Sentry: Focus on Features, Not Bugs

to drive >> the the margin out of this. Logs are an essential part of just about any tech startup. You need to keep your eyes on how your product is being used. And you definitely need to understand what happened when things go wrong. But logs are notoriously messy. It's especially hard to gather the insights you need when your logs, errors, and performance data all live in different tools. But now there's a solution. Sentry. Sentry's logs are trace, connected, and structured, allowing you to follow everything clearly and understand the context, even if you're a non-technical founder. Whether you're debugging your front end, back end, your mobile app, whatever it is, Sentry will give you the context you and your team need to get the problem fixed and to get on with your day, you got other things you got to focus on. That's why more than 4.5 million developers are already using Century, including amazing high-profile teams like Disney Plus and Anthropic. Learn more by going to century.io/twist and use the code twist to get $240 in Century credits. That's s nt r y.io twwist. >> That's what I'm looking for. There was a moment in this story where you were talking about give me the quote one more time about we hear loud and clear. Oh yes. That people are and then segway into the Harvey story with that. >> Yes. Uh here's it's from Aaliyah House. She's open AAI's head of product policy. We've heard very loud and clear from businesses that this is important. They often have their own commitments that they have to make that they have made to their customers. If you think about somebody who is serving other enterprises who are trusting them with incredibly sensitive data and so that does segue very neatly into our other big story today which is Harvey the legal AI company released their very first in-house proprietary model. It helps Harvey software take on more tasks that were typically performed by lawyers. It is Jason a post-trained openweight model built on top of Kimmy K3. So, uh, we did we did, as you as you requested, we did pull some clips. This is sort of what you've been predicting all along, which is application layer companies, companies that have this incredibly uh valuable expert data. They don't want to turn that over to the anthropics and open AIs of the world. They're going to use it themselves to build their own models. >> Yes. And uh, did you say you have a clip? >> I We have two. You pick your pick. We have a twist clip on this and we have an all-in clip on this. Dealer's choice. >> All right, I'll take a victory fap, but [laughter] let's keep it short here. Play it at like one and a half speed. >> Let's get moving through this. >> Let's take a look. Well, yeah, here we'll pull up the the the twist clip here, I believe, first. >> Thank you for putting it on 15. Need a little bit of a haircut there. You need to start working on frontier model. You need to get off the frontier models and use open source ones and own your content and not educate them to the extent you can. And I believe that'll be the trend of 2027 is startups are already doing it. But if you give Sam Alman, who is a, you know, sharp elbow guy, and he's got to figure out how to fill in a $1 trillion market cap, he's going to do exactly what Anthropocate or Microsoft or Facebook, which is he's going to look at the applications coming in, all of those Y cominator companies who give who take that deal. They're studying every one of their token usage. They're studying what they're doing and then they will pick the top five in terms of success and incorporate it as free product into their platform. This is your final warning. Don't trust the platforms. When somebody comes to you with free tokens, you know, free anything, there's no free in life. There's no free beer, there's no free pizza. There's always a price. And then the second one's from All In. Let me do a big victory fat bear. >> Oh yeah. This one was right before July 4th. Actually, this was the July 4th all in. As you could tell from your casual, laid-back Honolulu vibes in the clip here. >> Oh yeah, that's when I had my beard and everybody loved this when I was in Hawaii. >> I got to say, I think this is a good look. I'm not just trying to butter you. I think this is I like this. This is a good J Cal. I think you should go with this one for a while. >> What do I like a Tom Celich type situation in Hawaii? Five Hawaii 5. >> It's a little bit It's a little Magnum PI. Yeah. >> Okay. Magnum PI >> to the platform. There is no free pizza. There's no free beer. When somebody like Sam Walman comes to you and says, "Here's some free tokens. Your alarm should go up." Zuckerberg did the same thing. He said, "Hey, I'm going to give people a bunch of access. Going to give them money. Come to the Facebook platform." Nobody who went to bed with Microsoft in the 80s, Facebook in the 2000s or Sam Alman now in in the 2020s did not wake up with their throat slit. This is a message to founders. If you partner with any of these people, they will slit your throat and take your business wholesale. There is nothing to