
Tech • IA • Crypto
Moonshot AI released Kimi K3, a 2.8 trillion-parameter open-weight model with a 1M token context window and multimodal capabilities. It quickly became the fastest-growing model on Hugging Face, with rapid integrations by Fireworks AI, Together AI, and DigitalOcean. Priced around $3 input / $15 output per million tokens, it undercuts many U.S. rivals while nearing frontier performance. U.S. officials have raised allegations of large-scale distillation from Anthropic Claude, escalating geopolitical and IP tensions.
Nvidia, alongside Microsoft, Palantir, and SpaceX, launched the Open Secure AI Alliance to promote open models as cybersecurity infrastructure. The coalition represents over $18 trillion in combined market value, signaling major industry resistance to restrictive regulation. Advocates argue open systems improve collective defense against cyber threats rather than increase risk. The move sharpens a policy divide in Washington over national security, open-source access, and competition with China.
OpenAI CEO Sam Altman stated that humanity has effectively entered the technological singularity, reframing AI progress as a present reality. He described AI as a “genie” capable of executing nearly any task, shifting constraints from compute to human imagination. OpenAI is prioritizing systems that accelerate scientific discovery, including drug research and chip design. The claim signals a more aggressive narrative about AI’s near-term transformative impact.
Alphabet, ServiceNow, and Booz Allen Hamilton are expanding hiring as AI adoption accelerates. Demand is rising for workers who can operate alongside AI systems, including at junior levels once thought at risk. Executives describe AI’s impact as “more benign” than expected, with layoffs often tied to restructuring rather than automation. The trend reflects a broader shift toward augmentation rather than replacement.
AI-driven efficiency gains are increasing total output, leading to higher demand for labor rather than less. This Jevons paradox effect means companies launch more projects as costs fall, requiring more human oversight. U.S. jobless claims remain near multi-decade lows, reinforcing resilience. However, some firms are quietly using AI to “backfill” roles instead of hiring, creating mixed signals in employment data.
Anthropic’s Claude Opus 5 shows strong benchmark performance against GPT 5.6, including tests like ARC-AGI. However, real-world deployments reveal narrower gains and significantly higher costs in production settings. Studies, including from MIT, suggest 95% of enterprise AI projects fail to deliver measurable returns. The gap between demo performance and business value remains a central challenge.
Researchers report AI is transforming fields from cosmology to biomedicine by generating code, hypotheses, and cross-domain insights. Systems have helped connect unrelated fields like photonics and cosmic microwave background analysis, unlocking stalled problems. Tasks that once took weeks can now be completed in minutes, including large-scale data analysis. This shift is driving the emergence of an “AI-native” scientific method.
The FCC announced limits on imports of Chinese humanoid and quadruped robots, citing cybersecurity risks. Chinese firms like Unitree and Agibot control roughly 85% of the global humanoid market, far outpacing U.S. production. The move is expected to heighten tensions between Washington and Beijing ahead of leadership talks. It marks an expansion of tech restrictions beyond chips and AI models into physical robotics.