We believe that the most important assets in our company are our employees— that’s why we’re launching our new Team Spotlight blogpost series, where we sit down with a member of our team to discuss their background, and what they do here at Seek.
This week, I’m on the record with Raz Besaleli, our co-founder and Director of AI Research!
Raz joined Seek AI in March 2022 as the founding NLP Research Scientist and then as Head of AI in September 2022. Prior to Seek, they worked on research in computational semantics and NLP applications for indigenous and low-resource languages. At Seek, Raz is interested in investigating how valuable business questions can be understood as structured representations, and how such questions can be discovered via process automation using foundation models and reinforcement learning.
Check out Raz’s personal website here.
Hello Raz! To begin, tell me a little bit about yourself!
Well, going all the way back, I was born in Maryland to musician parents, and spent most of my childhood moving around— from Alabama to North Carolina, then from New Hampshire to Vermont, then back to Maryland and eventually to New York. I actually left high school early and took some community college classes instead, completing my undergraduate degree at St. John’s College in Annapolis, MD. I’m currently a final-semester MS student in computational linguistics and a graduate research assistant at the NLP Lab at Montclair State University, on top my work here at Seek.
We have a few Johnnies here at Seek (including myself!). Reflecting back on your time at St. John’s, how would you say your “Great Books” education help you succeed in what you do here?
At St. John’s, we all graduate with a double major in Philosophy, History of Mathematics and Science and a double minor in Classics and Comparative Literature— and I happened to focus on the math and science! I actually wrote my senior thesis on Cauchy’s residue theorem, which is a topic in complex analysis. So while my focus on mathematics definitely equipped me with the foundational knowledge relating to what I work on everyday, I’d say that the most valuable thing St. John’s taught me was to question assumptions and to not rely on them, which I’ll get to in a second.
In true St. John’s fashion, I’ll have to agree with you on the risks of relying on assumptions then— we’ll circle back to this. Well, you’re one of the co-founders of Seek AI now, tell me a little bit about how that came to be! What's your Seek story?
I actually met Sarah Nagy about three months into grad school (shoutout Montclair State Uni) while I was working as a research assistant. We met through AngelList, got to talking, and then the rest is history!
That's amazing! So what exactly do you do here at Seek?
Basically, I direct the construction of our natural language-to-SQL engines. In the end, it gets down to semantic parsing and information retrieval— understanding how we can understand natural language utterances or queries for data as structured representations. You could say that it’s a lot of coding, and a lot of crying over math.
In all seriousness, though, in addition to the usual day-to-day meetings, I’m involved in very exciting research that we’re working on right now, so keep an eye out for that!
You have an impressive background in NLP research— so while we’re on this topic, what advice would you give to someone interested in pursuing a career within this field?
I think it’s really important to understand the distinction between natural language processing, so creating statistical accounts of language, and statistical processing of things for which natural language is a medium. So I’d say that when you’re trying to do something as complicated as semantic parsing, you need to understand that you’re trying to generate things based off of surface-level representations. In a way, it’s kind of like Ptolemy’s methodology in his work on astronomy, The Almagest— Ptolemy doesn’t care about what the celestial bodies are per se; he cares about tracking their movements with a perfect circle. It really is a phenomenal science in the classical sense.
But with that being said, if you want to get into natural language processing, think. about. language. I’d even say that perhaps the first thing you need to learn is why language works the way it does you know? It doesn’t really have rules— there are very much latent structures which are completely unobservable, and it’s very rare that we ever figure out how language works.
In the end, I’d say it all comes down to three things: focus on language, don’t make assumptions, and start with the basics. Take that entry to linguistics class, get good at coding, and get up to speed on DevOps, and read as many papers as you can— for every Medium article, you could be reading an academic paper.
That’s great advice— I hear research, research, research 😉 Now that you’ve emphasized the importance of research in this field, and with AI being all-the-talk, I’d like to ask— what’s the biggest misconception about AI in your opinion?
This is my opinion, and my opinion only— artificial intelligence is not something intelligent in the sense that we know it. We don't make artificial intelligence. What defines artificial intelligence is the perception of systems as artificially intelligent as humans. It’s purely an artifact of human psychology, and very much has to do with how humans perceive something as “intelligent”. This concept goes all the way back to the 1960s, when one of the first “AI”s, ELIZA, came out— it simulated Rogerian psychotherapy and it was written in purely in regular expressions, being entirely deterministic. But you know, people back then started calling it sentient just like they call LaMDA sentient a few months ago.
In the end, I’d say that AI is nothing but the interaction of humans and things that they want to be human, and it very much to do with the fact that we are desperate to assign empathy and sentience to non-living beings.
And one more thing— AI is really not going to take over the world. Trust me.
What about the potential risks and threats of AI then?
You know, there's a lot of talk going around these “sparks of AGI”, but it’s important to keep in mind that there really isn’t a good definition of what general artificial intelligence is. There are working definitions, sure, but artificial intelligence is not a thing that’s coming soon.
I think when people worry about risks of AI, they immediately think of some sort of Terminator, Skynet-level of destruction like AI is going to burn us to the ground and become our evil overlord. That’s not happening.
What is likely, however, is for people to start training LLMs and create irreversible damage to the environment because it takes an incredible amount of energy to compute— thousands of metric tons of CO2 is released when training an LLM. They also can be used to generate misinformation or harmful content in the hands of the wrong people in industrial scales— it wouldn't be hard to destabilize even governments with one.
AIs don't need to be sentient to commit irreparable damage to marginalized communities, or the environment— the real threat is what AIs can already do. Instead of pausing research for a period of time, I think it’s much more important to acknowledge what they’re capable of doing now and how delicate our society is in the face of this threat.
Good to hear that our AI overlords aren't out to get us!! Before we conclude, I know that we’re hiring right now, and more positions will open up soon— is there anything else you’d like to say about working here at Seek?
That’s a great question! I’ll have to say, if you’re interested in working at Seek, and if you’re BIPOC, neurodivergent, or queer, please contact me! AI isn’t AI without BIPOC and queer people— if we’re trying to imitate ourselves and build representation for “humans”, we have to include everyone in society, we can’t single out a huge portion of the population. Seek is a very, very diverse workplace in every aspect you can think of. I myself am neurodivergent and nonbinary; and at Seek, we all advocate for each other. We’re constantly seeking more talented individuals from diverse backgrounds to join our team— that’s what makes us who we are. We even have people working across three continents and time-zones!
💙Want to join our team? Check out our job openings here 💙