A comprehensive index of artificial intelligence and machine-learning research with AI-generated summaries, citation metrics, and direct links to papers and code.
Lingjiao Chen, Matei Zaharia, James Zou
This paper evaluates the March and June 2023 versions of GPT-3.5 and GPT-4 across diverse tasks, revealing substantial performance and behavior drift over time and highlighting the need for continuous monitoring.
John M. Dagdelen, Alexander Dunn, Sang‐Hoon Lee, et al.
Fine-tuning pretrained LLMs (GPT-3, Llama-2) enables joint named entity recognition and relation extraction from scientific text, producing structured records for materials chemistry.
Yan Hu, Qingyu Chen, Jingcheng Du, et al.
This paper shows that task-specific prompt engineering, incorporating medical knowledge and few-shot examples, significantly improves GPT-3.5 and GPT-4 performance on clinical NER tasks, though still below BioClinicalBERT.
Unknown
NuNER is a compact NER foundation model pre-trained via contrastive learning on GPT-3.5-annotated data, outperforming larger models in few-shot settings.
Unknown
Phi-3 is a series of language models trained on heavily filtered web and synthetic data, achieving performance comparable to much larger models like Mixtral 8x7B and GPT-3.5.
Unknown
DBRX is a 132B open sparse MoE LLM that surpasses GPT-3.5 and competes with Gemini 1.0 Pro.
Unknown
Phi-1 is a code-focused LLM trained on textbook-quality web data and synthetic textbooks/exercises generated by GPT-3.5.
Unknown
Introduces ChatGPT, an interactive conversational model built on GPT-3.5.
Unknown
Open Pre-trained Transformers (OPT) provides a suite of decoder-only models from 125M to 175B parameters, with OPT-175B matching GPT-3 performance.
Unknown
WebGPT fine-tunes GPT-3 to browse the web via text, using imitation learning and human feedback to answer long-form questions more factually.
Luciano Floridi, Massimo Chiriatti
This commentary analyzes GPT-3's nature using reversible/irreversible questions, shows it fails mathematical, Turing, and ethical tests, and argues it is not general AI.
Tom B. Brown, Benjamin Mann, Nick Ryder, et al.
Showed that scaling language models to 175 billion parameters (GPT-3) enables in-context learning — performing new tasks from just a few examples in the prompt, without fine-tuning.