Why Teams Will Break This Year (and How to Fix Them) | Battery Venture’s Dharmesh Thakker

**Ben Lloyd Pearson** (0:08)
Welcome to Dev Interrupted. I'm your host, Ben Lloyd Pearson.

**Andrew Zigler** (0:11)
And I'm your host, Andrew Zigler. And Ben, what's on your mind this week?

**Ben Lloyd Pearson** (0:17)
Yeah, so I have something to confess. I have a credit addiction. I signed up for a new AI service that promised all sorts of agents and other things that would do various tasks and solve problems for me, and they gave me a whole bunch of monthly credits to hire these agents. But it really sent me down this rabbit hole of the future of work and how we value things, I think. Are we going to be tokenizing ourselves or our work or our tasks that we carry out every week? Basically, everything is priced in tokens now, and hard work costs more tokens, but you never really actually know what the cost is until the prompts have already been sent. So yeah, somehow I find myself wondering, how can I maximize the amount of tokens that I spend? Because spending more tokens means you're accomplishing more work, right? But anyways, maybe we'll talk about this more later. Let's dive into this week's news.

**Andrew Zigler** (1:14)
Yeah, we can dive into this week's news. I got a great one about OpenAI CEO Sam Altman in a recent Reddit AMA. But before we get there, this token thing you're talking about is stuck in my brain. It's so funny to me because if you think about it in some ways, when you go and you have your tokens and you're trying to get the result you want, it kind of becomes like the modern loot box. Like you have all these tokens and you're trying to get that rare shiny thing at the end and it only comes out every once in a while so you got to put a lot of tokens in the machine. And so the idea that doing more work, spending more tokens means that you're more valuable, that's fascinating. Maybe it's about the things that you can't get out of the machine that you put the tokens into.

**Ben Lloyd Pearson** (1:55)
I wonder how many tokens it's going to cost for me to have an agent that figures out how to best spend my tokens.

**Andrew Zigler** (2:01)
Well, I really hope that people will let us know in the comments on Substack this week about how many tokens that they think that would be worth. But now to move on to this week's news, I do want to cover an interesting one from a Reddit AMA. Now we all know Reddit AMAs, right? Where somebody in a high position somewhere, they descend into the throng of the masses and they answer questions for the common folk. We had one last week from Sam Altman, and there was someone who asked, would you consider releasing some model weights and publishing some research about it? And in this AMA, Sam indicated that he thinks that they're on the wrong side of history right now when it comes to open sourcing models and technology around AI, and indicated that that's something that they might do in the future. Now, of course, in Reddit AMA, you should take everything with a grain of salt, but it definitely sparked a lot of interesting conversations online.

**Ben Lloyd Pearson** (2:55)
Yeah. Honestly, my opinion is that the risk of open sourcing these models in some of these ways is probably extremely low. Like the real commercial value comes from all of the training and the reasoning that they add to these models. So I do kind of agree that he's probably is on the wrong side of history, and we are already seeing a lot of innovation and rapid iteration coming out of the open source space on this. But so yeah, I kind of feel like this is a no brainer.

**Andrew Zigler** (3:22)
When it comes to the open sourcing AI, you know, it's a complicated formula. It's not like other open source technologies in the past where you release the source code, you release it under, you know, a license that may or may not be permissive, that may or may not let someone build something of commercial value on it. But ultimately, when you release those things, someone else can replicate your success and carry the torch of that technology forward. But with AI, when you open source it, it's a little more complex, because you might open source the model and even maybe the weights that you used. But if you don't open source the training code that you use or the hyperparameters you use to get it into that state, or if you don't share your training data so that folks can understand what went into the model and then maybe further train it, then you're only giving them a really small part of the equation. And when companies release their technology in that way and call it open source, they still hold all the keys because they're the ones that can run that model, continue to train it and make it succeed at scale.