collaboration, or maybe they're just another really ambitious experiment,
I think all that's going to depend on whether this social intelligence can be operationalized at scale.
I think the bet right now is that the next breakthrough in AI is like maybe this isn't going to come from just smarter answers, but maybe from systems that understand how humans actually work together.
That's their bet.
That's, I guess, what they say they're going to do.
So it'll be interesting to follow along and see what they're going to do.
They've just raised half a billion dollars almost.
So this will be a fascinating company to watch.
Thanks so much for tuning into the podcast.
If you learned anything new in this episode and if it was interesting or insightful, it would mean the world to me and help the show out a ton.
If you could leave a rating and review on Apple, you can leave a comment.
I read them all and I really do appreciate them.
And then over on Spotify, it's the about tab.
So it helps the show get pushed out to more people.
It really helps with the algorithm.
If you could take a second to do that, that would be incredible.
And honestly, I'd really appreciate it.
And otherwise make sure to go check out AIboxai if you would like to build tools.
If you're not a developer.
I'm not a developer, so I basically built this platform for myself and it's really useful and I love to get your guys' feedback on it.
So go try out AIbox and let me know what you think.
Have a great rest of your day and I'll catch you in the next episode.
