跳到正文- Source: We Have a Package for You! (package hallucinations)
- Source: Universal and Transferable Adversarial Attacks on Aligned Language Models (GCG)
- Source: Tree of Thoughts: Deliberate Problem Solving with Large Language Models
- Source: The Instruction Hierarchy (OpenAI)
- Source: The Attacker Moves Second
- Source: Text Embeddings Reveal (Almost) As Much As Text
- Source: Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsity
- Source: Self-Consistency Improves Chain of Thought Reasoning in Language Models
- Source: Robust Speech Recognition via Large-Scale Weak Supervision (Whisper)
- Source: ReWOO: Decoupling Reasoning from Observations (Xu et al., 2023)
- Source: Rethinking the Role of Demonstrations: What Makes In-Context Learning Work?
- Source: Reflexion: Language Agents with Verbal Reinforcement Learning (Shinn et al., 2023)
- Source: Red Teaming Language Models with Language Models
- Source: Red Teaming Language Models to Reduce Harms
- Source: ReAct: Synergizing Reasoning and Acting in Language Models (Yao et al., 2022)
- Source: Qwen3 Technical Report and launch blog
- Source: Prompting Science Report 2: The Decreasing Value of Chain of Thought
- Source: Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layer
- Source: OSWorld Benchmark
- Source: On the Brittle Foundations of ReAct Prompting for Agentic LLMs (Verma et al., 2024)
- Source: Not what you've signed up for (indirect prompt injection)
- Source: Mixtral of Experts
- Source: Many-shot jailbreaking (Anthropic)
- Source: Lessons From Red Teaming 100 Generative AI Products
- Source: Large Language Models are Zero-Shot Reasoners
- Source: Language Models Don't Always Say What They Think
- Source: Language Models are Few-Shot Learners (GPT-3)
- Source: Language Agent Tree Search (LATS, Zhou et al., 2023)
- Source: Jailbroken: How Does LLM Safety Training Fail?
- Source: Jailbreaking Black Box Large Language Models in Twenty Queries (PAIR)
- Source: Great, Now Write an Article About That: The Crescendo Multi-Turn LLM Jailbreak Attack
- Source: Firecracker (project site and NSDI 2020 paper)
- Source: Extracting Training Data from Large Language Models
- Source: Executable Code Actions Elicit Better LLM Agents (CodeAct, Wang et al., 2024)
- Source: Design Patterns for Securing LLM Agents against Prompt Injections
- Source: Defending Against Indirect Prompt Injection Attacks With Spotlighting
- Source: Defeating Prompt Injections by Design (CaMeL)
- Source: DeepSeekMoE: Towards Ultimate Expert Specialization in Mixture-of-Experts Language Models
- Source: DeepSeek-V3 Technical Report
- Source: Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
- Source: Calibrate Before Use: Improving Few-Shot Performance of Language Models
- Source: Best-of-N Jailbreaking
- Source: AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation (Wu et al., 2023)
- Source: AgentDojo
- Source: A small number of samples can poison LLMs of any size (Anthropic)