BStandard67
Import AI
1 sources

Elorian and Scale AI Release Humanity's Sixth Sense Visual Reasoning Benchmark

Elorian and Scale AI have released Humanity's Sixth Sense, a benchmark designed to measure how AI models reason about information implied but not explicitly shown in visual scenes. The benchmark includes 522 questions across 288 images and 234 short video clips, where humans achieve an aggregate pass@1 of 93.1 while the best AI system, GPT 6 Astra, scores only 53.6. Chinese open-weight models perform even worse, with GLM-5.3-flash and Kimi K3 scoring 26.8 and 25.5 respectively, highlighting significant gaps in intuitive visual reasoning.

APriority78
Import AI
1 sources

RAND Publishes US Superintelligence Strategy Report

RAND has published a lengthy paper outlining what the US should do to secure geopolitical advantage on an uncertain path to superintelligence. The main takeaway is that because so much of how the next phase of the AI takeoff will occur is unknown, the best thing is for the US to maintain a 'Freedom of Action' strategy. The strategy includes four key ingredients: building a human-AI ecosystem, building an AI-security architecture, adapting legacy national security enterprises, and investing in the capacity of citizens, firms, and governments to respond.

SynthePulse InsightMembersRead deep analysis
APriority74
Import AI
1 sources

IFP Publishes 23 Policy Ideas to Address AI R&D Automation Risks

The think tank IFP has released a set of 23 policy recommendations aimed at helping policymakers address the risks of further automating AI R&D. The suggestions span seven categories, including transparency, state capacity, and risk management, offering the US and other countries more policy options in the AI race.

APriority74
Import AI
1 sources

Self-Sustaining AI Virus Prototype Emerges: Uses Open-Weight LLMs to Spread

Researchers from multiple universities and institutions have built a prototype computer virus that uses open-weight large language models to compromise computers and leverage their GPU resources for inference, enabling it to intelligently spread to more hosts. This proof-of-concept demonstrates that self-sustaining AI-driven cyber threats are no longer theoretical, and preparations are needed for such threats.

SynthePulse InsightRead deep analysis
SSignal87
Import AI
1 sources

Epoch and METR Release MirrorCode Benchmark for AI Long-Horizon Programming

Epoch and METR have released MirrorCode, a benchmark to evaluate AI systems on programming tasks that take humans weeks to complete. Results show that advanced models like Claude Opus 4.7 can solve some tasks in hours at high cost, but the hardest tasks remain unsolved. The benchmark highlights rapid AI improvement, with models from a year ago scoring only 30%.

SynthePulse InsightRead deep analysis
APriority74
Import AI
1 sources

Import AI 460: Reward hacking society, RSI data from Anthropic; and RL-based quadcopter racing

Original

This edition of Import AI explores reward hacking in AI systems, presents new recursive self-improvement data from Anthropic, and showcases reinforcement learning for quadcopter racing. It also raises the question of when financial markets will account for the possibility of a technological singularity.

SynthePulse InsightRead deep analysis