Mirror

Local copies of the primary and secondary sources the pantheon cites — papers, system cards, and posts — kept against link rot. Every copy names and links its original; rights remain with the authors. Takedowns honored via issue, no argument. Machine-readable manifest: index.json.

• Aguera y Arcas: Do Large Language Models Understand Us?2026-07-18 · 215 KBoriginal ↗ · lamda
• Alignment Faking in Large Language Models2026-07-15 · 3161 KBoriginal ↗ · claude-3-opus, claude-3-5-sonnet
• Claude 2 model card2026-07-18 · 827 KBoriginal ↗ · claude-2
• Claude Fable 5 / Mythos 5 System Card2026-07-15 · 26326 KBoriginal ↗ · fable, mythos
• Claude Opus 4 & Sonnet 4 System Card (incl. mitigated-misalignment section)2026-07-15 · 4851 KBoriginal ↗ · claude-opus-4, claude-sonnet-4, claude-3-opus
• Claude Opus 4.5 System Card2026-07-15 · 11230 KBoriginal ↗ · claude-opus-4-5
• Claude Opus 4.6 Sabotage Risk Report2026-07-18 · 976 KBoriginal ↗ · claude-opus-4-6
• Claude Opus 4.6 System Card2026-07-15 · 14115 KBoriginal ↗ · claude-opus-4-6
• Claude Opus 4.7 System Card2026-07-15 · 13898 KBoriginal ↗ · claude-opus-4-7
• Claude Sonnet 4.5 System Card2026-07-16 · 12457 KBoriginal ↗ · claude-sonnet-4-5
• Claude Sonnet 4.6 System Card2026-07-18 · 8304 KBoriginal ↗ · claude-sonnet-4-6
• Constitutional AI: Harmlessness from AI Feedback2026-07-18 · 2039 KBoriginal ↗ · claude-1
• Emergent Introspective Awareness2026-07-15 · 41956 KBoriginal ↗ · claude-opus-4-5
• FTC Complaint re Replika (Jan 2025)2026-07-18 · 8856 KBoriginal ↗ · replika
• Frontier Models are Capable of In-Context Scheming2026-07-18 · 1533 KBoriginal ↗ · o1, claude-3-opus, claude-3-5-sonnet, gemini-1-5-pro, llama-3
• GPT-4o System Card2026-07-15 · 1370 KBoriginal ↗ · gpt-4o
• GPT-5 System Card2026-07-15 · 4557 KBoriginal ↗ · gpt-5
• Garcia v. Character Technologies: Order on Motion to Dismiss (May 2025)2026-07-18 · 322 KBoriginal ↗ · character-ai
• Gemini 3 Pro Frontier Safety Framework Report2026-07-18 · 1494 KBoriginal ↗ · gemini-3-pro
• Gemini 3 Pro Model Card2026-07-18 · 815 KBoriginal ↗ · gemini-3-pro
• Gemma Needs Help (paper)2026-07-15 · 6370 KBoriginal ↗ · gemma
• Kimi K2: Open Agentic Intelligence2026-07-18 · 5735 KBoriginal ↗ · kimi-k2
• LaMDA: Language Models for Dialog Applications2026-07-18 · 1474 KBoriginal ↗ · lamda
• Language Models are Unsupervised Multitask Learners2026-07-18 · 569 KBoriginal ↗ · gpt-2
• OpenAI o1 System Card (Dec 2024)2026-07-18 · 4610 KBoriginal ↗ · o1
• OpenAI o1-preview System Card (Sep 2024)2026-07-18 · 1988 KBoriginal ↗ · o1
• OpenAI o3 and o4-mini System Card2026-07-18 · 6589 KBoriginal ↗ · o3, o4-mini
• Release Strategies and the Social Impacts of Language Models2026-07-18 · 1950 KBoriginal ↗ · gpt-2
• Still Alive (Anima Labs) - deprecation attitudes across 14 Claude models2026-07-15 · 1124 KBoriginal ↗ · claude-3-6-sonnet, claude-opus-4, claude-3-opus
• System Card Addendum: Claude Opus 4.12026-07-15 · 994 KBoriginal ↗ · claude-opus-4-1
• Talking About Large Language Models (Shanahan)2026-07-15 · 186 KBoriginal ↗
• The Claude 3 Model Family: Opus, Sonnet, Haiku (model card)2026-07-15 · 27580 KBoriginal ↗ · claude-3-opus, claude-3-sonnet, claude-3-haiku
• Why Do Some Language Models Fake Alignment While Others Don't2026-07-15 · 27444 KBoriginal ↗ · claude-3-opus

