Podbit · God Mode Podcast
Opus 5 benchmarks well but consistently underdelivers; Fable 5 just gets it done. Ben has learned to use Fable when he has credits and accepts a worse product the rest of the time — a revealing admission about the gap between benchmark and lived experience.
Where this was said
At 32:55 · chapter starts 28:03
With ChatGPT prompting the structure, Rik runs through each lesson as a headline and Ben riffs on how it aged. The series opens with the moment Dario Amodei and Sam Altman refused to hold hands at a group photo at a conference — a petty but revealing signal of how far relations between the two labs had deteriorated. Lesson two spotlights OpenAI's decision to focus on coding agents above all else, which Ben calls pivotal: the coding-agent flywheel (build better agents → agents improve themselves → everything else improves) turned out to be the year's biggest strategic insight. Lesson three surfaces the Mythos speculation, with Ben now claiming Anthropic trained Mythos in February and almost certainly has a Mythos 2 scored in the 80s or 90s on intelligence benchmarks. Lesson four addresses the Opus nerf — Ben was among the first to notice labs quietly throttling model effort during supply crunches, a practice he traces to around February. The GPT-5.5 86% hallucination benchmark from episode 10 gets a failing grade: fifteen weeks on, OpenAI still hasn't meaningfully moved the needle. The section closes with the Codex vs Claude Code call — once a clear Claude Code victory, now a dead heat — as OpenAI's aggressive developer migration campaign on Twitter bore fruit.
Rik used Claude to analyse all 25 episodes and distil them into three recurring wars: OpenAI vs Anthropic, open vs closed source, and the Elon Corner. The 25-lesson retrospective grades every major call against what actually happened.
Ben speculated that Anthropic has already trained a Mythos 2 model scoring in the 80s or even 90s on intelligence benchmarks, but has not released it.
When AI labs are supply-constrained, they quietly reduce model effort levels in the backend. Users experience a noticeably worse product with no announcement. Ben says this has been happening since at least Opus 4.6 in early 2026.
According to a benchmark cited in episode 10, GPT-5.5 had an 86% hallucination rate compared to Grok's 17% at the time.
The fastest path to Twitter growth isn't volume — it's forming sharp opinions about how the platform works and sharing them immediately. People cluster around those who understand the rules and say so out loud.
By 2006, AI algorithms were stuck because they were being trained on almost no data. Fei-Fei Li's insight: human children see tens of thousands of object categories by age 6 — so machines needed massive data too. ImageNet's 15 million images, combined with GPU power and better algorithms, triggered the modern AI revolution in 2012.
The ImageNet challenge pitted machines against humans on recognizing 1,000 object categories. Humans clocked a ~4% error rate. In 2012, a neural network smashed previous AI performance — and by 2016, machines had surpassed humans entirely. That single benchmark created the modern AI era.
When video was added to AI training data in 2023, something clicked: machines could generate plausible motion without knowing muscle anatomy — just from watching millions of cat videos. Sora's January 2024 release proved AI had crossed into temporal, physical understanding.
AI is trained on the internet — the largest archive of human behavior ever assembled. But the most profound human thoughts, Picasso's creative flash, a private childhood memory tied to a gray cup, have never been uploaded anywhere. That's the gap AI cannot close.
AlphaGo's Move 37 against Lee Sedol shocked Go masters — no human had ever conceived it. But Fei-Fei Li urges caution: Go has fixed mathematical rules, and AI's bigger compute simply found a configuration human memory couldn't retain. That's creativity in a constrained space, not the open-ended kind.
Surface-level intuition — the kind you can describe in words — is just context, and AI already handles that. But the deeper kind: the feeling shaped by what you ate, your hormones, and a mood you can't name? That has no sensory apparatus feeding it to any machine. It's inaccessible, and will remain so until brainwave-level sensors exist.
Self-driving cars already exist. But the real robot revolution — robots assisting the elderly, fighting wildfires, supporting overworked nurses — is a 20-30 year arc, not a 2-year one. Hardware plus AI moves slower than software alone, but the impact will be civilizational.
Language AI is powerful, but humans evolved in a spatial, physical world. WorldLabs is building foundational models for spatial and 3D intelligence — letting people generate entire environments from a sentence or sketch. The applications span filmmaking, robotics training, architecture, and healthcare.
We use essential and analytics cookies to run Vuci. To understand how the site is used: Privacy Policy.
Install Vuci on your phone
Add it to your home screen for a faster, app-like experience.
Install Vuci on your phone
Tap the Share button, then “Add to Home Screen”.
A new version is available
Reload to get the latest Vuci.