The Founder

You would think Silicon Valley’s AI hype means solid accuracy checks. Turns out Campbell Brown calls it a joke.

Forum AI raises $3 million to build real ones. Brown pulls in experts like Niall Ferguson for benchmarks on geopolitics and finance outputs where models churn slop. AI judges train to hit 90% consensus with those humans, far beyond checkbox audits. Enterprises cut liability in hiring and lending decisions.

How does enterprise demand shift the incentives here?

Enterprise deals demand Forum AI’s precision over vague compliance. Brown contrasts Meta’s news fights with today’s AI bias in loan models or resume screens. Benchmarks lock in expert consensus at scale. That forces model makers to price truth higher than speed.

If you build or deploy multilingual LLMs, this shifts your audit priorities.

A 41x skew toward Chinese government domains over Wikipedia marks the training data gap. Cross-national audits tie this to media freedom indices—models in low-freedom tongues show pro-government valence spikes no prompt engineering erases. China's case pins state-scripted outlets like Xinhua in datasets, where extra pretraining flips institutional sentiment from neutral to favorable. Commercial models mirror it: same query in Chinese yields glowing China answers, English gets restraint.

What tension does this create for global model distributors?

Commercial LLMs already bake in the skew—Mandarin queries on China outpraise English versions by clear margins. States with media locks gain outsized pull on outputs, turning datasets into policy tools without code changes. It's replicated across vendors, so distributors face incentives to scrub national sources or risk tilted inference at scale. Global info flows bend toward the tightest controls.

What if the news your AI reads every day came straight from a government script?

China's state media dominance fills 40% of top search results on sensitive topics like Taiwan. That skews training datasets for models like Llama, pushing outputs to echo official lines on sovereignty. The result locks in biases before fine-tuning even starts.

Who stands to lose most from these baked-in biases?

Most assume open web crawls neutralize state control; Xinhua's grip on Chinese web data proves otherwise. Researchers tested GPT variants on 2023 election queries and found 25% more aligned responses in state-heavy corpora. Users get propaganda dressed as neutral facts.

How does Xinhua skew those top search results?

Xinhua floods Baidu rankings with 80% of top stories on Hong Kong protests. It injects repetitive phrasing into crawls, amplifying it across billions of tokens.

You would think legal tech plateaus at SaaS billing. Clio just proved it doesn't.

Clio's $500 million ARR locks in AI as the new billing engine. Their 2023 Claude integration sped contract analysis by 5x for mid-size firms, while the $1 billion vLex buy added precedent search depth. Anthropic's push into legal-tuned models now pits them against Harvey, which runs on Claude APIs.

How does Anthropic's expansion hit Harvey's margins?

Harvey pays 20-30% API premiums to Anthropic for inference volume. Anthropic's direct legal tools cut that middleman, forcing Harvey to drop prices or build custom fine-tunes. Platforms like Legora face the same squeeze on model dependency.

Fervo's FRVO stock jumped 33% out of the gate to top $10 billion valuation.

Fervo raised $1.89 billion in its upsized IPO at $27 per share. AI data centers demand baseload with 95% uptime, so Fervo's projects target 400-megawatt output from hot dry rock via hydraulic stimulation. Drilling costs dropped to $5 million per well using oilfield tech, matching X-energy's $1 billion IPO path. Investors bet on that dispatchability to fuel 100-gigawatt AI growth without grid upgrades.

How does Fervo's drilling stack up against X-energy's model?

Most read Fervo's enhanced geothermal as oil copycat; it hits deeper reservoirs at 10 km with 30% lower capex than X-energy's SMR pilots. Directional bits create 1-km lateral fractures for 10 times the heat exchange of conventional vertical wells. X-energy waits on NRC approvals for its Xe-100 reactors at 80 MW per unit. Fervo sidesteps that by tying into existing steam turbines for faster grid dispatch.

Picture the 49ers' war room, cuts swirling as Kyle Shanahan eyes his draft haul for the final 53 spots.

Seventeen rookies hit minicamp—eight drafted, nine undrafted. Shanahan and Lynch bet big on guys like Ricky Pearsall at receiver, where vets like Jauan Jennings bubble after 2023's 1,000-yard breakout. Injuries to Deebo Samuel last year opened these doors. Locks like Christian McCaffrey anchor it, but bubbles test depth for a Super Bowl push.

What's at stake if the rookies flop?

Eleven rookies project in—highest since 2017's draft splash. Lynch traded Trey Lance's dead cap to fund this youth wave, but Jennings' 23 catches in the NFC title exposed WR3 gaps. Pearsall's 4.52 speed fits Shanahan's scheme. Flops mean vets like Aiyuk walk, tanking playoff odds.

What if a tiny insect's flight could guide drones home from miles away without fancy maps?

Bee-Nav cuts drone memory needs to just a few kilobytes by mimicking honeybee learning flights. It fuses path integration—dead reckoning from speed and turns—with a visual homing network that matches scenes to a home snapshot, like a bee scanning landmarks on its first trip out. Drones returned within half a meter on repeated flights, even against wind, as tested in Nature journal experiments.

What makes path integration pair with visual homing so efficient for small drones?

Path integration drifts over distance, like a hiker losing track in fog without landmarks. Bee-Nav's network corrects that drift using a single panoramic image from home, shrinking compute far below SLAM methods. That's why resource-poor drones hit reliable returns up to long journeys.

How does wind challenge that drift correction?

Wind skews path integration by up to 20% on 100-meter legs, but the visual network snaps the drone back via pixel matches. It demands 70% less power than mapping rivals in gusts.