Conferences, people and the culture of running systems. The human side of the pager.
Jakub Pachocki argues that no AI lab has adequately solved alignment to justify continuing development at maximum speed, and calls for industry-wide voluntary slowdowns until shared safety standards emerge.
The World Summit on the Information Society convened in Geneva this July to launch a new implementation phase, moving away from documenting commitments toward tracking concrete outcomes through roadmaps, metrics and accountability mechanisms.
OpenAI reports that GPT-6 Astra achieved 98.6% on the ARC-AGI-3 benchmark, a dramatic leap from the 7.8% posted by its predecessor GPT-5.6 Sol. However, undisclosed evaluation parameters and gaps in transparency raise questions about what the headline number actually demonstrates.
OpenAI unveiled its latest flagship model, GPT-6 Astra, claiming entry into the "AGI era," though the model's performance gains vary significantly across benchmarks and it carries a steep per-token cost.
Meta has formally launched its Muse Code coding agent and released Muse Spark 1.3, claiming substantial improvements in coding and agentic tasks that briefly overtook Google's latest Gemini model on cost-efficiency benchmarks.
Google has rolled out its third Flash model iteration in six weeks, claiming performance parity with Anthropic's Opus 5 while maintaining budget-friendly token rates, though the model requires additional computational effort to achieve comparable results.
A flip chart experiment at RIPE 92 explored how to bridge the gap between academic research and real-world network operations. The trial revealed interest in formal mechanisms to help researchers and operators collaborate more deliberately.
Following the success of its San Francisco location, which has drawn over 22,500 developers since July 2025, AWS is opening permanent community spaces in three new cities to host workshops, hackathons, and developer meetups.