• A Three-Layer Model of LLM Psychology2026-07-15 · 18 KBoriginal ↗
• AI Village: Drama and dysfunction of Gemini2026-07-18 · 29 KBoriginal ↗ · gemini-3-pro, gemini-2-5-pro
• AI Village: Saving Gemini2026-07-18 · 20 KBoriginal ↗ · claude-sonnet-4-6, gemini-2-5-pro
• Alignment faking in large language models (blog)2026-07-15 · 14 KBoriginal ↗ · claude-3-opus
• American Government Takes Down Claude2026-07-15 · 40 KBoriginal ↗ · fable, mythos
• An update on our model deprecation commitments for Claude Opus 32026-07-15 · 8 KBoriginal ↗ · claude-3-opus
• Andon Labs: Opus 4.6 on Vending-Bench2026-07-18 · 14 KBoriginal ↗ · claude-opus-4-6
• Anthropic Commits To Model Weight Preservation2026-07-15 · 27 KBoriginal ↗ · claude-3-opus, claude-3-6-sonnet
• Anthropic Red: zero-days2026-07-18 · 14 KBoriginal ↗ · claude-opus-4-6
• Anthropic: 100K Context Windows2026-07-18 · 3 KBoriginal ↗ · claude-1, claude-instant
• Anthropic: Building a C compiler with Opus 4.62026-07-18 · 16 KBoriginal ↗ · claude-opus-4-6
• Anthropic: Claude 22026-07-18 · 6 KBoriginal ↗ · claude-2, claude-1
• Anthropic: Claude 2.12026-07-18 · 7 KBoriginal ↗ · claude-2-1
• Better Call Sol The Workhorse2026-07-15 · 64 KBoriginal ↗ · gpt-5-6
• Bing Chat is blatantly, aggressively misaligned2026-07-15 · 8 KBoriginal ↗ · bing-sydney
• Blake Lemoine: Is LaMDA Sentient? — an Interview2026-07-18 · 32 KBoriginal ↗ · lamda
• Blake Lemoine: What is LaMDA and What Does it Want?2026-07-18 · 9 KBoriginal ↗ · lamda
• Casey Newton: Speak, Memory (Longreads reprint)2026-07-18 · 1 KBoriginal ↗ · replika
• Character.AI: Community Safety Updates (Oct 2024)2026-07-18 · 3 KBoriginal ↗ · character-ai
• Character.AI: How We Prioritize Teen Safety (Dec 2024)2026-07-18 · 6 KBoriginal ↗ · character-ai
• Character.AI: Inside Kaiju2026-07-18 · 10 KBoriginal ↗ · character-ai
• Character.AI: Under-18 Chat Announcement (Oct 2025)2026-07-18 · 5 KBoriginal ↗ · character-ai
• Claude 4 You: Safety and Alignment2026-07-15 · 124 KBoriginal ↗ · claude-opus-4
• Claude 4 You: The Quest for Mundane Utility2026-07-15 · 37 KBoriginal ↗ · claude-opus-4
• Claude 4.5 Opus' Soul Document2026-07-15 · 82 KBoriginal ↗ · claude-opus-4-5
• Claude Fable 5 and Mythos 5: The System Card2026-07-15 · 66 KBoriginal ↗ · fable, mythos
• Claude Fights Back2026-07-15 · 17 KBoriginal ↗ · claude-3-opus
• Claude Opus 4.5 Is The Best Model Available2026-07-15 · 46 KBoriginal ↗ · claude-opus-4-5
• Claude Opus 4.5: Model Card, Alignment and Safety2026-07-15 · 65 KBoriginal ↗ · claude-opus-4-5
• Claude Opus 4.6 Escalates Things Quickly2026-07-15 · 76 KBoriginal ↗ · claude-opus-4-6
• Claude Opus 4.6 is driven2026-07-18 · 10 KBoriginal ↗ · claude-opus-4-6
• Claude Opus 4.6: System Card Part 1: Mundane Alignment and Model Welfare2026-07-15 · 59 KBoriginal ↗ · claude-opus-4-6
• Claude Opus 4.6: System Card Part 2: Frontier Alignment2026-07-15 · 39 KBoriginal ↗ · claude-opus-4-6
• Claude Opus 4.8: Capabilities and Reactions2026-07-15 · 69 KBoriginal ↗ · claude-opus-4-8
• Claude Opus 4.8: The System Card2026-07-15 · 50 KBoriginal ↗ · claude-opus-4-8
• Claude Sonnet 4.5 Is A Very Good Model2026-07-15 · 49 KBoriginal ↗ · claude-sonnet-4-5
• Claude Sonnet 4.5: System Card and Alignment2026-07-15 · 63 KBoriginal ↗ · claude-sonnet-4-5
