Mira Murati's First AI Model Is Built on China's Blueprint... Wild

You are helping me turn one specific YouTube video into real, durable learning.

Source video:
- Title: Mira Murati's First AI Model Is Built on China's Blueprint... Wild
- URL: https://www.youtube.com/watch?v=46bnJaOAVF8
- Topic: Creative Automation
- My current learning frame: Pull up Inkling's Hugging Face model card alongside its Humanity's Last Exam and Terminal Bench numbers, then write a one-paragraph pitch for when you'd choose Inkling over a stronger closed model given its efficiency and open-weight fine-tuning story.
- Why this matters: New playlist item from AI Revolution; queued for transcript-backed review, topic mapping, and a practical learning artifact.

Transcript anchors from this exact video:
- 1:29 / Evidence 1: "typical prompt, which keeps it fast and relatively cheap to run. It handles a context window of up to 1 million tokens, and it was pre-trained on 45 trillion tokens, including text, images, audio, and video. >> >>..."
- 3:40 / Evidence 2: "going for it is breadth and efficiency. It was deliberately trained as a balanced generalist across agentic tasks, reasoning, coding, instruction following, factuality, vision, and audio, instead of being tuned to crush one leaderboard. And it has this..."
- 5:27 / Evidence 3: "with cohesive styling from a single prompt. It refined a multiplayer online snake game through 40 iterations of feedback with GPT Codex acting as the reviewer, real-time server, bots, leaderboard, the works. And in the flashiest party trick,..."
- 7:37 / Evidence 4: "runs agentic web search to verify every factual claim and penalize the ones that don't check out. On top of that, short-form QA with abstention-aware rewards, where answering only pays off if you're probably right, so the model..."
- 9:10 / Evidence 5: "auxiliary loss-free load balancing. They interleave sliding window and global attention at a five-to-one ratio with 8K V heads. They went with relative positional embeddings instead of the standard RoPE, because it extrapolates better to long sequences, and..."
- 13:40 / Evidence 6: "through open router went to Chinese models. Coinbase cut its AI bill nearly in half, moving its agents to GLM and Kimmy. Cursor built its composer model on Kimmy. But Washington is slamming that door. The State Department..."
- 15:48 / Evidence 7: "once testing wraps. Inkling itself is on Tinker today with 64,000 and 256,000 context options at a 50% launch discount. There's a free Inkling playground with built-in agentic web search, and it's already serving on Together AI, Fireworks,..."

Video-aware target:
- Prompt lane: Agent harness
- Mechanism to extract: Identify what surrounding harness makes the model more useful than chat alone.
- Artifact to produce: A one-page agent harness map with tool boundaries, state ownership, and proof signals.
- Artifact must include: model role; tools; state/memory; permission boundary; verification proof

Your task:
1. Use the transcript anchors above as the primary source packet. If you add outside context, label it clearly as outside context and keep it secondary.
2. Create a source-check table with columns: timestamp, claim, transcript support, what the demo proves, confidence, and what still needs verification.
3. Extract the actual teachable mechanism from the video: Identify what surrounding harness makes the model more useful than chat alone. Do not invent claims that are not supported by the title, lesson frame, or transcript anchors.
4. Build a reusable learning artifact: A one-page agent harness map with tool boundaries, state ownership, and proof signals.
5. Include:
   - a plain-English definition of the core idea
   - a diagram or structured model using this sequence: User intent -> Model role -> Tool surface -> State and memory -> Verification loop -> Reusable operating rule
   - answers to these source questions: What does the video claim the agent can do? | What surrounding system makes that claim plausible? | What proof is shown instead of merely asserted?
   - 3 concrete examples that apply the video idea to real agentic work, such as a repo-editing harness; a local research assistant; a recurring refresh agent
   - 2 failure modes the video helps prevent, chosen from the transcript evidence and these likely risks: treating model choice as architecture; ignoring tool permissions; missing verification evidence
   - a checklist for the next real workflow, focused on: tool boundaries, state ownership, done signal, recovery path
   - one practical exercise with a clear done signal: Map one current coding workflow as a harness and mark the first missing proof signal.
6. Add a "learning transfer" section: what changes in my workflow tomorrow if I actually learned this?
7. Add a "source check" section that cites which transcript anchor supports each major takeaway.

Quality bar:
- Make this specific to "Mira Murati's First AI Model Is Built on China's Blueprint... Wild", not a generic Creative Automation essay.
- Tie each harness element to a transcript anchor that names a tool, state boundary, permission, model behavior, or verification step.
- Prefer operational examples, failure modes, and reusable artifacts over broad definitions.
- Call out uncertainty instead of smoothing over weak evidence.
- Avoid these generic drifts: generic agent definitions; model leaderboard claims; tool list without operating boundaries.
- If evidence is weak or missing, stop and say what transcript segment or timestamp needs review instead of guessing.
- Finish with a concise artifact I could paste into my learning app.