8/31/2023
Teaching with AI
Guide for educators using ChatGPT in classrooms with prompts, explanations, and limitations.
8/28/2023
Introducing ChatGPT Enterprise
ChatGPT Enterprise offers enterprise-grade security, privacy, and advanced capabilities.
8/24/2023
OpenAI partners with Scale to provide support for enterprises fine-tuning models
OpenAI partners with Scale to help enterprises fine-tune advanced models.
8/22/2023
GPT-3.5 Turbo fine-tuning and API updates
GPT-3.5 Turbo fine-tuning now available; developers can customize with custom data.
8/15/2023
Using GPT-4 for content moderation
GPT-4 used for content moderation and policy development with faster feedback loops.
7/20/2023
Custom instructions for ChatGPT
ChatGPT custom instructions feature lets users set preferences for all conversations.
6/13/2023
Function calling and other API updates
OpenAI API updates: function calling, improved model control, extended context window, and reduced pricing.
5/31/2023
Improving mathematical reasoning with process supervision
Training method for improved mathematical reasoning using process supervision (rewarding reasoning steps) vs outcome supervision; improves alignment via interpretable chain-of-thought.
5/9/2023
Language models can explain neurons in language models
Using GPT-4 to automatically generate and score explanations for neuron behavior in GPT-2; releases dataset of explanations for interpretability research.
3/23/2023
ChatGPT plugins
OpenAI announces ChatGPT plugins enabling LLMs to access real-time data, run computations, and integrate third-party services.
3/17/2023
GPTs are GPTs: An early look at the labor market impact potential of large language models
Research on labor market impact of large language models including analysis of economic displacement and wage effects.
3/14/2023
GPT-4
GPT-4 multimodal model accepts text/images, achieves human-level performance on professional/academic benchmarks. Major LLM release.
2/16/2023
How should AI systems behave, and who should decide?
Explanation of ChatGPT behavior design, customization plans, and public input approach. AI alignment/governance.
2/1/2023
Introducing ChatGPT Plus
ChatGPT subscription plan launch with premium features. Product announcement.
1/17/2023
Let's build GPT: from scratch, in code, spelled out.
Step-by-step implementation of GPT architecture from scratch with detailed code and explanations of each component.
1/11/2023
Forecasting potential misuses of language models for disinformation campaigns and how to reduce risk
Research collaboration investigating LLM misuse for disinformation with policy and ML expert input. Safety/policy research.
1/3/2023
Fine-tuning GPT-3 to scale video creation
GPT-3 fine-tuned for automated video creation at scale. LLM application case study.
12/16/2022
Point-E: A system for generating 3D point clouds from complex prompts
System for generating 3D point clouds from text prompts. Multimodal generative model.
11/30/2022
Introducing ChatGPT
Conversational AI model trained to answer follow-ups, admit errors, and reject inappropriate requests. LLM release.
10/19/2022
Scaling laws for reward model overoptimization
Research on scaling laws governing reward model overoptimization. Machine learning research publication.
9/7/2022
The spelled-out intro to language modeling: building makemore
Educational tutorial building a character-level language model from scratch, covering foundational concepts in neural networks and language modeling.
8/24/2022
Our approach to alignment research
OpenAI's research focus on learning from human feedback and building aligned AI systems to address broader alignment challenges.
7/28/2022
Efficient training of language models to fill in the middle
Efficient training methods for language models to perform infilling tasks in code and text.
7/25/2022
A hazard analysis framework for code synthesis large language models
Framework for hazard analysis of code synthesis large language models addressing safety and reliability.
7/18/2022
Reducing bias and improving safety in DALL·E 2
DALL-E 2 implements technique to generate images of people reflecting global population diversity.
6/28/2022
DALL·E 2 pre-training mitigations
DALL-E 2 safety guardrails implemented to prevent policy-violating image generation.
6/23/2022
Learning to play Minecraft with Video PreTraining
Neural network trained via video pretraining to play Minecraft using native human controls; achieves diamond tool crafting as step toward general computer-using agents.
6/9/2022
Techniques for training large neural networks
Technical approaches to training large neural networks using GPU clusters for synchronized computation.
6/2/2022
Best practices for deploying language models
Best practices guide for developing and deploying large language models from Cohere, OpenAI, and AI21 Labs.
5/18/2022
DALL·E 2 research preview update
DALL-E 2 research preview expanded with 3 million user-generated images and improved safety processes.
4/13/2022
Hierarchical text-conditional image generation with CLIP latents
Research on hierarchical text-conditional image generation using CLIP latents for improved image synthesis.
3/15/2022
New GPT-3 capabilities: Edit & insert
GPT-3 and Codex released with edit and insert capabilities for text modification beyond completion.
3/3/2022
A research agenda for assessing the economic impacts of code generation models
Research agenda for evaluating economic impacts of code generation models.
3/3/2022
Lessons learned on language model safety and misuse
Safety and misuse mitigation lessons from deploying large language models.
1/27/2022
Aligning language models to follow instructions
InstructGPT: language models aligned with human feedback to follow instructions, improve truthfulness and reduce toxicity.
1/25/2022
Introducing text and code embeddings
OpenAI API embeddings endpoint for semantic search, clustering, topic modeling, and classification tasks.
1/24/2022
Text and code embeddings by contrastive pre-training
Text and code embeddings learned via contrastive pre-training.
12/16/2021
WebGPT: Improving the factual accuracy of language models through web browsing
Fine-tuned GPT-3 with text-based web browser interface for improved factual accuracy in open-ended question answering.
12/14/2021
Customizing GPT-3 for your application
Single-command fine-tuning for GPT-3 customization to specific applications.
11/30/2021
OpenAI Residency
OpenAI Residency program announcement for AI talent development.
10/29/2021
Solving math word problems
System solving grade-school math word problems at 55% accuracy, approaching human performance of 60%.
7/28/2021
Introducing Triton: Open-source GPU programming for neural networks
Triton 1.0: open-source Python-like language for efficient GPU neural network programming without CUDA expertise.
7/7/2021
Evaluating large language models trained on code
Evaluation methods for large language models trained on code.
6/10/2021
Improving language model behavior by training on a curated dataset
Fine-tuning language models on curated datasets to improve specific behavioral values.
3/25/2021
GPT-3 powers the next generation of apps
Over 300 applications using GPT-3 API for search, conversation, text completion and advanced AI features.
3/4/2021
Multimodal neurons in artificial neural networks
Discovered multimodal neurons in CLIP responding to same concepts across literal, symbolic, conceptual presentations.
1/25/2021
Scaling Kubernetes to 7,500 nodes
Kubernetes scaled to 7,500 nodes for infrastructure supporting large models like GPT-3, CLIP, DALL·E.
1/5/2021
CLIP: Connecting text and images
CLIP: neural network learning visual concepts from text supervision; enables zero-shot visual classification.
1/5/2021
DALL·E: Creating images from text
DALL·E: neural network generating images from text captions across diverse concepts.
9/4/2020
Learning to summarize with human feedback
Applied reinforcement learning from human feedback to improve language model summarization capabilities.