The Polylog AI Intelligence Brief

Morning Edition · Thursday, August 13, 2026

Markets

Nvidia Enlists Six Wall Street Firms to Mobilize More Than $500 Billion for AI Data Centers

The chipmaker is guaranteeing part of the resale value of its own accelerators, moving compute financing off its own balance sheet and into structured credit markets.

3 sources
Corroborated

Tech

Three Frontier Models Shipped in One Day, and Two of Them Undercut the Leaders on Price

DeepSeek V4 Pro 0813 posted 87.9 on Terminal-Bench 2.1 at $0.87 per million output tokens, Grok 4.6 took the top spot on a knowledge-work evaluation, and Alibaba moved its 2.4-trillion-parameter model to open weights.

2 sources
Plausible

Markets

Texas Freezes a 474-Gigawatt Grid Connection Queue That Is Almost Entirely Data Centers

Governor Greg Abbott ordered an audit before any new project advances, putting roughly a fifth of the United States data center pipeline at risk of delay.

1 source
Corroborated

Tech

OpenAI Says Codex Now Generates Most of Its Enterprise Output Tokens, and Non-Developers Are Driving the Growth

The company reports Codex produced 64 percent of combined Codex and ChatGPT enterprise output tokens in June, with legal teams adopting it faster than engineering.

3 sources
Plausible

Tech

Google Trains Clinical AI Through Simulated Residency Instead of Exam Questions

Reinforcement learning against multi-turn simulated patient encounters targets diagnosis, management, communication, documentation and safety together, a step past the single-answer benchmarks that made medical AI look solved.

2 sources
Corroborated

Tech

Researchers Argue Agent Safety Belongs at Runtime, Not in the Training Run

A new paper contends that reinforcement learning from human feedback, direct preference optimization and Constitutional AI are structurally insufficient for agents that execute code and mutate files, and a separate wave of tools is already moving control outside the model.

2 sources

Tech

DeepMind Ships a Sign Language Model Into a Phone, With the Vision Stage Running on Device

SL2T translates American Sign Language to English text in Gboard and Live Transcribe on Pixel 11, trained on more than 100,000 hours across over 50 sign languages and scoring 70 on BLEURT (a machine-translation quality score) on the FLEURS-ASL benchmark.

1 source

Tech

A Paper Retrofits Recurrent Depth Into an Already-Trained Language Model

Researchers split Qwen2.5-0.5B-Instruct into a prelude and a weight-tied recurrent block, and report the learned iterative latent transition survives outcome-only annealing.

1 source

Tech

Meta's Open Perception Models Move Into Assistive Robotics at the University of Pittsburgh

Researchers are building wheelchair-mounted manipulation around Segment Anything and DINO rather than training task-specific vision systems from scratch.

2 sources

Tech

OpenAI Loses Its Only Dedicated Ethicist and Does Not Plan to Replace Her

Chloé Bakalar left in July after less than a year, the third senior safety-adjacent departure the company has absorbed without an announcement.

1 source
Corroborated

Tech

Spotify Will Badge AI-Generated Artists and Keep Them Out of Recommendations by Default

From mid-September, profiles judged to represent synthetic identities carry an "AI Persona" label and are excluded from editorial and algorithmic playlists unless a user follows them.

1 source
Corroborated

Tech

An Autonomous Research Agent Takes On Conway's 99-Graph Problem and Reports Verifiable Bounds, Not a Solution

The work narrows the search space for a longstanding open question in combinatorics, and its value rests entirely on results a machine can check.

2 sources