Venture with Grace
00:44:56 · Ori Goshen
[00:44:56] S1: I think the process of defining what you want, defining kind of the rule of the game, what's valid, what's not, what's what good looks like, how do judge a good output, how do you judge a good outcome, is the bottleneck. And I think even if you have good conceptual understanding of that, translating it into an eval and an eval framework is hard. And and I think that's... And then and then you need to apply all the different training algorithms, right, and know which one to apply when and which would be effective when and so on and so forth. So that's kind of the third bottleneck. But what we've been... And that goes back to your scalability issue. What we've been building ourselves is actually a platform that identifies the work of a researcher in the company. So every FDR in the company, he has two responsibilities. First and foremost, make sure the customers are happy and are getting huge amount of value from their work. And second is that they're being educated because part of the value is just by doing this, the company is doing it and being involved in the process. Now they suddenly create that skill. The third job of FDRs is to identify themselves. How would they, when they go to the next deployment, are able to streamline to make that process much more, much faster, much cheaper? How they when they solved an issue, when they applied an algorithm in a very specific way, how do they create that skill, if you will, so that when they come to the next deployment, they can repurpose, they can reuse that skill. So we're doing this on an ongoing basis, actually what we've seen where, let's say, mid year projects of pretty complicated RL training, you know, would take several months with all the debugging and the reward hacking and all the things that you need to address there. Now it takes a week or two for our experts using those tools. So that's the way you make it scalable. And it's less a question of this industry or the other, because we see a lot of similarities in the types of problems. Definitely, if you kind of translate them into the technical language, it's more about, you know, how do you, a sense, platformize and how do you identify the process so you can actually scale. And that's what we're seeing in the field.
00:48:21 · Grace Gong, Ori Goshen
[00:48:21] S0: I I find it... Like, I can totally see it because I feel like nowadays, like, because of, like, it's really hard to define, like, scope of, like, Wordtune, Wordtune AI. And, like, lot of times, like, there's, like, human in the loop and then there's, like, you know, your agent have to talk to, like, different agents. Like I'm personally trying to figure out how to like actual, actually have like agent connectivity, like your Codex can talk to your, know, hot coder or something. [00:48:46] S1: When [00:48:49] S0: guess, like when you were thinking about ensembling your own team, so obviously I heard you guys have some really great talent. Yes. Why don't we talk about the hiring aspect of things? How do you evaluate the talent? And I think from East Of Rio, maybe like the culture is different than like The US. In The US, obviously, like if you work at like any of the frontier lab that you're good. Or like, you know, you go to school or you're training, including blah, blah, blah. Like, is there like a hiring hack beside people who went to like 8,200? Like, that's kind of like a... [00:49:26] S1: Yeah. 8,200 is a good hack. Another one is... [00:49:31] S0: You're just hiring the alumni. It's like, do you just go to, like, the school? Or, like, I don't I don't think it's a school. It's like a kind of like an intelligence services. [00:49:37] S1: Yeah. It's [00:49:37] S0: a... Yeah. So you just go there and then be like, hey, alumni hiring.
00:49:37 · Grace Gong, Ori Goshen
[00:49:37] S0: a... Yeah. So you just go there and then be like, hey, alumni hiring. [00:49:40] S1: Yeah. And it's also a lot of friends bringing... Like, friends. Friend of friends. Friend of friends, exactly, type of hiring naturally. There's one thing to caveat, there is a lot of cybersecurity talent. [00:49:54] S0: Yeah, yeah, In Israel. All the top cyber company kind of like you have to have the badge too.
00:49:54 · Grace Gong, Ori Goshen
[00:49:54] S0: Yeah, yeah, In Israel. All the top cyber company kind of like you have to have the badge too. [00:50:00] S1: Yes. Yeah. And it's a very long conversation. And now also cybersecurity, I think, is changing rapidly and it's becoming more closer to AI, especially if you're trying to kind of address all the threats that are AI powered in a way. So I think that landscape is changing as well. But going back to your question, I think, 8200, there's a lot of very creative, very strong technical folks. Also, within the academic groups, People should not discount the academia. There's a lot of very good research groups and people that took that path, And there is some really good talent there as well. And another thing I would say, Israel is a very established ecosystem in hardware and low level software development. So really hardcore storage and networking. And some of the technical people there, you know, that have done various very algorithmic challenging type of roles, not all of them have the deep learning background or kind of they've used like modern AI type of, they've done modern type of AI development, but the transition is, and actually some of our strongest researchers and engineers came from that background. They used to be doing storage and networking and all these types of, you know, more legacy work. So so, yeah, I think over there, if you go to these companies, in some cases, these large corporations, there is a very strong talent that could be applicable in AI as well. So that's another act.
00:52:13 · Grace Gong
[00:52:13] S0: Would you say, when you're thinking about, like, ensembling the team, like, do you find it, like... I guess, like, how do... Do you train them since you mentioned about, like, the people coming from, like, legacy kind of, like, tech background? And do you feel like you have to train them with AI or do people just adopting it into the new... I guess the newest technology very fast?
00:52:42 · Ori Goshen
[00:52:42] S1: Right. I think there are two aspects if we need to unpack it. One is being familiar with, you know, training deep neural networks and all that stuff. This is something you you definitely need training. And we've actually done this in the company for many years now. So we obviously use, you know, open resources and courses out there, but we also have our kind of learnings and gotchas and things that we've learned along the way. So I think that's definitely something that calls for training and we're doing that as a company. Every member of our technical staff is going through that type of training. The other thing is how to use the AI tools in that development cycle. And that I think comes to people more naturally. I think there is still a lot of gotchas there. And while obviously we're in the age where no code needs to be written manually directly by a human, still there are some really important do's and don'ts, and you easily write Slop, right? It can easily start a project and very fast, very quickly, it becomes just unsustainable, unmanageable, very hard to maintain, that creates a problem. And we've actually experienced that pain ourselves. There are tools that we built that we decided, okay, we'll vibe it and so on, and we pay the price. So it's a very delicate balance. I think the entire industry is learning how to manage that balance, how on one hand leverage those AI tools that are essentially becoming must have, right? You can't, I would say you can't live with them, you can't live without them, right? So how do you do that? How do you leverage? How do you enjoy the benefits without creating a lot of slop and kind of going through that slippery slope, if you will, of code. So... And I think we have... We've built a lot of skills within the company and a lot of very good processes to make sure we maintain that, but we improve over time. But it's something that is an ongoing improvement and learning curve for all of our engineers, unrelated to what they actually work with, whether it's a tool or code for training a neural net or code for, you know, inference. It's it's... I think the the same the same logic, the same rules apply.