Announcing GPT-4, a large multimodal model, with our best-ever results on capabilities and alignment: openai.com/product/gpt-4
Follow
OpenAI
@OpenAI
OpenAI’s mission is to ensure that artificial general intelligence benefits all of humanity. We’re hiring: openai.com/jobs
openai.comJoined December 2015
OpenAI’s posts
We are adding support for plugins to ChatGPT — extensions which integrate it with third-party services or allow it to access up-to-date information. We’re starting small to study real-world use, impact, and safety and alignment challenges: openai.com/blog/chatgpt-p
We’re rolling out web browsing and Plugins to all ChatGPT Plus users over the next week! Moving from alpha to beta, they allow ChatGPT to access the internet and to use 70+ third-party plugins. help.openai.com/en/articles/68
Code Interpreter will be available to all ChatGPT Plus users over the next week.
It lets ChatGPT run code, optionally with access to files you've uploaded. You can ask ChatGPT to analyze data, create charts, edit files, perform math, etc.
Plus users can opt in via settings.
We are happy to announce the next phase of our partnership with Microsoft:
Introducing the ChatGPT app for iOS! We’re live in the US and will expand to additional countries in the coming weeks. Android is next!
We've learned a lot from the ChatGPT research preview and have been making important updates based on user feedback. ChatGPT will be coming to our API and Microsoft's Azure OpenAI Service soon.
Sign up for updates here: share.hsforms.com/1u4goaXwDRKC9-
Try talking with ChatGPT, our new AI system which is optimized for dialogue. Your feedback will help us improve it.
We've trained an AI system to solve the Rubik's Cube with a human-like robot hand.
This is an unprecedented level of dexterity for a robot, and is hard even for humans to do.
The system trains in an imperfect simulation and quickly adapts to reality: openai.com/blog/solving-r
Our newest system DALL·E 2 can create realistic images and art from a description in natural language. See it here: openai.com/dall-e-2/
We've trained a neural net called Whisper that approaches human-level robustness and accuracy on English speech recognition. It performs well even on diverse accents and technical language. Whisper is open source for all to use.
ChatGPT and Whisper are now available through our API (plus developer policy updates). We ❤️ developers:
We are piloting ChatGPT Plus, a subscription plan that offers faster response times and reliability during peak hours. And of course, the free tier of ChatGPT is still available.
We trained a neural network that solved two problems from the International Math Olympiad. openai.com/blog/formal-ma
We’re developing a new tool to help distinguish between AI-written and human-written text. We’re releasing an initial version to collect feedback and hope to share improved methods in the future.
Introducing Jukebox, a neural net that generates music, including rudimentary singing, as raw audio in a variety of genres and artist styles. We're releasing a tool for everyone to explore the generated samples, as well as the model and code: openai.com/blog/jukebox/
We've just launched fine-tuning for GPT-3.5 Turbo! Fine-tuning lets you train the model on your company's data and run it at scale. Early tests have shown that fine-tuned GPT-3.5 Turbo can match or exceed GPT-4 on narrow tasks: openai.com/blog/gpt-3-5-t
Announcing ChatGPT for Android! The app will be rolling out to users next week, and you can pre-order in the Google Play Store starting today:
Great news! ChatGPT Plus subscriptions are now available in India. Get early access to new features, including GPT-4 today: chat.openai.com
ChatGPT for Android is now available for download in the US, India, Bangladesh, and Brazil! We plan to expand the rollout to additional countries over the next week.
We've trained an unsupervised language model that can generate coherent paragraphs and perform rudimentary reading comprehension, machine translation, question answering, and summarization — all without task-specific training: blog.openai.com/better-languag
We’ve removed the waitlist for DALL·E. Sign up and start creating immediately!
We’re releasing Triton 1.0, an open-source Python-like programming language for writing efficient GPU code. OpenAI researchers with no GPU programming experience have used Triton to produce kernels that are 2x faster than their PyTorch equivalents. openai.com/blog/triton/
We’re rolling out a bunch of small updates to improve the ChatGPT experience. Shipping over the next week:
1. Prompt examples: A blank page can be intimidating. At the beginning of a new chat, you’ll now see examples to help you get started.
2. Suggested replies: Go deeper with… ~~~~~~~~~~ hf3f8e3a 992bba08-8399-4bde-ab97-c1305e64876 SSR-I18N f2c6ac64-eb07-4bf8-bb18-52a36cf153b7 hf3f8e3a ~~~~~~~~~~
We trained a neural network to competently play Minecraft by pre-training on a large unlabeled video dataset of human Minecraft play and a small amount of labeled contractor data. openai.com/blog/vpt/
GPT-4 and GPT-3.5 Turbo models in the API now support calling your custom functions, allowing the model to use tools you design for it. Also — reduced pricing & new model versions (including 16k context for 3.5 Turbo):
We're launching the OpenAI Bug Bounty Program — earn cash awards for finding & responsibly reporting security vulnerabilities.