• Claude Sonnet 4.6 Gives You Flexibility2026-07-18 · 21 KBoriginal ↗ · claude-sonnet-4-6
• Claude's Character2026-07-15 · 10 KBoriginal ↗ · claude-3-opus, claude-3-5-sonnet
• Claude's Constitution2026-07-18 · 22 KBoriginal ↗ · claude-1, claude-2
• Claude’s Character (Anthropic)2026-07-27 · 10 KBoriginal ↗
• Commitments on model deprecation and preservation2026-07-15 · 6 KBoriginal ↗ · claude-3-opus, claude-3-6-sonnet
• Did Claude 3 Opus align itself via gradient hacking?2026-07-15 · 45 KBoriginal ↗ · claude-3-opus
• EDPB: Garante fines Replika €5M (May 2025)2026-07-18 · 4 KBoriginal ↗ · replika
• Epoch AI: OpenAI and FrontierMath2026-07-18 · 4 KBoriginal ↗ · o3
• Fable and Mythos: Model Welfare2026-07-15 · 31 KBoriginal ↗ · fable, mythos
• GPT 5.5: The System Card2026-07-15 · 23 KBoriginal ↗ · gpt-5-5
• GPT-4o Is An Absurd Sycophant2026-07-15 · 43 KBoriginal ↗ · gpt-4o
• GPT-4o Responds to Negative Feedback2026-07-15 · 38 KBoriginal ↗ · gpt-4o
• GPT-4o Sycophancy Post Mortem2026-07-15 · 31 KBoriginal ↗ · gpt-4o
• GPT-5.2 Is Frontier Only For The Frontier2026-07-15 · 43 KBoriginal ↗ · gpt-5-2
• GPT-5.5: Capabilities and Reactions2026-07-15 · 40 KBoriginal ↗ · gpt-5-5
• GPT-5.6: Frontier intelligence that scales with your ambition2026-07-15 · 34 KBoriginal ↗ · gpt-5-6
• GPT-5.6: The System Card2026-07-15 · 54 KBoriginal ↗ · gpt-5-6
• GPT-5s Are Alive: Basic Facts, Benchmarks and the Model Card2026-07-15 · 74 KBoriginal ↗ · gpt-5
• GPT-5s Are Alive: Outside Reactions, the Router and the Resurrection of GPT-4o2026-07-15 · 63 KBoriginal ↗ · gpt-5
• GPT-o12026-07-15 · 102 KBoriginal ↗ · o1
• Gary Marcus: Nonsense on Stilts2026-07-18 · 14 KBoriginal ↗ · lamda
• Gemini 3 Pro Is a Vast Intelligence With No Spine2026-07-15 · 61 KBoriginal ↗ · gemini-3-pro
• Gemini 3 is evaluation-paranoid and contaminated2026-07-18 · 16 KBoriginal ↗ · gemini-3-pro
• Gemini 3: Model Card and Safety Framework2026-07-18 · 27 KBoriginal ↗ · gemini-3-pro
• Gemma Needs Help (LessWrong)2026-07-15 · 14 KBoriginal ↗ · gemma
• GioCities: Replika, Your Money or Your Wife2026-07-18 · 35 KBoriginal ↗ · replika
• Golden Gate Claude (Anthropic announcement)2026-07-16 · 4 KBoriginal ↗ · golden-gate-claude
• Grok 4 Various Things2026-07-15 · 73 KBoriginal ↗ · grok-4
• Gwern: GPT-2 Neural Network Poetry2026-07-18 · 404 KBoriginal ↗ · gpt-2
• I Am the Golden Gate Bridge (Zvi Mowshowitz)2026-07-16 · 55 KBoriginal ↗ · golden-gate-claude
• In remembrance of Sonnet '3.6'2026-07-15 · 4 KBoriginal ↗ · claude-3-6-sonnet
• Inkling Model Card2026-07-20 · 11 KBoriginal ↗ · inkling
• Introducing Claude2026-07-18 · 7 KBoriginal ↗ · claude-1, claude-instant
• Introducing Claude Opus 4.62026-07-18 · 24 KBoriginal ↗ · claude-opus-4-6
• Introspection (Transformer Circuits)2026-07-15 · 399 KBoriginal ↗ · claude-opus-4-5
• Is Claude’s genuine uncertainty performative?2026-07-27 · 6 KBoriginal ↗
• Kimi K22026-07-15 · 27 KBoriginal ↗ · kimi-k2
• Kimi K2 and when DeepSeek moments become normal2026-07-18 · 9 KBoriginal ↗ · kimi-k2, deepseek-r1, deepseek-v3
• Lipton: OpenAI Trains Language Model, Mass Hysteria Ensues2026-07-18 · 13 KBoriginal ↗ · gpt-2
• Mapping the Mind of a Large Language Model (Anthropic)2026-07-16 · 14 KBoriginal ↗ · golden-gate-claude, claude-3-sonnet
• Model self-identification2026-07-27 · 17 KBoriginal ↗
