Good morning, welcome to this week’s Lore Brief, your 3-minute brief of the most important moves in AI and tech.
This issue is brought to you by Factory, the fastest way to ship software with autonomous engineering agents.
Jensen Huang’s first X post backs open models letter → NVIDIA’s CEO used his debut post to share a letter arguing open-weight models boost safety cybersecurity and innovation. The letter has drawn broad support from major tech firms and investors. Anthropic remains the main frontier lab that has not signed. Read more here →
Nvidia invests $5 billion in Ilya Sutskever’s Safe Superintelligence → The chipmaker is putting substantial equity into the stealth lab as part of a long-term partnership. SSI will gain access to Vera Rubin systems that expand its compute by an order of magnitude. Sutskever said the research is now ready to scale. Read more here →
Moonshot AI releases full Kimi K3 weights → The 2.8 trillion parameter MoE model comes with native visual understanding and a one-million-token context window. A new architecture delivers about 2.5 times more intelligence per unit of compute. The company is also open-sourcing attention kernels and agent infrastructure. Read more here →
Entire game demo created by Claude Opus 5:
An interesting idea:
Opus 5 for creating product videos:
Anthropic shares new context engineering rules for Claude 5 models → The team removed more than 80 percent of Claude Code’s system prompt with no drop in coding performance. Newer models work better with less rigid rules and more room for judgment. Progressive disclosure and cleaner tool design now replace long lists of examples and constraints. Read more here →
Michael Dell highlights local frontier-scale AI → Shopify’s Tobi Lütke is running the 753-billion-parameter GLM-5.2 model on a Dell workstation. It delivers 40 tokens per second with no cloud or API. Frontier intelligence is becoming available on personal hardware. Read more here →
Pipe Network makes Kimi K3 runnable on Mac Studio → The team used streaming conversion so the full model never loads at once. Expert pruning then keeps only the most useful of the 896 experts for a given workload. This shrinks the 1.6 TB model enough to run locally on Apple Silicon. Read more here →
Boring Company seeks funding at $20 billion valuation → Elon Musk’s tunneling firm is in talks to raise about $4 billion. The new valuation is more than triple its previous $5.7 billion mark from 2022. The company continues expanding tunnel networks beyond Las Vegas. Read more here →
Microsoft launches MAI-Cyber-1-Flash for security → The specialized model finds hard vulnerabilities in complex codebases. Combined with multi-agent systems it reaches top performance at half the previous cost. Project Perception turns this into continuous agentic monitoring and remediation. Read more here →
Claude finds real weaknesses in cryptographic schemes → Anthropic’s Mythos Preview discovered a new attack that halves the security of the post-quantum HAWK signature scheme. It also sped up attacks on reduced-round AES by hundreds of times. Most of the work happened autonomously at a cost of roughly $100,000 per result. Read more here →
