Danny Geisz: On interning at Imbue, going full-time, and making magic

October 3, 2023
For this post in our Humans of Imbue series, I sat down to chat with Danny Geisz, previous intern at Imbue, now a Member of Technical Staff. We discussed his passion for creating magical experiences and how that translates into his work exploring LLMs and coding agents at Imbue.

I heard about Danny before I met him—he was an undergrad at UC Berkeley and a collaborator of Jamie Simon (an Imbue team member), and the team was abuzz about the idea of bringing him on for a summer internship.

Jamie’s testimonial was strong. “He’s quick, asks good questions—I never have to repeat anything. The man is just sharp and has strong opinions. He’s very much like, ‘Here are things I believe in, here are things I want to do’. He comes fully stocked with interesting and useful opinions, which is not how undergrads usually work.”

The team validated these lofty claims during the interview process, and Danny joined Imbue.

Danny loves magic

As I got to know Danny, it became apparent that he loves magic. Not necessarily in the literal sense (though he is a skilled magician), but in the sense that he loves magical things—and creating them.

This love can be traced back to Danny’s childhood. “Magic tricks were a huge thing for me from a very young age.” He also explored a plethora of equally imaginative hobbies, from yo-yo’ing to Rubik’s cubes, unicycling to ukulele. “I was constantly trying to wow other people with different things I created or did.”

By fifth grade, Danny knew he wanted to become an engineer capable of creating new and interesting things. “I don’t even know where I learned the term ‘engineer’, but putting different things together and creating something new out of nothing was very salient for me.”

Danny explores subatomic particles and quantum mechanics

One serendipitous day in fifth grade, Danny happened upon a mysterious relic in the book drop at his school’s library—a book, far too advanced for elementary schoolers, about subatomic particles. As he dove into the material, the gears of his imagination started to turn. “It spurred me to try to come up with my own theory for how things worked.”

Obsessed with gaining knowledge, Danny blazed his own trail. He was put on an unprecedented fast track through his high school’s math curriculum. By sophomore year, he was taking BC Calculus and AP Physics, all while teaching himself quantum mechanics on the side.

Danny recounted the time his AP Physics teacher recorded the class for a student who was missing that day. He was caught red-handed glued to his phone the entire class, but upon further investigation, it was revealed that Danny was actually reading ahead in the textbook, working through problems. Danny was on a quest for knowledge, and it seemed that nothing could hold him back.

Danny goes to Berkeley

Like many, Danny was excited to go to college to explore his passion for physics. But despite taking a challenging course load and doing research on the side, Danny started to become disillusioned with the field. He started asking himself, “Why does any of this matter? Is this thing that I’m spending so much time on actually going to change anything?”

Dissatisfied and unhappy, Danny set his sights on more practical projects. Inspired by a computer science course that he described as a “breath of fresh air”, he started taking on bigger engineering tasks.

Forever treading his own path, Danny decided to take a gap year to explore various entrepreneurial endeavors. Every signpost pointed him in the direction of becoming a founder, but something didn’t feel right to him. “There was this tension between a desire for agency and freedom, but there were definitely aspects of being a founder that did not energize me.”

Returning to Berkeley, Danny decided to give physics another shot, and after a fortuitous cold email, he met Jamie. Jamie went to work getting Danny up to speed on deep learning theory, but once again, Danny couldn’t be contained. “He quickly got bored of deep learning theory and graduated to wanting to do a bunch of wildly experimental stuff,” Jamie recalled. “Man’s more of a creator, I think.”

After some time, Jamie decided he wanted to introduce Danny to the Imbue team. When I asked Jamie why, he reflected, “Kanjun (Imbue’s cofounder & CEO) had waved a magic wand and completely unblocked me creatively by giving me extra resources and encouragement and guidance and stuff. I felt like I could like pass that on.”

Jamie and Danny (who Jamie refers to as his “lad”) at Fun Friday.

Danny interns at Imbue

Reflecting on his internship, Danny’s most powerful memories are of personal growth and meaningful relationships.

The first thing Danny discovered was that Imbue is unlike other companies in terms of leadership. “My model of companies was that the people above me had a particular vision for what they were trying to do, and I basically needed to fit myself into actualizing that vision. Kanjun told me that wasn’t true at all in our first one-on-one.”

Through their one-on-ones, Danny developed a deep connection with Kanjun. “The felt truth of Kanjun’s optimism about humanity, the extent to which Kanjun cares about humanity—it’s a very human, emotionally salient vision. Amidst all the chaos, it really anchors Imbue. It’s the sort of energy that I really do not feel about the other major AI labs.”

Danny’s eyes were also opened to the concept of creative agency. “Everyone was very, very supportive of giving me agency, and they did whatever they could to make me feel trusted and safe. Even if my perspective wasn’t immediately obvious or was initially confusing to people, there was a lot of trust given for my process.”

As an intern, Danny found a close collaborator in Josh (Imbue’s cofounder & CTO). “Our minds work so differently, and we bring a lot to each other. He gives me enough trust to share my very different way of going about things, but also simultaneously challenges the way that I do things in a way that’s very constructive and productive.”

On paper, Danny spent the summer experimenting and incubating his wild ideas about coding agents, but he claims, “A lot of what I did as an intern just properly set me up to actually start to work.”

Danny joins Imbue full-time

The last day of Danny’s internship was Friday, August 15, and his first day as a full-time Member of Technical Staff was the following Monday. “It felt like the most natural thing in the world. I just flowed from what I was doing as an intern directly into what I was doing as a full time person. There was no change in trajectory because I knew what I was about as an intern, and I just continued working towards that as I moved full time.”

Today, Danny is continuing his work on coding agents, but that statement doesn’t convey his deeper ambitions. On this topic, Danny says it best:

Right now, we don’t know what we’re doing in terms of working with LLMs. The way we’re working with large language models is very, very primitive because we don’t know how to reliably extract information from them. We don’t know how to reliably integrate them with other systems.

There have been a lot of different pushes that make people think we know what we’re doing in terms like prompt engineering or different tree of thought practices but we’re very much in like a dark wild west. There’s just so much to explore in the space.

For a long time, I’ve been very interested in what it would look like for a computer to generate code. The notion that a machine is able to comprehend and architect a code base is very salient to me.

And so basically, I’m trying to make this happen. I’m trying to figure out how to use LLMs in conjunction with other tools and other systems in a way that just feels good—like is something you can trust, is something that you can rely on, is something you don’t have to constantly babysitting, is something that feels magical. Where you somehow just express your intentions and what you want to see to the computer, and the computer just reliably makes that vision happen in terms of what the code should be like to accomplish your particular vision.

I very strongly feel like the sorts of things that we’ll learn from doing this with code, because they’re so reasoning focused, will actually generalize very well to other domains as well.

I’m trying to figure out how to actually use LLMs in a way that is actually cool and magical and does amazing things. I want to build something where the hype is warranted because it’s f***ing magic.”

If you’re interested in creating magical experiences at the interface between humans and computers, we’re hiring for positions in both research and product. We would love to hear from you!