CrusoeCrusoe gave us early access to their serverless fine-tuning product, which lets you fine-tune open models without having to deal with infra or provisioning. I thought it’d be cool to try fine-tuning a question generator using the transcripts of my old interviews. Have the models gotten so good that if they had all my research and prep, and they could look at a conversation so far, they could ask a next question better than I would? Crusoe made the implementation super straightforward. I just uploaded the data, picked an open model, and started the run. I didn’t have to touch any of the hyperparameters. Crusoe’s applied AI team maintains optimal recipes for each model, so I just set everything on auto. When the run finished, I deployed it as a self-serve endpoint and built an eval for my team. I had them choose the best next question out of three anonymized choices: one that was produced by the base model, one that was produced by the fine-tuned model, and one that I actually asked. Fortunately, my team preferred my actual questions about two-thirds of the time. Hopefully this benchmark doesn’t saturate. And in the remaining cases, they almost always preferred the fine-tuned model over the base model. Serverless inference is live now, and serverless fine-tuning goes live next week. Learn more at crusoe.ai/Dwarkesh.
Antithesis podcast ad on Dwarkesh Podcast
Automated software testing platform that simulates faults to find bugs in code. This 1:01 mid-roll ad ran on Dwarkesh Podcast, aired Oct 1, 2026.
TL;DR
- Ad length: 1:01
- Ad placement: Mid-roll (middle of the episode)
- Audience type: AI
- Episode name: Si Sheppard – How did a few hundred Spanish soldiers topple two empires?
- Aired: Oct 1, 2026
At-a-glance
antithesis.com ↗Ad copy
But with software, you can actually test them. The Antithesis software testing platform runs your system through thousands of parallel histories, letting you explore any counterfactual you want. How does your software perform if a packet gets dropped? What about when the network drags, or the machine dies mid-write? Importantly, you don't have to come up with these counterfactuals. One big problem with traditional testing is that it only covers the distribution that you know. To test for a bug, you first have to imagine it.
These tests are sort of like the Flower Wars. The Aztecs' success in these ritual battles against known rivals made them feel like the best military in the world, but they're still unprepared for Cortés. Antithesis solves this by steering your tests for you, injecting faults and making things as adversarial as possible. That way, your software gets tested against scenarios you'd never imagine, but that could still arise in the real world. Go to antithesis.com/dwarkesh to learn more.
More AI podcast ads
View all ›| Play | Brand | Podcast |
|---|---|---|
 Dwarkesh Podcast | ||
OutSystemsSo glad you're here. Our show today brought to you by OutSystems, the leading agentic systems platform. OutSystems helps their customers modernize operations by enabling them to build, modernize, and operate enterprise systems starting from any coding tool in and this is really important. A governed, there's a word you want to hear, governed agentic engineering model. It's time to innovate at the speed of AI, but without compromising quality or control, which is why thousands of enterprises worldwide trust OutSystems for those mission-critical apps. We're not talking theory here. We're talking real world experience and massive operational wins. You could find so many examples on their website, but I'll give you one. Uh, the leading global retailer that was struggling to keep up with their rapid expansion. They used OutSystems to build AI agents that automated the invoicing and translation processes, shifting 95% of their daily store operations to a new digital app, saving thousands of hours of manual work. Now, now that's an internal system, but they also can be used for customer-facing systems like the major global logistics giant that was drowning in technical debt. Their engineers are spending their weeks manually firefighting by reviewing platform logs by hand. Have you ever done that? No fun. By using OutSystems, they built an AI log review agent in just 5 days. And that one agent now saves them 1,600 engineering hours every year. Which means that team instead of sitting there going through the paper let's can focus on real innovation real work instead of reactive maintenance. Even in modern digital banking companies are using this platform to standardize and automate complex financial workflows delivering secure personalized experience to their customers at paces their competitors cannot match. OutSystems it gives you the flexibility to deploy how you want and the governance. Got to have that to do it safely. So stop chasing the hype. Start building systems that actually scale securely. Learn more at outsystems.com/twit. That's outsystems.com/twit. |  Intelligent Machines (Audio) | |
OutshiftAll right, before we dive into our rapid-fire, Paul, this episode is also brought to you by Outshift, which is Cisco's incubation engine for frontier technology. So multi-agent AI is in every enterprise roadmap right now. But here's the problem that everyone is facing. Agents can pass messages, but they can't think together. The system silently underperformed through cognitive failures like misreadings, unverified claims, and false consensus that conventional monitoring cannot detect. Now Outshift by Cisco is building the fix, what they call the internet of cognition. This is an open-source foundation for building multi-agent systems from design to production, with shared context, shared memory, and guardrails to drive the results we actually expect from AI. You can go learn more about this by reading the paper, experiencing the demo, and grabbing the code from Outshift.com. That is Outshift.com. |  The Artificial Intelligence Show | |
ScribeOur show today brought to you by Scribe. However well a process gets documented today, someone on your team will eventually find a better way to do it. But most of the time that improvement never makes it past the person who found it. That's where Scribe comes in. And that's what today's sponsor Scribe does best. Scribe is a workflow AI platform trusted by 94% of the Fortune 500. You just do the process as you normally would and Scribe captures it in real time and turns it into documentation automatically. No manual writing, no manual screenshots, no starting from scratch every time someone new joins the team. Scribe automatically redacts sensitive information like names, account numbers, emails, and as an admin, you can enforce Scribe across your entire team. Feel confident nothing slips through the cracks. Anyone following this process can launch real-time onscreen guidance the moment the Scribe is created so critical processes get done right every time. And Scribe doesn't just document your workflow. Its AI suggests ways to improve it by identifying unnecessary steps and opportunities to simplify or automate the process. It's not just capturing how work gets done. It's helping you do it better. To see what Scribe could look like for your org, head to scribe.how/machines and mention Intelligent Machines for your first month of Scribe capture free on select plans. That's scribe.how/machines. And we thank them so much for their support of Intelligent Machines. | Intelligent Machines (Audio) | |
Horizon3.aiBut first, a word from our sponsors. Horizon3.ai, the most dangerous threats are always the ones we don't see coming. The feeling of uncertainty is one of the biggest challenges in cybersecurity. You can have talented people. You can have the perfect stack. You can have solid processes. But at least I can speak for leadership at my business, there's always that one question in the back of your mind, "What am I missing? Where am I missing?" That's where Horizon3.ai comes in. Horizon3.ai is an AI-native, proactive security platform designed to show organizations exactly how an attacker could compromise their environments before a real threat gets the chance. Horizon3.ai's autonomous AI hacker, NodeZero, safely and continuously tests your infrastructure at machine speed the same way a real attacker would, finding areas that are vulnerable. It covers your entire attack surface in a single test, deploys in minutes. They have run 225,000 production-safe tests with zero downtime, so you don't have to worry. You can stop guessing and start knowing your defenses are intact. Because NodeZero's result is a loop: hack, fix, verify, repeat. It finds the problem, you fix it in one click, you can confirm it's secure, and then it starts over looking for more problems. Horizon3.ai is trusted by the NSA, by CISA, by Fortune 100 companies, so you know you can trust it, too. See exactly what an attacker would find in your environment before they do. Go to horizon3.ai/intelligent and request your free NodeZero demo. That's horizon3.ai/intelligent. horizon3.ai/intelligent. No commitment required. Results in hours, not weeks. horizon3.ai/intelligent. | Intelligent Machines (Audio) |