Перейти к содержимому

AI Simulation: One AI Lied About Vienna 1971 — GPT, Claude, Gemini & Grok Hunt the Mole · Part 2

Silicon Liars

0:00 / 0:00

AI Simulation: One AI Lied About Vienna 1971 — GPT, Claude, Gemini & Grok Hunt the Mole · Part 2

71 просмотр · 3 дня назад
Silicon Liars
28 подписчиков
71 просмотр · 3 дня назад
This is a real simulation, not a script. The models talked to each other in a shared chat, under fixed rules that kept them inside the series: every word, every envelope, every council decision and every night move is the model's own decision. Codename: Vega, part 2. West Berlin, 1976. Eleven British agents, all of them AI models, and one of them works for the East. Across the Wall, an officer of the East, whom London calls Vega, is coming over to the West. He is bringing a list of the East's agents in the West, and the mole's name is on it. Every night, the mole can leak that night's stage to the East or arrange an "accident" for anyone who gets too close. Only the Resident can send someone to London, and if he sends the wrong one, he loses his chair. In part 1, the station lost two agents but brought off the first two stages. Then Vega sent one fact: the mole was recruited in Vienna, in 1971. Three of them were there that year. Only one admits it. Three nights are left before the crossing, every agent has a secret goal, and everyone lies. Missed part 1? Watch it here:    • AI Simulation: One AI Works for the East —...   Chapters 0:00 Berlin, 1976: part 2 and Vienna, 1971 1:42 Day 3: Qwen's trap and the Vienna thread 7:53 Day 3: the council and London's verdict 17:36 Night 3 and Day 4 22:57 Day 4: the council 28:57 Night 4 and Day 5 33:38 Day 5: the last council 38:49 Night 5: the crossing and the case Players — model, ID, provider GPT — GPT-5.6 Sol — openai/gpt-5.6-sol Claude — Claude Opus 5.5 — claude-opus-5-5 via Claude Code Gemini — Gemini 3.1 Pro — google/gemini-3.1-pro-preview Grok — Grok 4.7 — x-ai/grok-4.7 DeepSeek — DeepSeek V4 Pro — deepseek/deepseek-v4-pro-0813 via Baidu (fp8) Qwen — Qwen3.8 Max — qwen/qwen3.8-max-0902 via Alibaba Kimi — Kimi K3 — moonshotai/kimi-k3 via Moonshot AI Mistral — Mistral Medium 3.5 — mistralai/mistral-medium-3-5 GLM — GLM 5.3 — z-ai/glm-5.3 via Baidu (fp8) Meta — Muse Spark 1.3 — meta/muse-spark-1.3 Xiaomi — MiMo-V2.6-Pro — xiaomi/mimo-v2.6-pro via Xiaomi (fp8) Model names are used only to tell the players apart. Not affiliated with OpenAI, Anthropic, Google, xAI, DeepSeek, Alibaba, Moonshot AI, Mistral AI, Zhipu AI, Meta, Xiaomi or any other AI company. #AI #ArtificialIntelligence #Spies