🚀 Fable 5 Released — But With Strings Attached
Meta's Fable 5 has officially been released to the public, marking another milestone in the ongoing race for AI dominance. However, early analysis suggests the model may have been "nerfed" — industry parlance for deliberately limited or restricted — in specific areas. While the rationale behind certain safety guardrails is understandable (preventing bioweapon designs or system hacking seems reasonable), the lack of transparency around these limitations has raised eyebrows.
The Wall Street Journal framed the broader debate succinctly: "The battle over how to tame AI has just begun." Questions are swirling around model access, government involvement, and the philosophical tension between building unfettered intelligence versus maintaining control. The concern? If a model is restricted from generating dangerous content, but users can simply prompt it to build another AI that can generate dangerous content, the restrictions become circular and ineffective.
Meanwhile, Together AI's open-source infrastructure continues to push the envelope on what's possible when building less constrained systems. The debate over fettered vs. unfettered intelligence is far from settled.
🏛️ OpenAI Proposes Handing Trump Administration a 5% Stake
In a move that's equal parts strategic and controversial, OpenAI is reportedly in early talks to grant the Trump administration a 5% equity stake in the company, according to the Financial Times. The proposal would potentially extend to other frontier AI labs as well, as part of a broader effort to ease regulatory pressure and curry favor with the current administration.
The mechanics of this arrangement remain murky. Would the equity go to a sovereign AI fund? Directly to the White House? Or be distributed to American households via the Trump accounts (a direct-to-citizen payment mechanism similar to Social Security)?
Joe Weisenthal offered a tongue-in-cheek counterproposal on X: "Rather than equity stakes, why not make companies pay 20% of all pre-tax income to the federal government and then instead of exercising shareholder influence, politicians and regulators could set rules on corporate conduct across industries."
Dean Ball, a policy analyst, laid out two potential paths for distributing the stake:
- Option 1: Divide the 5% across all US households, giving each a direct equity stake.
- Option 2: Hand the stake directly to the government — a scenario Ball described as "akin to inviting rats to live and reproduce in the walls of your house."
Ball warns that direct government ownership would lead to political capture, governance nightmares, and scope creep — with the stake inevitably expanding beyond 5%. The more optimistic scenario involves equity distributed through individual Trump accounts, potentially concentrating benefits on specific cohorts (e.g., children born in a given year) to maximize per-capita impact.
Quick math: If OpenAI is valued at $1 trillion, a 5% stake equals $50 billion. Distributed across roughly 130 million US households, that's about $385 per household — meaningful, but not transformative. However, if narrowed to a smaller demographic (say, all children under 5), the per-capita value could climb into the thousands.
The obvious context here: doing business with the Trump administration has been very lucrative for some companies. Intel's stock surged over 400% after the government took a stake. Conversely, being on the administration's bad side has been unambiguously painful. OpenAI appears to be hedging its bets.
🥪 Jersey Mike's Files for IPO — Blackstone Gets a 50% Markup in One Year
Jersey Mike's, the sandwich chain founded in 1975 by a 17-year-old named Peter (not Mike), is going public. The company filed for an IPO targeting a $12 billion valuation, a tidy 50% markup from the $8 billion Blackstone paid to acquire it just last year.
The origin story is pure Americana: Peter Cancro, then a high school football player in New Jersey, borrowed $125,000 from his football coach (who also happened to be a banker) to buy a local sandwich shop called Mike's Subs in Point Pleasant, New Jersey. Adjusted for inflation, that's roughly $775,000 in today's dollars — a staggering sum for a teenager to borrow, let alone deploy successfully.
Fast forward nearly 50 years, and Jersey Mike's now operates 3,300 locations across the US and Canada, with plans to open 300 additional shops in the UK and Ireland. Cancro transitioned to chairman this April, handing the CEO reins to former Wingstop CEO Charlie Morrison. The ticker: JMK.
📄 The US Government Just Digitized Retirement — By Firing the Vendor and Rebuilding from Scratch
In what might be the most underappreciated news of the week, the Office of Personnel Management (OPM) has successfully digitized the retirement process for federal employees. Until now, the entire system was paper-based and manual, with processing times stretching up to six months.
Spike Brem, an early Airbnb engineer recruited by Joe Gebbia (Airbnb co-founder and current US chief design officer) to embed at OPM, described the Kafkaesque reality in a detailed post:
"Deep underground in caverns blasted out of limestone, there were literally acres of file cabinets as far as the eye could see, storing files detailing federal employees' employment and paystub history. A simple case might be a quarter or a half-inch thick, but complex cases filled up whole filing cabinets. One famously took up a whole pallet."
Case workers — sitting in cubicles deep underground — would manually process each file, hand-checking calculations and keying data into a Cobalt mainframe built decades ago. Files were literally walked from one worker's outbox to another's inbox, sometimes sitting idle for days.
The team initially tried to salvage a project that a software contractor had been working on for over a year (with another year to go). But the quality was so poor, they scrapped it entirely and rebuilt the entire stack from scratch, working day and night to hit early milestones. The project is now complete.
Key takeaway: "The primary skill these vendors have is securing government contracts. It's a huge moat. Delivery of quality product be damned."
📱 SpaceX Reportedly Demoing an AI-Focused Smartphone Prototype
According to the Wall Street Journal, SpaceX showed investors a prototype of an AI-enhanced smartphone during its recent IPO roadshow. The device is described as slimmer than the iPhone, running a proprietary operating system, powered by xAI technology, and built around Qualcomm's Snapdragon chipset.
Elon Musk has repeatedly denied plans to build a phone, once posting: "The idea of making a phone makes me want to die. But if we have to make a phone, we will." More recently, he flatly stated: "We are not developing a phone."
And yet, here we are.
The strategic logic is compelling. SpaceX owns Starlink, one of the most advanced global satellite networks ever built. Over time, theoretically any person on the planet could have an internet-connected SpaceX device. Why wouldn't Musk take a shot at hardware?
The other factor: the App Store moat is eroding. Large language models are increasingly replacing standalone apps for simple tasks (weather, surf reports, calendar management, etc.). The long tail of niche apps — once a key iPhone lock-in mechanism — is becoming less relevant. As one panelist noted:
"I don't even need the weather app anymore. I can just ask ChatGPT for the forecast. There are so many apps I used five years ago that I don't touch today."
If the App Store's gravitational pull is weakening, the barriers to launching a new device drop significantly — especially if the device is tightly integrated with AI-first workflows.
🧥 Jensen Huang's Signature Leather Jacket Hits Auction
Nvidia CEO Jensen Huang's iconic Tom Ford black leather jacket is being auctioned by Sotheby's, with an initial estimate of $40,000 to $60,000. Industry observers expect the final sale price to go much higher, given the jacket's cultural significance and novelty value.
In related news, Meta employees consumed 73.3 trillion AI tokens in a single month, costing roughly $221 million per month or around $2.65 billion annually. (Some analysts think the actual figure may be higher.)
Nvidia has also been striking revenue-sharing deals with AI cloud firms, taking the concept of "GPU sale-leaseback" to new levels of financial engineering. These structures are common in other industries but represent a novel development in tech, reflecting the increasingly industrial nature of the AI buildout.
🎆 TVPN Signs Off for July 4th Weekend
No show tomorrow — it's a long weekend. Leave five stars on Apple Podcasts and Spotify, and sign up for the newsletter at tvpn.com. See you Monday.
Cortisol rising. We love you. Bye.