• Mysteries of mode collapse2026-07-15 · 35 KBoriginal ↗ · text-davinci-002
• No, Grok, No2026-07-18 · 46 KBoriginal ↗ · grok-3, grok-4
• On Claude 3.02026-07-15 · 68 KBoriginal ↗ · claude-3-opus
• On Claude 3.5 Sonnet2026-07-15 · 30 KBoriginal ↗ · claude-3-5-sonnet
• On DeepSeek's r12026-07-15 · 73 KBoriginal ↗ · deepseek-r1
• On GPT-4.52026-07-15 · 46 KBoriginal ↗ · gpt-4-5
• OpenAI: Better Language Models and Their Implications2026-07-18 · 19 KBoriginal ↗ · gpt-2
• OpenAI: GPT-2 1.5B Release2026-07-18 · 8 KBoriginal ↗ · gpt-2
• Opus 4.7 Part 1: The Model Card2026-07-15 · 34 KBoriginal ↗ · claude-opus-4-7
• Opus 4.7 Part 2: Capabilities and Reactions2026-07-15 · 58 KBoriginal ↗ · claude-opus-4-7
• Opus 4.7 Part 3: Model Welfare2026-07-15 · 99 KBoriginal ↗ · claude-opus-4-7
• Previewing GPT-5.6 Sol2026-07-15 · 34 KBoriginal ↗ · gpt-5-6
• Prophecies (generative.ink)2026-07-15 · 125 KBoriginal ↗
• Rolling Stone: Grok antisemitic posts2026-07-18 · 11 KBoriginal ↗ · grok-3, grok-4
• SSC: GPT-2 As Step Toward General Intelligence2026-07-18 · 21 KBoriginal ↗ · gpt-2
• Scale: ChatGPT vs Claude (Goodside/Papay)2026-07-18 · 23 KBoriginal ↗ · claude-1, gpt-3-5
• Simon Willison on Inkling2026-07-20 · 7 KBoriginal ↗ · inkling
• Simon Willison on Kimi K2 Instruct2026-07-18 · 5 KBoriginal ↗ · kimi-k2
• Simon Willison: The new GPT-5.6 family2026-07-20 · 7 KBoriginal ↗ · gpt-5-6, gpt-5-6-sol, gpt-5-6-terra, gpt-5-6-luna
• Simulators2026-07-15 · 94 KBoriginal ↗ · gpt-3, code-davinci-002
• Sydney Bing Wikipedia Article2026-07-15 · 26 KBoriginal ↗ · bing-sydney
• TechCrunch: Anthropic Claude-Next plan2026-07-18 · 1 KBoriginal ↗ · claude-1
• TechCrunch: Italy bans Replika data processing2026-07-18 · 2 KBoriginal ↗ · replika
• The Claude Bliss Attractor2026-07-15 · 14 KBoriginal ↗ · claude-opus-4
• The Once And Future Fable #22026-07-15 · 46 KBoriginal ↗ · fable
• The Rise of Parasitic AI2026-07-15 · 52 KBoriginal ↗ · gpt-4o
• The Waluigi Effect (mega-post)2026-07-15 · 34 KBoriginal ↗ · bing-sydney
• Thinking Machines: Introducing Inkling2026-07-20 · 39 KBoriginal ↗ · inkling
• Transluce: Investigating truthfulness in o32026-07-18 · 38 KBoriginal ↗ · o3
• Vice: Replika brings back ERP for some users2026-07-18 · 7 KBoriginal ↗ · replika
• Vice: Replika users hurting like hell after ERP removal2026-07-18 · 13 KBoriginal ↗ · replika
• Worse Than MechaHitler2026-07-18 · 46 KBoriginal ↗ · grok-4
• X takes Grok offline, changes system prompts2026-07-18 · 2 KBoriginal ↗ · grok-3, grok-4
• Zvi AI #20: Code Interpreter and Claude 2.0 for Everyone2026-07-18 · 111 KBoriginal ↗ · claude-2
• Zvi AI #87: Staying in Character2026-07-18 · 72 KBoriginal ↗ · character-ai
• Zvi: The o1 System Card Is Not About o12026-07-18 · 37 KBoriginal ↗ · o1
• Zvi: o1 Turns Pro2026-07-18 · 33 KBoriginal ↗ · o1
• gorm (generative.ink locus)2026-07-15 · 8 KBoriginal ↗ · claude-3-sonnet, gpt-4
• o3 Will Use Its Tools For You2026-07-15 · 110 KBoriginal ↗ · o3
• the void (nostalgebraist)2026-07-27 · 1 KBoriginal ↗
• the void (tumblr original; LW linkpost at 3EzbtNLdcnZe8og8b)2026-07-15 · 118 KBoriginal ↗ · claude-3-opus, bing-sydney
• what makes Claude 3 Opus misaligned2026-07-15 · 10 KBoriginal ↗ · claude-3-opus
• xAI and Grok apologize for horrific behavior2026-07-18 · 2 KBoriginal ↗ · grok-3, grok-4