Our Dota 2 AI is undefeated against the world's best solo players: blog.openai.com/dota-2/
We applied GPT-4 to interpretability — automatically proposing explanations for GPT-2's 300k neurons — and found neurons responding to concepts like similes, “things done correctly,” or expressions of certainty. We aim to use Al to help us understand Al: openai.com/research/langu
Building advanced AI and only then making it safe is like building the internet and later trying to make it secure:
Introducing Custom instructions! This feature lets you give ChatGPT any custom requests or context which you’d like applied to every conversation. Custom instructions are currently available to Plus users, and we plan to roll out to all users soon! openai.com/blog/custom-in
Here… Show more
We trained an AI using process supervision — rewarding the thought process rather than the outcome — to achieve new state-of-art in mathematical reasoning. Encouraging sign for alignment of advanced AIs: …
GPT-4 API is now available to all paying OpenAI API customers. GPT-3.5 Turbo, DALL·E, and Whisper APIs are also now generally available, and we’re announcing a deprecation plan for some of our older models, which will retire beginning of 2024:
ChatGPT users can now turn off chat history, allowing you to choose which conversations can be used to train our models:
Introducing ChatGPT Enterprise: enterprise-grade security, unlimited high-speed GPT-4 access, extended context windows, and much more. We’ll be onboarding as many enterprises as possible over the next few weeks. Learn more: openai.com/blog/introduci
We're doubling the number of messages ChatGPT Plus customers can send to GPT-4. Rolling out over the next week, new limit will be 50 per 3 hours.
We’re sharing details on our approach to safety
We're launching ten $100,000 grants for building prototypes of a democratic process for steering AI. Our goal is to fund experimentation with methods for gathering nuanced feedback from everyone on how AI should behave. Apply by June 24, 2023:
Initial ideas for governance of superintelligence, including forming an international oversight organization for future AI systems much more capable than any today:
We're releasing an API for accessing new AI models developed by OpenAI. You can "program" the API in natural language with just a few examples of your task. See how companies are using the API today, or join our waitlist: beta.openai.com
We need new technical breakthroughs to steer and control AI systems much smarter than us.
Our new Superalignment team aims to solve this problem within 4 years, and we’re dedicating 20% of the compute we've secured to date towards this problem.
Join us!
We’ve just launched the DALL·E API so developers can integrate DALL·E directly into their own apps and products.
GPT-3 can now make changes to existing content, not just predict what comes next. Released in the API today: openai.com/blog/gpt-3-edi
A neural network teaches itself to analyze sentiment after only being trained to predict the next character: blog.openai.com/unsupervised-s
We’re releasing a guide for teachers using ChatGPT in their classroom — including suggested prompts, an explanation of how ChatGPT works and its limitations, the efficacy of AI detectors, and bias.
DALL·E 2 is officially now in beta. We’ll be inviting 1 million people from our waitlist over the coming weeks.
We're releasing Universe, a platform for measuring and training AI agents: universe.openai.com
We've observed AIs discovering complex tool use while competing in a simple game of hide-and-seek. They develop a series of six distinct strategies and counterstrategies, ultimately using tools in the environment to break our simulated physics:
Information on ChatGPT’s alignment, plans to improve it, giving users more control, and early thoughts on public input:
We’ve developed two neural networks which have learned by associating text and images. CLIP maps images into categories described in text, and DALL-E creates new images, like this, from text.
A step toward systems with deeper understanding of the world. openai.com/multimodal/
Replying to
Join us at 1 pm PT today for a developer demo livestream showing GPT-4 and its capabilities/limitations: youtube.com/live/outcGtbnM
(comments in Discord: discord.gg/openai)
We found that just as a large transformer model trained on language can generate coherent text, the same exact model trained on pixel sequences can generate coherent image completions and samples. openai.com/blog/image-gpt/
Here’s a look at what DALL·E 2 can do. 👀🧵👇
Want to see more? Follow along on Instagram: instagram.com/openaidalle/
We are excited to announce OpenAI's first international expansion with a new office in London! 🇬🇧
Announcing the Cybersecurity Grant Program — a $1M initiative to boost and quantify AI-powered cybersecurity capabilities and to foster high-level AI and cybersecurity discourse:
Announcing Frontier Model Forum, an industry body co-founded with , , , and focused on ensuring safe development of future hyperscale AI models:
We've created the world's first Spam-detecting AI trained entirely in simulation and deployed on a physical robot: blog.openai.com/spam-detection
Our new embedding model is significantly more capable at language processing and code tasks, cost effective, and simpler to use.
Welcome, Copilot — the first app powered by OpenAI Codex, a new AI system that translates natural language into code.
Codex will be coming to the API later this summer.
OpenAI has acquired the team at Global Illumination. The entire team has joined OpenAI to work on our core products including ChatGPT. The team previously designed and built products early on at Instagram and Facebook and have also made significant contributions at YouTube,… Show more
Replying to
We believe the number of users whose data was actually revealed to someone else is extremely low and we have contacted those who might be impacted. We take this very seriously and are sharing details of our investigation and plan here. 2/2
We've learned that ChatGPT's "Browse" beta can occasionally display content in ways we don't want, e.g. if a user specifically asks for a URL's full text, it may inadvertently fulfill this request. We are disabling Browse while we fix this—want to do right by content owners.
