Interleaves chain-of-thought reasoning with tool-using actions in one LLM loop: the model plans, queries a source like Wikipedia, then revises from results. Cuts hallucination versus reasoning-only prompting and beats trained agents on interactive tasks.
AI Agent Papers2024
Treats the interface between an LM agent and a computer as a design variable. A custom agent-computer interface (ACI) with concise file-edit, repo-navigation, and test commands plus compact feedback reaches 12.5% pass@1 on SWE-bench, 87.7% on HumanEvalFix.
Zirui Guo, Lianghao Xia +3University of Hong Kong, Beijing University of Posts and Telecommunications
A graph-based RAG framework pairing a knowledge graph with vector retrieval and a dual-level (low/high) query mode. New documents merge into the graph via set operations instead of triggering a rebuild, cutting the cost of keeping the index current.
The 2017 paper that replaced recurrence with pure self-attention, making sequence models fully parallelizable — and, almost as a side effect, laying the architectural foundation for nearly every large language model that followed, from BERT to GPT.
Embeds multi-head self-attention inside an LSTM-style memory, so stored memories can attend to one another instead of just sitting in separate slots — sharpening relational reasoning and topping WikiText-103, Project Gutenberg, and GigaWord.
Introduced the two-stage recipe behind the GPT lineage: unsupervised generative pre-training on unlabeled text, then supervised fine-tuning per task. A single 12-layer Transformer decoder beat bespoke architectures on 9 of 12 NLP benchmarks.
A 1.5B-parameter model trained only to predict the next token on diverse web text does translation, summarization, and QA zero-shot, with no fine-tuning. It recast NLP tasks as conditional language modeling and sparked the staged-release misuse debate.
Demonstrated that language model loss falls as a smooth power law in model size, data, and compute across more than seven orders of magnitude — turning "make it bigger" from a hunch into a budget you can plan, and justifying the GPT-3 scale-up.
At 175 billion parameters, this autoregressive model becomes a strong few-shot learner: it handles translation, QA, and reasoning from a few prompt examples with no gradient updates, establishing in-context learning as an alternative to fine-tuning.
Showed that fine-tuning a GPT model on public GitHub code yields a capable program synthesizer, and introduced HumanEval — the docstring-to-function benchmark that still anchors code-generation evaluation. A production variant powers GitHub Copilot.
Made reinforcement learning from human feedback (RLHF) the standard alignment recipe: collect demonstrations and preference rankings, train a reward model, then optimize with PPO. A 1.3B aligned model was preferred over the 175B GPT-3 by human raters.
A multimodal model that accepts image and text inputs and returns text, scoring at human level on professional exams — including a bar exam in the top 10%. Its performance was forecast from models using 1/1000th the compute, showing predictable scaling.