Daily Brief

Moonshot AI released Kimi 3.1 Pro as open weights, a 104B active / 2.8T total parameter MoE model benchmarking at GPT-4.1 (Fable-5.1) class performance. The release includes a "Pro Highspeed" API variant delivering 2-3× faster inference than standard. Latent Space frames the frontier gap between Chinese open-weights and closed Western models as now approximately one month.

Apollo Global's chief economist assessed that despite massive capital deployment, AI has produced no measurable financial return for the vast majority of corporate America, with gains confined to tech sector companies. The statement lands as enterprise AI spending continues to accelerate and hyperscaler capex projections remain at record levels.

A new study forced Grok (xAI) and GPT (OpenAI) to make candidate recommendations across nearly 500 election races; the two models disagreed in more than 90% of cases, with Grok overwhelmingly favoring Republicans and GPT favoring Democrats. Neither company has acknowledged systematic political alignment in their models.

Governor Josh Shapiro signed the Artificial Intelligence Deceptive Advertising Act, requiring disclosures on any ad — image, video, audio, or text — "significantly modified or generated by artificial intelligence," with the label formatted to match the ad medium. The law applies to national brands and local businesses alike.

The Verge's detailed breakdown documents how Google's AI search rollout has severed the click-through pipeline that digital publishers spent 15 years building, with Google capturing the answer layer that previously drove traffic to third-party sites. The piece connects AI Overview expansion directly to revenue collapse at independent outlets.

Block released a voice model described as "eerily human" that can run fully locally on an Apple MacBook M4 Max, requiring no cloud inference. The model represents Block's push into on-device AI capabilities outside its core fintech business.