Morning Edition

The Futurist

Sunday, August 23, 2026
AI & Technology Markets & Crypto Ideas Worth Keeping
S&P 500$7,674.37▲ 0.43%
Dow Jones$53,277.01▲ 0.98%
Nasdaq$26,180.46▲ 0.43%
VIX$15.13▼ 5.50%
BITCOIN$77,168.00▼ 1.31%
RIPPLE$1.51▲ 8.77%
ETHEREUM$2,422.70▲ 0.67%
DOGECOIN$0.09▲ 8.74%
01

AI & Technology

404 MEDIA

"Subtlefakes": Slightly Altered Nonconsensual AI Images Are Taking Over X

A new category of AI-generated abuse imagery is emerging — slightly modified versions of nonconsensual deepfakes designed to evade detection. The technique exploits the gap between what moderation systems flag and what human eyes recognize as harmful.

SIMON WILLISON

Linus Torvalds on AI-Assisted Debugging

Torvalds used AI to help crack a debugging session he called "from hell," crediting it with handling significant grunt work. Notably, the AI repeatedly declared the problem "impossible and unsolvable" — and was wrong every time.

A16Z

The Case for AI That Improves Itself

Former Google and Anthropic researchers at Mirendil are building AI systems designed to meaningfully contribute to their own development. The framing shifts AI from productivity tool to recursive infrastructure — systems that accelerate themselves.

404 MEDIA

Township Passes Electrical Moratorium to Block Nuclear Weapons Data Center

A local township is using a six-month freeze on major electrical infrastructure as a last-ditch effort to stall a data center built to support America's nuclear arsenal. It's a novel legal maneuver in what's becoming a broader battle over where AI-era compute gets built.

SIMON WILLISON

More Than Just Code Review

Willison argues the core skill for working with coding agents isn't reading every line they write — it's confidently directing and verifying outputs through multiple lenses. A quiet but important reframe of what "human oversight" actually means in practice.

404 MEDIA

Brand Hype Has Existed Since the Bronze Age

Bronze Age Arabian pottery shows all the hallmarks of modern brand strategy: consistent materials, distinct visual identity, and cultivated cultural reputation. Marketing didn't start with Madison Avenue — it started with clay.

02

X / Twitter Signal

AI agents now burn nearly 5x the tokens that humans do, and that figure is up 14x since February. This isn't a usage trend — it's a structural shift. Agents are becoming the primary consumers of AI infrastructure, and the economics of every model provider need to be rethought around that reality.

Mollick flags the coming contradiction cleanly: people will tell pollsters they hate AI while quietly using it for everything. This is the tobacco-era gap between stated preference and revealed behavior — and it has massive implications for how AI policy, regulation, and public sentiment will diverge from actual adoption curves.

Ng published what he considers the most important skills for building and deploying AI applications. In a field drowning in framework tutorials, a foundational skills map from one of its architects is worth slowing down for — especially as the gap between "using AI" and "deploying AI" widens fast.

Gemini 3.7 Flash is Google's fastest-growing model launch ever, per Kilpatrick on the inside. After years of DeepMind being the critical darling that couldn't find a product, this signals a genuine inflection — and puts real competitive pressure on OpenAI's mid-tier offerings.

03

The Thread

The week's signal is this: AI is no longer primarily a human-facing tool. A16z's data showing agents consuming 5x more tokens than people — up 14x in six months — means the dominant user of AI infrastructure is now AI itself. Linus Torvalds debugging with an AI that kept insisting the problem was unsolvable, then being proven wrong, is a perfect microcosm: these systems are powerful assistants and unreliable oracles simultaneously. The skill isn't trusting them. It's knowing when not to.
"The AI several times stated flat out that this was impossible and unsolvable. I suspect those moments were its least useful contribution." — Linus Torvalds
Meanwhile, the subtlefakes story deserves more attention than it's getting. The arms race between AI-generated abuse content and detection systems has entered a new phase — one where slight perturbations render harmful images invisible to automated moderation. This is what happens when safety tooling chases yesterday's attack vector. The township blocking a nuclear weapons data center with an electrical moratorium is the same pattern at infrastructure scale: legacy governance frameworks scrambling to contain forces they weren't designed for. Gemini 3.7 Flash breaking growth records and Grok Bot opening enterprise access in the same week suggests the model wars are accelerating into deployment wars. The question is no longer which model is smartest — it's which ecosystem locks in enterprise workflows first. Stripe buying OpenRouter, Cloudflare building agent payment rails, Cursor challenging GitHub: the picks-and-shovels layer is consolidating fast, and whoever owns the agent plumbing owns the margin.