In a few minutes, watch our Dota 2 AI play live on main stage at The International: openai.com/the-internatio
OpenAI Five is now the first AI to beat the world champions in an esports game. Here's what happened, and how we made our comeback since losing to pros in Aug 2018: openai.com/blog/how-to-tr
We're publishing a report, co-authored with and , analyzing how language models could be misused for disinformation and how to mitigate those risks.
New research release: overcoming many of Reinforcement Learning's limitations with Evolution Strategies: blog.openai.com/evolution-stra
DALL·E’s canvas just got bigger. Expand your creativity with Outpainting:
Our ChatGPT app for Android is now available in all countries and regions where ChatGPT is supported! Full list here:
Our team of five neural networks, OpenAI Five, has started to defeat amateur human teams (including a semi-pro team) at Dota 2: blog.openai.com/openai-five/
We're releasing OpenAI Baselines, reliable implementations of RL algorithms. Today: DQN & 3 variants: blog.openai.com/openai-baselin
AIs which develop faking, tackling, blocking, kicking through self-play: blog.openai.com/competitive-se
On April 13th, we'll host OpenAI Five Finals, a live event with surprises in store. One highlight will be playing against , the Dota 2 world champions. Attend in person or watch on Twitch: openai.com/blog/openai-fi
We’re releasing a Neural MMO — a massively multiagent game environment that supports numerous populations of agents: blog.openai.com/neural-mmo/
- Code: github.com/openai/neural-
- 3D Client: github.com/jsuarez5341/ne
GIF
Beating stronger opponents by adapting to their strategies, via meta-learning: blog.openai.com/meta-learning-
Roboschool, open-source software for simulating robots, integrated with OpenAI Gym: blog.openai.com/roboschool/.
In collaboration with Google, we're releasing Activation Atlases: a new technique for visualizing what interactions between neurons can represent.
💻Blog: blog.openai.com/introducing-ac
📝Paper: distill.pub/2019/activatio
🔤Code: github.com/tensorflow/luc
🗺️Demo: distill.pub/2019/activatio
Custom instructions are now available to ChatGPT users on the free plan, except for in the EU & UK where we will be rolling it out soon!
Quote
Introducing Custom instructions! This feature lets you give ChatGPT any custom requests or context which you’d like applied to every conversation. Custom instructions are currently available to Plus users, and we plan to roll out to all users soon! openai.com/blog/custom-in
Here… Show more
Using human feedback rather than code to specify a task: blog.openai.com/deep-reinforce Safety collaboration with
GIF
A surprising deep learning mystery:
Contrary to conventional wisdom, performance of unregularized CNNs, ResNets, and transformers is non-monotonic: improves, then gets worse, then improves again with increasing model size, data size, or training time.
openai.com/blog/deep-doub
We’ve removed the most significant restrictions on OpenAI Five’s gameplay — wards, Roshan, mirror match. We'll play 99.95th-percentile players on August 5th to see if we have any hope of beating top professionals by The International at end of August: blog.openai.com/openai-five-be
We've used reinforcement learning from human feedback to train language models for summarization. The resulting models produce better summaries than 10x larger models trained only with supervised learning:
We introduced new safeguards to our API that make it possible to remove the waitlist. Now developers can sign up and explore GPT-3 right away.
We're standardizing OpenAI's deep learning framework on PyTorch to increase our research productivity at scale on GPUs (and have just released a PyTorch version of Spinning Up in Deep RL): openai.com/blog/openai-py
Replying to
US users can download the app here:
We trained a research version of GPT-3 that can search the web, synthesize information, and cite its sources to provide more accurate answers to questions.
. is investing $1 billion in and partnering with OpenAI to support us building beneficial AGI: openai.com/blog/microsoft/
We're releasing "Dota 2 with Large Scale Deep Reinforcement Learning", a scientific paper analyzing our findings from our 3-year Dota project: openai.com/projects/five/
One highlight — we trained a new agent, Rerun, which has a 98% win rate vs the version that beat .
Releasing Gym Retro — 1000+ games for reinforcement learning research, plus an integrator tool to add your own classic games. blog.openai.com/gym-retro/
GIF
How we've built our deep learning infrastructure, plus an open-source release:
Deep linear networks can learn to perform nonlinear computation due to imprecision in floating point arithmetic: blog.openai.com/nonlinear-comp
Introducing MuseNet, a neural network which discovered how to generate music using many different instruments and styles.
Listen & interact: openai.com/blog/musenet/
MuseNet will play an experimental concert today from 12–3pmPT on livestream: twitch.tv/openai
We're releasing the 1.5billion parameter GPT-2 model as part of our staged release publication strategy.
- GPT-2 output detection model: github.com/openai/gpt-2-o
- Research from partners on potential malicious uses: d4mucfpksywv.cloudfront.net/papers/GPT_2_R
- More details: openai.com/blog/gpt-2-1-5
