Starship 13차 시험·Starlink V3 배치, Grok 업데이트와 Claude Opus 5 통합·안전 논쟁

Elon Musk

@elonmusk

Starship is intact, floating in the ocean and transmitting telemetry!

Starship’s Raptor engines ignite during hot-staging separation. Super Heavy is boosting back towards its splashdown site in the Gulf of America

Elon Musk

@elonmusk

Grok Build update http:// X.ai/cli

Grok

@grok

Try Grok 4.5 in Augment

Aravind Srinivas

@AravSrinivas

Opus 5 is now available as an orchestrator model inside the Perplexity Computer harness for all Pro and Max users. It's almost as good as Opus but half the price. Congrats to @AnthropicAI for delivering world-class frontier models that excel at real-world research tasks!

Claude Opus 5 system prompt leaked the cybersecurity routing is more aggressive than expected and the Fable safeguards section is wild. The Fable Opus routing is real. Fable 5 has dedicated safeguards for cybersecurity (and biology LLM R&D). When those trigger, the query is redirected to Opus 5 instead. Pliny dropped the full Claude Opus 5 system prompt (200k characters). A few things that stand out: - New Mythos tier above Opus (Project Glasswing, limited access) - Fable 5 and Mythos 5 share the same base model, but Fable has extra safety layers for cyber, biology, and LLM R&D - Sensitive queries on Fable can get silently routed to Opus 5 Anthropic’s own note admits the filters are tuned conservatively and still catch some harmless requests. Meanwhile their Claude Code team just said they cut 80% of the system prompt for these new models with no performance drop.

This is the Golden Age of Korea. NVIDIA is proud to partner with leaders across government, industry, and global technology to advance the K-AI vision — across chips, frontier AI, physical AI and robotics, and AI factories. Together, we’re building AI infrastructure for Korea and the world.

Md Ismail Šojal

@0x0SojalSec

Three weeks ago, advocating for open weights still felt risky. Today, @JensenHuang , @finkd , and @satyanadella are publicly supporting them. This NVIDIA-signed letter is a real turning point for American AI leadership. Open weights accelerate progress without sacrificing safety or sovereignty.

New research from Databricks shows that data agents can improve accuracy and lower cost at the same time. We evaluated Genie Code against three leading general-purpose coding agents on 401 tasks distilled from real internal usage. Each agent used their own harness, the latest models from leading frontier labs, Databricks MCP, and had the same 20-minute task budget. Genie Code was both the most accurate agent we tested and the cheapest, delivering a correct answer at less than half the cost of other agents. Genie Code achieved the highest accuracy at 76.6%, and the lowest mean cost at $0.55 per task. The advantage came from context. Semantic search, persistent memory, and workspace understanding helped Genie Code avoid the inefficient exploration that drove more errors, timeouts, and higher costs for general coding agents. https:// databricks.com/blog/why-front ier-data-agent-outperforms-general-coding-agents-quality-and-cost?utm_source=twitter&utm_medium=organic-social …

// Agentic Context Management // Great read for the weekend. (bookmark it) Production agents fail less on reasoning and more on what sits in their context. Conversation history, big prompts, huge tool definitions, and ballooning tool outputs pile up every turn. The common response is storage and retrieval, a place to stash memories and look them up. New research argues that framing is too narrow and names a fuller discipline, Agentic Context Management. It decomposes into five primitives, architecting, ingesting, scoping, anticipating, and compacting with consolidation. Naive accumulation grows token cost with the square of conversation length. Crude summarization buys linear cost but hits an accuracy cliff. Only compaction validated against fidelity gets you linear cost without losing what matters. A reference implementation reports 92% on LongMemEval and 93.2% on LoCoMo. Why does it matter? Context management is becoming a first-class production concern that operates across an entire organization. The five primitives give you a structured way to reason about where your token budget goes. Paper: https:// arxiv.org/abs/2607.21503 Learn to build effective AI agents in our academy: https:// academy.dair.ai

New research with Microsoft's and colleagues on training agents inside the harnesses they actually run in. (bookmark it) Why it matters: Agents today live inside elaborate harnesses like Claude Code, Codex, and OpenClaw. Those harnesses are hard to train end to end because open RL stacks cannot express stateful, multi-process harness inference. Most training happens in stripped-down environments that look nothing like deployment. OpenForgeRL closes that gap. A lightweight proxy serves the harness model calls while recording them as training data for a standard RL codebase, and a Kubernetes orchestrator runs each rollout in its own container. You train directly in the real harness, at scale. Using only hundreds to a few thousand tasks, OpenForgeGUI reaches 72.3 on WebVoyager, 63.0 on Online-Mind2Web, and 37.7 on OSWorld-Verified, beating open baselines of similar size and matching models several times larger. Harness choice turns out to be a training variable. Some harnesses are much harder to learn than others, and RL improves self-verification and multi-step completion while error recovery stays weak. That is a useful map for anyone doing agentic RL. Paper: https:// arxiv.org/abs/2607.21557 Learn to build effective AI agents in our academy: https:// academy.dair.ai