Between 2 Servers
Parse & Prejudice w/ Justine Gehring
Wendy: Hello and welcome back to another episode of Between 2 Servers. I'm your host Hayden Baillio. I'm here with Justine–
Host: Wendy, what the hell?
Host: This is my show we talked about this already.
Wendy: Oh..you left your notebook and I just thought I would–
Host: You just thought you what? You're the set designer, get out of my chair. Give me my book! This is awkward but um...hi I'm I'm so sorry. Hi my name is Hayden Baillio and this is another thrilling episode of Between 2 Servers. I'm I already said my name, damn it. Um, I'm, anyways I'm here with Justin.
Guest: Justine.
Host: What?
Guest: Justine.
Host: Yeah what I said. Um, thanks for coming on the show.
Guest: My pleasure.
Host: Sorry that really threw me off the whole thing.
Guest: I understand.
Host: Um so you research um, ML for code and G and N's is that actual science or just a really elaborate way to avoid writing documentation.
Guest: Yeah, it's real science, it's a lot of math and how it ties into code and all that but...
Host: Yeah MC squared and stuff like that.
Guest: Yeah exactly, every day.
Host: Yeah, two plus 2 and things.
Guest: Yeah.
Host: I know math yeah, yeah. Okay, okay. Your talk is about AI being a force multiplier. In my experience, that force is usually anxiety. Can you comment on that?
Guest: I think the whole plan is to remove that anxiety by using AI at scale but you know AI has good, AI has bad.
Host: Wise words.
Host: You got your master's degree from um McGill and Miller, Miller, Miller...
Guest: Yeah Miller.
Host: Did you also get a master's in making simple concepts sound terrifyingly academic or was that just a minor that you took?
Guest: That was just a minor.
Host: Okay.
Guest: A specialty.
Host: Yeah.
Guest: A certificate maybe even.
Host: A certificate yeah a boot camp she did a boot camp for that one I think.
Host: Your talk mentions lossless semantic tree for Java is that just a fancy way of saying we parse some code or does adding the word lossless double your research funding?
Guest: No there's actually extra stuff in there hence, the lossless it's full Fidelity that's why we know, we know.
Host: Fidelity.
Guest: Fidelity, Full Fidelity!
Host: Oh.
Guest: Yeah.
Host: I also know about Fidelity. The bank.
Host: As well. You work in large scale code refactoring on a scale from slightly uncomfortable to existential dread how do developers react when you tell them an AI wants to rewrite their entire code base?
Guest: I mean that's the fun part. It's not the AI part that rewrites it. It's all deterministic. All those changes are done automatically, but without the AI so.
Host: Is that what you tell yourself?
Guest: That is what I tell myself.
Host: Okay.
Guest: And that's what my sales tell me to tell people.
Host: That makes sense I met one of your sales person–people. He was....
Guest: Tall?
Host: He had nice white pants at least um... you talk about AI having agency and autonomy are you preparing developers for the robot uprising or just for unemployment?
Guest: You know, it's all about helping people be more productive, and leaving you more time to do the fun stuff in code, and that's that's where the AI comes in.
Host: More time to do the fun stuff in life because you don't have a job right?
Guest: Hopefully that's not where that'll end.
Host: Maybe not. Open rewrite recipes sounds like a Food Network show where algorithms make dinner how often do your AI recipes serve up something completely inedible?
Guest: It's funny when I do you know, try to rephrase an email or something using, you know, an LLM. They'll sometimes think I'm talking about actual recipes about food and stuff but...
Host: Make sense.
Guest: Yeah.
Host: What's your favorite recipe that it's come up with?
Guest: Recently, I've been trying to get it to say something about carbonara, I don't know why.
Host: Yeah.
Guest: That's that's the recipe, yeah, but it's…
Host: Yeah.
Guest: It's one time out of two, actually give me a recipe.
Host: Do you make your carbonara with cream or no cream?
Guest: No cream.
Host: Thank you.
Guest: No cream.
Host: That's the best answer you could have made. Your bio says you research generating code under challenging circumstances is challenging circumstances code for the AI has absolutely no idea what it's doing but we ship it anyway?
Guest: Yeah, yeah, yeah, that was my thesis, yeah. Haha, in my grad school.
Host: Makes sense.
Guest: It's a tough problem.
Host: You mentioned using AI in beddings to get a bird's eye view of your code base is that because the view from ground level is too depressing?
Guest: It can be daunting.
Host: This episode is just all off right now.
Host: Okay, Justin how many developers have threatened to quit after hearing your presentation on autonomous AI refactoring? Just ballpark it for me.
Guest: Zero.
Host: Surprising.
Host: Your talk promises reliably using AI at scale in the AI world does reliably mean work 60% of the time all the time or is it just in demos?
Guest: Yeah there is a pandemic of of demos AI being only good for demos I hope we're not one of them I feel like it's actually useful but time will tell.
Host: Alright.
Host: As someone that's overseeing a partnership with an AI institute, have you considered that the first thing the AI will refractor is probably your job description?
Guest: And I would welcome that change. Maybe you'll give me a promotion in that description.
Host: That would be nice that would be nice. You can go back to the fun things of life.
Guest: The fun parts of coding.
Host: Because you'll be unemployed right.
Guest: [laughs uncomfortably]
Host: Yes.
Host: Well Justin, I really appreciate you uh. I hope that your talk went well or goes well at this conference. Depending on when it happened. I won't be there.
Guest: It went well, so.
Host: I didn't ask. Thank you, um. But uh, thank you so much.
Guest: Thank you. I appreciate it.