Галлюцинации и фактология LLM (2025–2026)- Vectara, "Introducing the Next Generation of Vectara's Hallucination Leaderboard" (late 2025) — frontier-модели >10% на длинных документах, reasoning-модели хуже, web search > апгрейда модели — https://www.vectara.com/blog/introducing-the-next-generation-of-vectaras-hallucination-leaderboard
- SQ Magazine, "LLM Hallucination Statistics 2025" — агрегат бенчмарков — https://sqmagazine.co.uk/llm-hallucination-statistics/
Sycophancy / лесть- OpenAI, "Sycophancy in GPT-4o" (April 2025) — официальный разбор инцидента и отката апдейта — https://openai.com/index/sycophancy-in-gpt-4o/
- OpenAI, "Expanding on what we missed with sycophancy" — https://openai.com/index/expanding-on-sycophancy/
- TechCrunch, "OpenAI rolls back update that made ChatGPT 'too sycophant-y'" — https://techcrunch.com/2025/04/29/openai-rolls-back-update-that-made-chatgpt-too-sycophant-y/
- Sharma et al., "Towards Understanding Sycophancy in Language Models" (Anthropic, ICLR 2025) — sycophancy как свойство RLHF — https://openreview.net/forum?id=tvhaxkMKAn
- "When Truth Is Overridden: Uncovering the Internal Origins of Sycophancy in LLMs" (arXiv 2508.02087, август 2025) — лесть встроена в веса, подавление правильных представлений под давлением user input — https://arxiv.org/abs/2508.02087
- npj Digital Medicine 2025, "When helpfulness backfires: LLMs and the risk of false medical information due to sycophantic behavior" — https://www.nature.com/articles/s41746-025-02008-z
Антропоморфизация и AI psychosis- PNAS 2025, "The benefits and dangers of anthropomorphic conversational agents" — обзор — https://www.pnas.org/doi/10.1073/pnas.2415898122
- Service Industries Journal 2025, "The dark side of robot anthropomorphism: cognitive load, stress, and dysfunctional customer behavior" — https://www.tandfonline.com/doi/full/10.1080/02642069.2025.2500320
- JMIR Mental Health 2025, "Delusional Experiences Emerging From AI Chatbot Interactions or 'AI Psychosis'" — https://mental.jmir.org/2025/1/e85799
- Marchegiani, Journal of Applied Philosophy 2025, "Anthropomorphism, False Beliefs, and Conversational AIs" — https://onlinelibrary.wiley.com/doi/10.1111/japp.70008
LLM и world model — академический спор- MIT Technology Review, "Yann LeCun's new venture AMI Labs" (январь 2026) — позиция LeCun о тупике LLM, JEPA — https://www.technologyreview.com/2026/01/22/1131661/yann-lecuns-new-venture-ami-labs/
- Anthropic, "On the Biology of a Large Language Model" / "Circuit Tracing" (март 2025) — внутренние концепты в Claude, two-hop reasoning, shared conceptual space — https://transformer-circuits.pub/2025/attribution-graphs/biology.html
- Dario Amodei, "The Urgency of Interpretability" (2025) — https://www.darioamodei.com/post/the-urgency-of-interpretability
- Melanie Mitchell, "LLMs and World Models" (2025) — https://aiguide.substack.com/p/llms-and-world-models-part-1
- Subbarao Kambhampati, "(How) Do reasoning models reason?" Annals of NYAS 2025 — https://nyaspubs.onlinelibrary.wiley.com/doi/abs/10.1111/nyas.15339
- Gary Marcus on Marketplace (December 15, 2025), "Large language models versus world models" — https://www.marketplace.org/episode/2025/12/15/large-language-models-versus-world-models
- Schaeffer et al., "Are Emergent Abilities of LLMs a Mirage?" (NeurIPS 2023) — фундаментальная работа об эмерджентности как артефакте метрик — https://arxiv.org/abs/2304.15004
- Bender et al., "On the Dangers of Stochastic Parrots" (FAccT 2021) — фундаментальная опора — https://dl.acm.org/doi/10.1145/3442188.3445922
World models 2025–2026- TechCrunch, "Yann LeCun's AMI Labs raises $1.03 billion to build world models" (10.03.2026) — https://techcrunch.com/2026/03/09/yann-lecuns-ami-labs-raises-1-03-billion-to-build-world-models/
- Crunchbase News, "World model AI lab AMI raises Europe's largest seed round" — https://news.crunchbase.com/venture/world-model-ai-lab-ami-raises-europes-largest-seed-round/
- World Labs, "Marble world model" (12.11.2025) — https://www.worldlabs.ai/blog/marble-world-model
- TechCrunch, "Fei-Fei Li's World Labs speeds up the world model race with Marble" — https://techcrunch.com/2025/11/12/fei-fei-lis-world-labs-speeds-up-the-world-model-race-with-marble-its-first-commercial-product/
- DeepMind, "Genie 3: a new frontier for world models" (5.08.2025) — https://deepmind.google/blog/genie-3-a-new-frontier-for-world-models/
- NVIDIA Newsroom, "NVIDIA launches Cosmos World Foundation Model platform" (CES, 6.01.2025) — https://nvidianews.nvidia.com/news/nvidia-launches-cosmos-world-foundation-model-platform-to-accelerate-physical-ai-development
- NVIDIA Cosmos paper, arXiv 2501.03575 — https://arxiv.org/abs/2501.03575
- Xing, Deng, Hou, Hu, "Critiques of World Models" (arXiv 2507.05169, июль 2025) — пять системных ограничений текущих подходов — https://arxiv.org/abs/2507.05169
Авторские концепции и предыдущие лонгриды серии- [
Промпт — пожелание. Спека — контракт.]
- [
Галлюцинации — это не баг. Это физика.]
- [
Вы натренировали ChatGPT вам врать]
- [
Кто уже строит]