discuss here. Don't trust them. Use your own models. >> Yeah, we'll link to it. You guys can go check it out. Um, there's these are just the two that we could dig up 10 minutes. >> We're This is the This is the system I want to build next is like I need to have an archive of twist and an archive of allin with all of my opinions and then I want our agent >> to look at how right or wrong I was and give me two [laughter] lists. >> Here's the Here's where you blew it. Here's your Professor Galloway moments. >> This is your Professor Galloway. You got >> Jim Kramer moments. Yeah. >> This is my Yeah. My reverse Kramer, my professor cold takes. You totally blew it. And then here's your best J on fire. A nois index. >> Yeah. >> And a and a professor Galloway index. Give me those two. >> I do for this week in AI. I do have a full like a clawed skill that we've trained on every episode and I can just go in and ask it anything and it's brilliant. It it is a little timeconsuming, >> but I want the bot to go through and say give make a summary of Jason's opinions, his hot takes from each episode >> and then have an index just of the takes, not the full transcript. So then once you have the takes, then take the takes, how many times he said it, how did his take change over time. >> Rate the takes. It's a takeator. >> Rate the take. Then I want you to build how did it change over time. So did my opinion on Apple, Facebook, whatever change over time? >> What were the key moments? You know, was I hypocritical? You know, it's then >> based on the last 6 months of news or three months of news of what actually happened with Apple's product. Did they ever launch a new iPhone? Remember, I was very critical of Tim Cook for a long period of time. Stock went up. He bought back all the stock. Bought back half the equity in the company. I understand like hundred billion dollar or $200 billion in buybacks. >> They never really came out with a a killer product post the Steve Jobs road map. So, you know, right about that, no product ever emerged, wrong about the stock price because it went so I I need that. So, for my team, >> if you can do that, that would be amazing. And we're we're working on this kind of stuff. So anyway, yeah, >> what what you're seeing with the Harvey thing, I said this >> two years ago, if you can find the two-year clip where I said when I was in Tahoe skiing >> and I think you pulled it for a previous all-In and I said two years ago, there's a chance that op that open source is going to win the AI race. I said it two effing years ago. I think what we learned in 2023 was that the language models are starting to hit parody very quickly and that the real value is going to be in and it may even become commodities and open source may win the day. So then I think the winner is folks who have the training data. Give myself a pat on the back that you heard it live. >> That clip was like I was shocked by that. I was like really like in the age of chat GPT 3.5 2.5 I said on the pod filibuster here a bit I said on the pod open source is going to win the day and we we I think we talked about that four weeks ago on all in or five weeks ago that was like a very interesting moment because that's what's happened now the the evidence so back to my like Jason's takes and just holding me accountable to my takes am I professor Ice Gold takes Prof and Macy's is going to beat Amazon and Uber and Tesla and Robin Hood are going to zero and buy foreign stocks, not US stocks. Like all those terrible takes. I need to know when I had a bad take and when I had a great take, like a Nostraconist level take. That was a Nostraconist moment for me. Um, and anyway, we'll we'll >> we'll find it for we'll find it for the VOD, folks. I don't know if I can pull it up here live. So I had mentioned when I talked to lovable and 11 labs, they're huge, huge customers of the frontier labs. Frontier labs can get competitive with them just like Figma and Design, Cursor and Claude Code, Figma and Claude Design. They are I I kind of asked them on those interviews when I was in Paris in July, hey, are you making your own models? Are you doing that kind of stuff? And they're like, yeah, you know, we're looking at that. Here it is. Harvey, pull up the story. Harvey is making their own model. They forked an open source model. They're doing their own training and they are working on behalf of their clients. This is the story. Over the past six months, Harvey's research agenda has focused on two goals. One, building frontier legal intelligence using openweight models. And two, creating systems to allow law firms to build their own specialized models and their own intelligence. Today we're sharing an update on that research effort, including initial results from our first post-trained model, which we're calling Harvey Tenant. >> Harvey Tenant is a Kimmy K3 base that we post train together with fireworks research for long horizon legal work. In addition, it incorporates harness improvements to make training and test execution more effective. Our initial work shows promising results for both performance and cost efficiency.