Unsupervised Learning · 2026-03-11 · 2 min
Key moments - from our scoring
Substance score
29 / 100
Five dimensions, 20 points each
Daniel Miessler makes a contrarian case against the enterprise trend of building small, specialized custom models. Instead, he advocates for deploying the best available general models - citing the Claude Opus, Sonnet, and Haiku tier structure as the likely future - paired with sophisticated context management. His reasoning draws an explicit parallel to human expertise: just as the most experienced security professional or email specialist brings broad life experience to their narrow domain, models with general capabilities outperform narrow specialists even on domain-specific tasks like threat detection, email labeling, or security event processing. He predicts that as SOTA models improve and costs decrease, including open-source alternatives, enterprises will gravitate toward larger general models for all organizational tasks rather than maintaining a patchwork of tiny, custom models. Miessler acknowledges this view is about 70% conviction rather than certainty, but he believes the efficiency and decision-quality gains from generalist models will outweigh the complexity of managing custom specialists.
Because specialized tasks still benefit from broad general reasoning and experience - just as the best human specialists rely on general life knowledge, AI models perform better on narrow domains when they're general intelligences paired with good context, not custom-trained specialists.
A tiered approach similar to Claude's Opus, Sonnet, and Haiku models - where the best general models at different capability levels handle all organizational tasks through context management rather than multiple small custom models.
The most experienced humans in narrow domains (security, email management) still draw on broad life experience to excel; similarly, general AI models outperform narrow custom models even on specialized tasks because they bring broader reasoning capabilities.
Improving SOTA model capabilities, decreasing costs (including open-source versions), and the efficiency of context management making it cheaper and simpler than building and maintaining custom models for each task.
About 70% convinced, not completely certain, but he believes the evidence points toward general models becoming the enterprise standard.
Our reviewer’s read on each dimension, with quotes from the episode.
The episode presents one substantive claim - that SOTA models with context management outperform custom/fine-tuned models because narrow tasks benefit from general experience - but repeats this core idea across ~120 seconds without adding layers, evidence, or counterarguments. A B2B operator learns the thesis but little beyond it.
Anytime you think you're using a small model for a small task, there's usually a whole lot more going into a given decision than just that individual area of expertise.
I think what's far more likely is more of an opus, sonnet haiku model, where the best of the best just keeps coming down in price, including going into open source.
The argument - that general intelligence beats narrow specialization - is intuitive and builds on observable industry trends (model scaling, cost curves) but is not contrarian or first-principles. The analogy to human expertise is familiar; the framing as a SOTA-vs-custom choice is directionally novel but not deeply original.
I think the future is not a whole bunch of extremely small, specialized models throughout the enterprise.
This is because most specialized tasks still benefit from the general life experience of the person doing the execution.
This is a solo commentary, not a guest interview. No guest is present to assess.
I'm not completely sure I'm right about this, but I've never been a big believer in training custom models.
The speaker cites task categories (email labeling, security events, threat searching) as examples but provides no concrete data, benchmarks, financial comparisons, or named projects. No metrics on cost savings, accuracy deltas, or deployment outcomes. Almost entirely abstract reasoning.
For example, labeling emails, writing reports, processing security events, searching for threats on a network.
I'm about 70% sure.
This is a monologue with no interlocutor, host questions, or dialogue. Conversational craft cannot be assessed without a second voice.
I'm not completely sure I'm right about this, but I've never been a big believer in training custom models.
Computed from the transcript - who did the talking, and the words that came up most.
I think the future is cheaper and Open Source SOTA models combined with context, not custom, narrow models. Become a Member: See omnystudio.com/listener for privacy information.
Transcribed and scored by The B2B Podcast Index.
WEBVTT - Why I Believe in SOTA Models Over Custom Ones I'm not completely sure I'm right about this, but I've never been a big believer in training custom models. I've also never believed in fine tuning going all the way back to 2023. My intuition has always pushed me towards the best state of the art model possible, combined with context management. I just finally crystallized my reasoning around this.
Anytime you think you're using a small model for a small task, there's usually a whole lot more going into a given decision than just that individual area of expertise. For example, labeling emails, writing reports, processing security events, searching for threats on a network. On one hand, I think these are specialized, but the fact is, the smarter and more experienced a human is who has this expertise, the better job they're going to do. This is because most specialized tasks still benefit from the general life experience of the person doing the execution.
This is why I think the future is not a whole bunch of extremely small, specialized models throughout the enterprise. I think what's far more likely is more of an opus, sonnet haiku model, where the best of the best just keeps coming down in price, including going into open source. And those smaller models are used in conjunction with context to perform all the different tasks in an organization at much lower cost. But I think they'll still be extremely general models, not tiny and narrow custom ones.
I think the Tldr here is when you think you're doing a narrow task, that narrow task is actually benefiting from a ton of general experience. And I think this applies to humans, and I think it also applies to models. I'm not completely convinced of this. I'm about 70% sure.
But yeah, I think this is the way it's going to go.
Other episodes covering the same guests and topics, from across The B2B Podcast Index.
Is RAG Dead? The Pioneer Who Invented AI's Memory Layer Answers - with Douwe Kiela, Co-Founder, Contextual AI {ICYMI}Making Data Simple · on Fine-tuning86 / 100
477. The Nitty Gritty of AI From an Attorney and AI Expert with Mike BrownThe Game Changing Attorney Podcast with Michael Mogill · on Context management81 / 100
The Evolution and Impact of AI and Machine Learning Across IndustriesBeyond the Screen · on Fine-tuning79 / 100
Episode 235: Who tf is Jevon?AB Testing · on open source models76 / 100
235. European Sovereign Neocloud - Jun26Redefining Energy · on open source models72 / 100
Albert Chun, Founder/CEO of AI Circle, on Training a Frontier Model, and Why "Everyone's B+" Without ExperienceAI for Business Leaders · on open source models61 / 100AI Predicts the Text of Answers52 / 100Most Companies Aren't Anywhere Near Ready for AI61 / 100We're All Building a Single Digital Assistant67 / 100Why AI Will Replace Knowledge Workers66 / 100AI Quality Inversion36 / 100