Ever wonder what it’s like to work on cutting-edge advancements in natural language processing (NLP)? Well, you’ve come to the right place — welcome to our fourth team profile article! In this blog post, we interview Behzad Golshan, a Senior Research Scientist at Megagon Labs. Behzad discusses how he began working at Megagon, the most exciting developments in his work right now, and the best spots to relax and enjoy some coffee in the Bay Area.
How did you arrive at your current role at Megagon?
In 2016, I was finishing up my Ph.D. in Computer Science at Boston University. I realized I wanted to continue working as a research scientist, but I was having a difficult time choosing between staying in academia or joining the tech industry. I started applying to both types of roles and hoped that the correct decision would reveal itself along the way. Thankfully, it did!
I learned of Megagon Labs through my advisor and was immediately captivated by the organization’s work. I began talking with employees here and, against the advice of many friends, even let them know about my dilemma. Instead of receiving a pitch on why I should work at Megagon, I got invaluable guidance on how to make the best decision possible. It was then that I knew I wanted to join the company — I couldn’t ask for more supportive and friendly team members who sincerely care about one another.
What's the most surprising aspect of your recent work here at Megagon?
Recently, I’ve been working on common-sense reasoning abilities of NLP models. And the more I work on this, the more fascinating it becomes. My main takeaway so far? Humans are really good at common-sense reasoning. But computers (and even our best models) are still terrible at it!
Here’s an example to illustrate this: “The principal asked Jane’s stepmother to come to the school immediately.” Humans infer so much from this single sentence:
Jane is presumably between 6- to 18-years-old.
This event is probably not happening on a weekend since the school is open.
We learn more detail about the relationship between Jane and her mother.
It’s likely that Jane either did something wrong or was involved in an accident.
I’m amazed by how much we can understand from so little information. Now, one question remains: How can we enable machines to do the same?
Are there any exciting work events or developments coming up?
Wow, this is a hard question to answer. There’s so much happening in NLP right now; progress is occurring at an unprecedented rate. It’s exhilarating to be working in a field that’s moving so fast.
To put this in perspective, we see the release of a newer, bigger, and better model every couple of months. But do you want to know what I find most interesting about this? When you apply these models to build or improve a product, you start to see their weaknesses.
For instance, a small change in your input text can result in a completely different outcome from the model. That’s not exactly desirable from a product standpoint. But it’s a great reminder that there’s still vast room for improvement in NLP.
What advice would you offer to aspiring research scientists?
Yes — the desire for success can be a double-edged sword. It pushes you forward, which is great. But it can also be a source of pressure and frustration. I believe that people who aren’t obsessed with success are generally happier, more productive, and, ironically, ultimately more successful!
So concentrate on doing good work. Don’t stress about success. If you focus on the process and not just the results, it will come soon enough. This not only goes for research but numerous other areas of your life.
What’s the most underrated activity or place in the San Francisco Bay Area?
I love going to see the ocean! It has such a calming effect on me. Admittedly, this may not be an underrated activity. But I think many locals don’t do this enough because of the driving required. Spoiler alert: It’s totally worth the drive!
Grab a cup of coffee, maybe a pastry or two, and head to Lands End or Baker Beach in San Francisco. If you’re okay with a little longer commute, then take a trip to Half Moon Bay. Basically, any beach in Santa Cruz will do. After you arrive, find a quiet place to relax, sip your coffee, and enjoy the view!
We hope you’ve enjoyed this special interview with Behzad! Check out our blog to see other team member profiles and learn more about Megagon’s recent work!
Written by Behzad Golshan and Megagon Labs