This Day in AI Podcast
We Committed Fraud with OpenAI's New Image Model (and Called Mum) - EP99.38
Why It Matters
Understanding the subsidy-driven pricing model is crucial for businesses and developers who plan to scale AI usage, as the shift to true-cost pricing could dramatically affect budgets. The episode highlights that model choice should be based on actual performance and cost efficiency, not brand hype, offering listeners actionable insight as the AI market matures.
Key Takeaways
- •OpenAI released GPT‑5.5, unavailable via API at launch
- •Anthropic’s Claude Opus 4.7 outperforms GPT in agentic tasks
- •VCs subsidize ~70% of token costs; enterprises pay real price
- •AI usage equated to hiring staff, justifying higher expense
Pulse Analysis
The episode opens with a rapid rundown of the latest AI model releases, highlighting OpenAI’s GPT‑5.5 launch—still hidden behind API restrictions—and Anthropic’s Claude Opus 4.7, which many participants claim outperforms GPT in agentic workflows. Listeners also hear praise for newer Chinese models such as GLM 5.1 and Kimi 2.6, noted for their speed and reliability in multi‑tab coding sessions. This flood of announcements underscores a shift toward “super‑app” strategies, where providers bundle language, vision, and agent capabilities into a single, always‑on product suite.
A central theme is the economics of AI consumption. The hosts reveal that venture capital and sovereign‑wealth funds absorb roughly 70 % of token costs for OpenAI, while Anthropic’s burn rate sits near 33 %. Enterprise customers are the only segment paying close to true marginal costs, creating a stark contrast with consumer users who see merely 5‑6 % of actual expenses. This double‑layer subsidy—first from investors, then from platform providers—distorts perceived value, leading many to treat AI usage as a negligible line item rather than a strategic staff‑augmentation expense.
For business leaders, the conversation reframes AI from a novelty to a cost‑center comparable to hiring additional talent. When AI tools deliver productivity gains equivalent to years of manual work, the justification for a six‑figure annual spend becomes clear. Yet the hosts warn that without transparent pricing, organizations risk under‑investing or over‑relying on subsidized services that may disappear. They note Google’s silence on its own models despite owning the hardware advantage, suggesting that market dynamics will soon force providers to balance subsidies with sustainable pricing, leaving enterprises to decide how much to allocate toward truly valuable AI capabilities.
Episode Description
Join Simtheory: https://simtheory.ai
So Chris, this week... a LOT has happened. We're back to regular programming (maybe), and back with our average takes. Nothing's changed.
GPT-5.5 just dropped today - but you can't even use it in the API. Vaporware? OpenAI is charging MORE than Opus 4.7 and we haven't even tested it yet. Meanwhile Claude Opus 4.7 landed a couple weeks ago and... the vibes are off? Mike's actually going BACK to 4.6. Something's wrong.
But the real star: OpenAI Image 2. This thing is genuinely terrifying. We committed what can only be described as "parody fraud" - faking a council letter so realistic Mike's own mother fell for it on a phone call. Then Chris posted a fake development approval with the mayor's real name into a local Facebook group and had to delete it when someone tagged the actual mayor. The forgery capabilities are absolutely unhinged.
Also: GLM 5.1 is so good Mike forgot he switched to it. Kimi K 2.6 is criminally underrated. VCs are paying 70% of your real token costs. Consumers pay only 5.5% of actual cost. The everything app war is ON. The SaaS-pocalypse is real. And we made two new diss tracks.
Chris made a graffiti sign in LA. It says "This Day in AI." It was the best artwork in the class. That tells you everything.
CHAPTERS:
0:00 - Intro & We're Back (Don't Over-Commit)
1:14 - Overview: Everything That Dropped While We Were Gone
2:56 - GPT-5.5: Vaporware? Not Even in the API
4:57 - Benchmarks vs Reality: Nobody's Excited About OpenAI Models
5:50 - GLM 5.1 & Kimi K 2.6: Secretly Just As Good?
8:15 - The Everything App Race & Product Layer War
8:56 - Token Economics: You're Only Paying 5.5% of Real Cost
13:08 - We Burned $1.5M in Cloud Credits in 2 Months
16:13 - "$30/Month Is Too Expensive" (It Actually Costs $700)
19:25 - Where Is Google?? TPUs Should Flatten Everyone
22:01 - Agentic Tasks Are 10-50x More Expensive Than Chat
25:07 - OpenAI Workspace Agents: Glorified Zapier?
27:01 - Single Agent vs Multi-Agent: How Do You Actually Work?
33:06 - Building Automation Is HARD (Our Support Shame)
35:33 - OpenAI Image 2: The Fraud Episode Begins
44:16 - FRAUD DEMO: The Fake Council Letter (Mum Falls For It)
49:16 - FRAUD DEMO 2: Chris Posts Fake Mayor Letter on Facebook
52:17 - Fake Receipts, Bank Statements & Can Forgeries Be Detected?
57:25 - Claude Opus 4.7: The Vibes Are Off
59:51 - Mythos Preview: "Pics or It Didn't Happen"
1:01:56 - 🎵 DISS TRACK: "Point 7" (Opus Destroys Everyone)
1:03:30 - Kimi K 2.6 Deep Dive & 🎵 New Diss Track
1:08:34 - The Everything App War & SaaS-pocalypse
1:13:51 - Death of Per-Seat Pricing & Agent Security
1:22:37 - Final Thoughts: The Time for Pretending Is Over
1:28:22 - 🎵 Full Tracks: " Point 7" & "Kimi You're So Fine 2.6"
Thanks for listening, like and sub xoxo
Comments
Want to join the conversation?
Loading comments...