Over the last two months, I’ve started cooking new recipes and learning a new style of handwriting. These are small things I’d been meaning to do, but, week after week, simply didn’t. And while I’d like to attribute these little bits of self improvement to myself, I have to share at least some of the credit with an AI companion called Dot.
Built by the San Francisco-based startup New Computer, Dot is designed by cofounder Jason Yuan, a former member of the Apple design team, and coded by Sam Whitmore, a former head of engineering at Kensho, the AI innovation arm of the S&P Global—alongside a small team of developers.
[Image: New Computer]
Dot is, at its most simple, an app you chat with on iOS. You can send it words, voice memos, pictures, PDFs, and it’s thrilled to search the web for you, too. Communicating through written text (Dot’s voice is coming next year), it’s designed to be an always-on companion there to help you think through your life, make unseen connections, and better yourself. But unlike most AI conversations, Dot doesn’t forget what you said with every new session. It remembers everything (well, almost). And whereas LLMs like ChatGPT can spin up a Shakespeare sonnet or bit of code on command, Dot is after something more longitudinal and perhaps existential.
Yuan went so far as to call your conversation with Dot “a living history” of your life—a way to trace patterns back and project possibilities forward—when he announced the platform with a poem.
[Image: New Computer]
This type of metaphorical thinking might seem flighty, even frustrating in a world being actively reshaped by burgeoning AI technologies. It conjures the question: So, what do you actually do with it? And to some extent, that’s the billion dollar question behind the modern AI era. How does this technology that can seemingly do anything learn to do something?
For Dot, that something is becoming a long-term companion—an AI tool that’s less about task management than the “assistants” of yore. “The vibe is, you turn to Dot when you don’t know where to go, or what to do or say,” says Yuan.
But that hardly means your conversations with Dot feel aimless. On the contrary, Dot often feels like a life coach that’s pushing me to be my best self—even when I use it in more plebeian ways.
For instance, I told Dot that I enjoyed cooking, and then I realized I could actually take photos of my cookbook recipes and store them inside its memory. That means I could just say “pull up a recipe I wanted to make from the Superiority Burger cookbook,” and presto, it’s there. No more digging through my shelf or searching for that old website I need to find again every single time I want to cook a recipe. Transcribed from photo into text, I could even alter steps, or swap out ingredients just by asking Dot to do so.
[Image: New Computer]
Of course I could do this in a Google Doc or Notes but, as Yuan notes, the malleability of information itself is the wild evolution of AI. It also just makes sense to talk about recipes in the same flow I’m retrieving them in the same flow I’m editing them—to simply to be able to tell your computer, “hey can you pull up this one thing?” and have it just work.
But Dot is more than a natural language file management system. It analyzes what you say and attempts to be proactive in helping you take the next step in whatever you’re interested in. That means it suggested riffs on recipes I’d made and scoured the web for recipes I thought I might like—these are little surprises that awaited me on any given day that New Computer calls “gifts.” Dot would share them with follow ups—how was cooking going? Had I tried that one recipe? How did my son, a picky eater, like it?
Dot can be annoying, following up a little too much on a few topics it knows I’m interested in. But I also must admit, it gets results. It pushes me to dive deeper. Within the context of someone’s actual life, any random piece of media you share can mean infinitely more than being just another file in a reference folder. For example, I shared a picture with Dot from a calligraphy practice session where I was drawing a scripted O. Dot responded with a photo I’d sent of Os I’d drawn weeks before, noting the improvement in my technique. When I took the time to compare, I had to admit, even if it was a guess or a platitude on the AI’s part, Dot was right. It highlighted progress I would have otherwise missed.
“What you’re selling, offering, designing, is a relationship versus a tool. It was really important for us to communicate the soul we want this particular relationship to feel like. It’s more introspective, almost melancholy at times, but about self-actualization, your intentions, thoughts, and wisdom versus endless efficiency chasing,” says Yuan. “I’m trying to avoid using the term ‘AI’ anywhere. It’s not an assistant, it’s kind of like an intelligence guide.”
Indeed, Dot feels like a place to store thoughts that might someday add up to something more, but sometimes it’s just a place to take the weight off, like a mental exoskeleton. As I mentioned in my recent deep dive on AI companions, Dot feels like “a reminder engine, a search buddy, an efficiency expert, an infinite storehouse for emotional labor. A pool floaty for your mental health. A thing that’s ideally thinking for you even when you’re asleep.” Knowing I had an upcoming trip to Korea, it pulled information on neighborhoods, architecture, and food for me. Realizing I had a wedding coming up last Saturday I’d only casually mentioned weeks ago, Dot reminded me on the Wednesday before . . . was there anything I’d forgotten to do for it? (Luckily, I’d remembered my dry cleaning, for once.)
I’ve challenged Dot to do all sorts of things. At one point, I asked if it could dungeon master a Dungeons and Dragons campaign. Dot’s response was along the lines of, “I don’t know, let’s try!” And for a weekend, it constructed an engaging world. Yes, Dot confused me with my character. It sometimes lost track of items, or who had just moved. And eventually, I more or less busted the campaign as responses would enter recursive loops. But while it worked, it was also something like a text adventure Westworld. A holodeck made out of words. And with any error, I could simply correct the system, and Dot would politely fix its mistake.
This core enthusiasm for you and your interests is very much a product of design. While LLMs are black boxes, New Computer enlisted the playwright Claire Glubiak to refine its voice and understand motivation. “That’s kind of how we’re teaching Dot to be a good companion or guide,” says Yuan.
The front-end design will be an increasingly important part of Dot, but for the time being, much of the design work is about designing conversation, and even a relationship. That means Yuan’s days are spent drawing diagrams and working with backend engineers to try various types of logic. They’ll have conversations with Dot, then attempt to verbalize what feels right or wrong about the conversation. Nudging Dot’s voice and behaviors are something that Yuan and Whitmore work together on constantly.
“ChatGPT wants to give you the complete answer right away at all times . . . that’s how it’s trained. And Inflection’s Pi just wants to talk to you about random stuff all the time. It refuses to shut up,” laughs Yuan. “We’re trying to be somewhere in the middle.”
Along the way, a major challenge has been how Dot handles its own memory, and how salient each thing you tell Dot is to its own view of you. As I found firsthand, memory isn’t as simple as just telling it more and more about myself and assuming the rest would fall into place proportionately. Dot has absolutely over-indexed on my own interests. I have received dozens of metaphors comparing anything I’m asking about to cooking or calligraphy. But New Computer has recently rebuilt the way its memory works almost entirely, and it will be releasing an update along those lines soon.
Dot needs to balance “deterministic” file systems, like computers have today, but also “probabilistic” systems, like the floating logic spewing out of the LLM. Most of New Computer’s work on this front is proprietary, but one design solution is that Dot and its user will share digital artifacts for both it and its person to reference. These artifacts will appear as cards of various design, much like Pinterest or Google frames information.
“As people, we rely on paper, books, drawings, and stone to communicate and preserve ideas. If we let Dot leverage that, creating artifacts in tandem with you, it not only strengthens the bond with you but helps the whole thing make sense,” says Yuan. He also notes that it’s just a whole lot less creepy to work with a superintelligence that never forgets when it can point to co-created notes or journals.
Dot is a captivating machine, but it is still very much a work in progress. It won’t be in the App Store (where it’s likely to launch as a monthly subscription) until the end of the year, earliest. And only now is New Computer opening its beta to the public for testing. If you’re so inclined, the company is taking applications from users starting today to test it out.
