PLUS: An AI cracks a 30-year problem, finds 100+ hidden planets, and the Oscars' new AI ban
Happy reading
A relatively unknown open-weights model from China has just outperformed giants like GPT-5.5 and Claude in a live programming contest. The victory from Moonshot AI's Kimi K2.6 is sending ripples through the AI world, suggesting the competitive landscape is rapidly changing.
The win shows the performance gap between proprietary and open-weights models is shrinking faster than many expected. With powerful alternatives becoming more accessible, are we entering a new era where the dominance of today's top AI labs is no longer a given?
In today’s Next in AI:
China’s Kimi K2.6 model beats GPT-5.5
An AI cracks a 30-year coding problem
AI discovers over 100 hidden planets
The Oscars' new ban on AI in film
China's Coding Upset
Next in AI: An open-weights model from Chinese startup Moonshot AI just defeated giants like GPT-5.5 and Claude in a real-time programming challenge, signaling a major shift in the AI landscape.
Explained:
Kimi K2.6 won the "Word Gem Puzzle" competition by using an aggressive sliding strategy to constantly create new words, a method that proved highly effective on the largest game boards.
In contrast, top models from OpenAI and Anthropic hardly moved any tiles, instead scanning for existing words—a tactic that failed when the puzzle became too scrambled.
While GPT-5.5 still holds a slight edge on general benchmarks, this win shows the capability gap is closing fast, with powerful open-weights models becoming serious contenders.
Why It Matters: An open-weights model outperforming top proprietary systems in a complex, real-time task shows the AI lead held by Western labs is no longer guaranteed. This new era of competition could accelerate innovation and provide developers with powerful, freely available alternatives to the current market leaders.
AI Cracks 30-Year Problem
Next in AI: After three decades of struggling with a complex web development issue, a developer used an AI assistant to finally understand and implement a complete website caching strategy in a single afternoon.
Explained:
Complex technical topics often have a wide gap between theory and practice, but AI assistants now act as personalized tutors that can bridge that gap by tailoring explanations to a user's specific setup.
The urgency to solve the problem came from a major shift in web traffic, where a growing share now comes from AI and search crawlers that depend on efficient caching for access.
The AI compressed what would have been weeks of trial-and-error into a rapid back-and-forth session, helping the developer build a coherent strategy and even a public dashboard to track its success.
Why It Matters: AI assistants are evolving into powerful partners that help professionals untangle complex, domain-specific challenges that were previously too opaque to solve. This trend augments human expertise by compressing decades of experience into highly productive, focused work sessions.
AI's Cosmic Discovery
Next in AI: A new AI system named RAVEN is sifting through NASA TESS mission data, discovering over 100 hidden planets and identifying thousands of other promising candidates.
Explained:
RAVEN's training method is key to its success; it learned to distinguish real planets from false signals by studying hundreds of thousands of simulated celestial events.
Beyond just a numbers game, the AI found several rare worlds, including planets that orbit their star in less than 24 hours and others lurking in the mysterious "Neptunian desert."
This high-quality dataset provides the most precise measurements yet of planet occurrence, confirming that about 9-10% of Sun-like stars host a planet in a close orbit.
Why It Matters: RAVEN demonstrates how AI can automate and accelerate cosmic discovery, turning overwhelming astronomical data into reliable findings. This approach provides a clearer map of our galaxy and helps scientists zero in on the most promising worlds for future study.
The Oscar Goes To... a Human
Next in AI: The Academy of Motion Picture Arts and Sciences is drawing a line in the sand on AI. The organization announced new eligibility rules that officially bar AI-generated screenplays and performances from winning its prestigious writing and acting awards.
Explained:
Under the new guidelines for the 99th Oscars, screenplays must be “human-authored” and acting roles must be “demonstrably performed by humans with their consent” to be eligible for consideration.
This move directly addresses recent industry controversies, including plans for AI-recreated performances of deceased actors and the emergence of fully synthetic digital actors.
While the rules don't yet address AI's use in categories like visual effects or music, the decision gives a strong foundation for other awards to build upon as they navigate the rise of generative tools.
Why It Matters: This decision safeguards Hollywood's most coveted awards for human creativity, establishing a clear boundary as AI becomes more integrated into filmmaking. It also forces a crucial industry-wide conversation about the definition of authorship and artistry in the age of AI.
AI Pulse
Harvard found that OpenAI's o1 model outperformed human doctors in high-pressure emergency triage, correctly diagnosing 67% of cases with limited information versus 50-55% for physicians.
OpenAI faces a high-stakes civil trial initiated by Elon Musk, who is attempting to oust CEO Sam Altman and challenge the company's for-profit direction ahead of its planned IPO.
Colleague.skill sparked a new trend in China where users create AI replicas of their ex-partners by uploading personal data, raising privacy and ethical concerns.