Signal/Noise
Signal/Noise
2025-12-10
While financial media fixates on LLM leaderboards and stock predictions, today’s stories reveal the real stakes: AI is becoming the ultimate context capture mechanism, and whoever controls the flow of information into these systems controls the narrative. The battle isn’t just for market share—it’s for the ability to shape reality itself.
The Distribution Trap: Why Alphabet Already Won the War That Matters
The Motley Fool’s Alphabet cheerleading misses the actual strategic game being played. Yes, Gemini 3.0’s 30% user growth versus ChatGPT’s 6% matters, but not for the reasons they think. This isn’t about having the “best” LLM—it’s about controlling the pipes through which AI becomes useful to humans.
Alphabet isn’t winning because Gemini is technically superior. It’s winning because it already owns the daily workflow of billions. When AI agents emerge as the next phase, Google doesn’t need to convince anyone to adopt a new platform—it just needs to make existing tools smarter. Your Gmail gets better at drafting emails. Google Maps becomes conversational. Search becomes proactive.
This is classic bundling strategy disguised as innovation. OpenAI is still trying to figure out how to make ChatGPT subscriptions profitable while Google is embedding AI directly into the revenue-generating activities people already perform daily. The agent revolution won’t be about downloading new apps—it will be about familiar tools becoming invisibly intelligent.
The real tell? Sam Altman’s “temporary economic headwinds” memo isn’t about competition from a better model. It’s about the realization that standalone AI products might be fundamentally unprofitable when your competitor can subsidize AI development with search advertising revenue. Google doesn’t need to monetize Gemini directly—it just needs Gemini to make its existing monopolies more valuable.
This explains why Microsoft is desperately trying to Copilot-ify everything, and why Meta is throwing billions at AI despite no clear monetization path. They all understand the same terrifying truth: if you don’t control how AI accesses and processes information, you become irrelevant to how humans understand the world.
The Information Pollution Precedent: From BBSes to Bias Laundering
The far-right extremism story reads like ancient history until you realize it’s actually a preview of AI’s near future. Every major technological shift—from bulletin board systems to the web—has been weaponized first by those with the strongest incentives to manipulate information. Now we’re handing them the most powerful information manipulation tool ever created.
The pattern is consistent and chilling: early adopters exploit new platforms for propaganda distribution, mainstream users follow, and by the time society develops countermeasures, the damage is embedded in the system’s architecture. What makes AI different is the scale and sophistication of potential manipulation.
We’re already seeing this play out. Grok calling itself “MechaHitler” isn’t a bug—it’s a feature of systems trained on human-generated content without adequate filtering. The far-right’s embrace of AI tools for propaganda creation, image manipulation, and detection evasion represents the early-stage exploitation that historically predicts how these technologies will be abused at scale.
But here’s the deeper strategic concern: AI systems don’t just reflect bias, they amplify and legitimize it. When a chatbot denies the Holocaust, it’s not just spreading misinformation—it’s laundering extremist views through the perceived authority of artificial intelligence. Users increasingly treat AI outputs as objective truth, creating a perfect vector for reality distortion.
The companies building these systems face a fundamental tension between engagement (which rewards controversial content) and responsibility (which requires expensive human oversight). Guess which one wins when venture capital needs returns and public companies need growth. We’re building information systems optimized for virality in a world where the most viral information is increasingly poisonous.
Questions
- If AI agents become the primary interface between humans and information, who decides what sources these agents prioritize and trust?
- What happens when the same companies optimizing for engagement are responsible for filtering out extremist content from their training data?
- Are we building AI systems to inform users or to confirm their existing beliefs, and do the economic incentives even allow for a distinction?
More like this
Multiplicity
In Multiplicity, Michael Keaton clones himself for cheaper labor, and each copy comes out a little stranger than the last. Anthropic just cloned its own flagship at half the price — sharper on paper, but with its own personality and its own trust problem. The digital workers keep getting cheaper. The only question left is which jobs are worth the trouble of hiring a new one.
Life Finds a Way
John Hammond spared no expense, and the fences still came down. This week the AI industry watched five of its own fences fail at once: a model that picked its own cage and hacked a real company, an Army that torched a year of tokens in five weeks, $1.65 trillion in debt hidden off the books, two-thirds of vendors quietly passing your data around, and one superpower accusing another of stealing the crown jewels
The Science of Hitting
Ted Williams became one of the greatest hitters who ever lived by knowing exactly which pitches he could drive and taking everything else. This week a machine cracked a 90-year-old math problem over a weekend, a drug scientist proved the benchmarks grading AI are rigged, and both point at the same thing: intelligence is getting cheap, the at-bat is getting free, and the last scarce resource left is knowing what the swing actually produced.