The training pipeline is genius: Stage 1: CPT on 500B tokens of structured code tasks
Stage 2: Two-stage SFT with error masking
Stage 3: RL with chunk-level optimization Each stage builds on the last. No shortcuts. Just systematic capability building.
INNOVATION
-

Systematic AI Training Pipeline Evolution
By
–
-

IPA: Chunk-Level Credit Assignment Breakthrough
By
–
Here's the breakthrough nobody expected: They introduced IPA (Interaction-Perceptive Agentic Policy Optimization) that assigns credit at the CHUNK level, not token level. Tokens are too fine. Trajectories are too coarse. Chunks align with actual tool-use semantics.
-

Building the ecosystem before the model
By
–
They didn't just train a model. They built the ENTIRE ecosystem first. → ROLL: RL training framework
→ ROCK: Sandboxed execution engine
→ iFlow CLI: Agent orchestration system ROME wasn't built in a day. The infrastructure came first. The model came last. -

AI Agent Demos Exposed as Fraud
By
–
The dirty secret of AI agents: Every demo you've seen is basically fraud. That "autonomous AI employee" your favorite startup showed off? It's 3 ChatGPT calls wrapped in marketing. Meanwhile, this team built ROME using actual production-grade agent infrastructure.
-

AI Infrastructure Breakthrough in 2025
By
–
Chinese AI labs just dropped a bombshell research paper that exposes why 99% of "AI agent" companies are building on broken infrastructure. The ROME model + ALE ecosystem might be the most important open-source release of 2025. Here's what nobody's talking about:
-
New Year’s Resolution: Automate Your Daily Tasks with Abacus AI
By
–
Make a NY resolution to automate your routine work… Ask the Abacus AI Deep Agent to automate you day to day tasks for you Scheduled tasks can take care of ALL YOUR REPETITIVE CHORES
-
Welcome to 2026: AI Absolute Mastery Year Leave Human Work
By
–
He publicado un episodio en @ivoox
: "Bienvenidos al 2026: Tu Año de Dominio Absoluto con IA (deja de trabajar como un humano) #podcast -
2025 AI Research Highlights: IMO Gold and Gemini Contributions
By
–
2025 recap / highlights
– imo gold medal – contributions to Gemini 2.5/3 and deep think (via research and sometimes captaining)
– hosting @JeffDean
, @quocleix
, @benoitschilling
, @denny_zhou in singapore
– meeting and chatting with the legend @leehsienloong at the istana
– -
Wayve’s 2025: Global Expansion and Commercial Breakthroughs in Autonomous Driving
By
–
2025 was a defining year for Wayve.
— Alex Kendall (@alexgkendall) 31 décembre 2025
We founded Wayve on the belief that embodied intelligence would unlock autonomy at scale. After eight years of pioneering deep tech research, this year marked a true inflection where Wayve became global and commercial.
Wayve went global 🌍
*… pic.twitter.com/cbCi5M2uoh2025 was a defining year for Wayve. We founded Wayve on the belief that embodied intelligence would unlock autonomy at scale. After eight years of pioneering deep tech research, this year marked a true inflection where Wayve became global and commercial. Wayve went global 🌍 * A year ago we were just driving in central London. In 2025 we drove autonomously in 500+ cities across Europe, Asia, and North America — 43% with zero prior training data, showcasing the generalisation of AV2.0 * Expanded operations across London, Silicon Valley, Tokyo, Vancouver, Tel Aviv, and Stuttgart, welcoming exceptional talent and partnering with the world’s leading automotive manufacturers Wayve went commercial 🚗 * Started work with global fleets and automakers to deploy our AI Driver in applications across all levels L1–L5 autonomy * Announced a landmark deal with Nissan, one of the largest global automakers, to bring our AI Driver to their global consumer vehicles from FY2027 * Partnered with Uber to launch robotaxis on their marketplace in cities around the world, starting in London Grateful to our team, investors, and partners for making this possible. Bring on 2026!
→ View original post on X — @ryan_browne_, 2025-12-31 23:57 UTC
-
Kling v2.6 Released: Advanced Text-to-Video and Image-to-Video Generation
By
–
Kling v2.6 is here
— Replicate (@replicate) 31 décembre 2025
Generate cinematic text-to-video and image-to-video clips with fluid motion, photorealistic details, and native audio pic.twitter.com/tZRJgCBTn5Kling v2.6 is here Generate cinematic text-to-video and image-to-video clips with fluid motion, photorealistic details, and native audio