Import AI 464: Fables writes GPU kernels; AI automation; and analog computation

2025-10-13 · Source: Import AI · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Software Development & Engineering · Depth: Advanced, long

Summary

Recent advancements highlight AI's growing capabilities across diverse technical and economic domains. Fable, an AI system, achieved an 18.71X speedup in GPU kernel design on KernelBench-Mega, outperforming other models like Claude Opus 4.8 (14.4X) and GPT 5.5 (4.34X), signaling progress in AI R&D automation. Concurrently, AI systems' success rate on the Remote Labor Index for online freelance tasks surged from 2.5% in October 2025 to 16.1% in July 2026, with Fable 5 leading at 16.1%, indicating significant economic impact potential. Furthermore, OSWORLD 2.0, a new benchmark, evaluates AI on multi-hour, multi-program computer tasks, where Claude Opus 4.8 achieved 20.6% binary accuracy, demonstrating AI's increasing competence in complex computer use. Finally, JD's Oxygen AI Item Center, managing billions of SKUs on Huawei Ascend NPUs, exemplifies large-scale, self-updating AI-driven business operations.

Key takeaway

For Directors of AI/ML evaluating strategic investments, these advancements signal a critical shift towards AI-driven automation across R&D and operational tasks. You should prioritize integrating AI systems capable of complex computer interaction and self-improving functions, as demonstrated by OSWORLD 2.0 and Fable's kernel design. Focus on developing "person-light, AI-heavy" organizational structures, leveraging AI for large-scale back-office functions like inventory management, to maintain competitive advantage against rapidly expanding AI capabilities.

Key insights

AI systems are rapidly advancing in complex technical tasks, online labor automation, and sophisticated computer interaction.

Principles

• AI capability expansion outpaces human comparative advantage.
• Benchmarks signal AI's recursive self-improvement potential.
• Deep learning integrates with structured systems for scale.

Method

JD's Oxygen AIIC uses human-AI ontology engineering, "Semantic Search then Discrimination," self-evolving LLMs/VLMs with expert modules, and a "Unified item tunnel" for inventory management.

In practice

• Evaluate AI agents on multi-hour, multi-program tasks.
• Track AI success rates on online freelance projects.
• Implement human-AI collaboration for ontology evolution.

Topics

• GPU Kernel Optimization
• AI R&D Automation
• Digital Labor Automation
• OSWORLD 2.0
• Enterprise AI
• Supply Chain AI

Code references

• xlang-ai/OSWorld-V2

Best for: Machine Learning Engineer, Research Scientist, CTO, AI Scientist, Director of AI/ML, AI Engineer

Related on AIssential

• Import AI 463: Self-improving robots; a 10k Chinese GPU cluster; and an elegiac essay for the human era — Advanced AI development increasingly relies on autonomous systems, large-scale infrastructure, and machine-readable data, while raising profound societal questions.
• The Shape of the Thing — Exponential AI capability growth is shifting work from human-AI collaboration to AI agent management, enabling radical new organizational models.
• Official Keynote Closing Video | GTC 2026 — NVIDIA's GTC keynote unveiled a future of AI factories, autonomous agents, and physical AI, driven by massive compute and efficient inference.
• Microsoft’s AI chief on the greatest game of catchup ever played — Microsoft is aggressively building proprietary AI models, chips, and integrated software to achieve cost leadership and deep product integration, aiming to surpass competitors.
• Plato closes $14.5M round to scale AI tools for distributors — Plato's AI operating system automates core workflows for wholesale distributors, improving efficiency and commercial performance.
• Import AI 448: AI R&D; Bytedance's CUDA-writing agent; on-device satellite AI — AI capabilities are advancing rapidly, necessitating new metrics for governance and enabling diverse edge computing applications.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Import AI.