Elie Georges retweeted
All Plus and Team users in ChatGPT
Elie Georges retweeted
Today we launched a preview of OpenAI o1—the first in a new series of AI models.
Because o1 can reason about our safety rules in context, it can apply them more effectively.
We've rigorously tested and evaluated o1-preview, and our Preparedness Framework identified it as safe to release because it doesn't facilitate increased risks beyond what's already possible with existing resources. openai.com/index/openai-o1-s…
Elie Georges retweeted
We're releasing a preview of OpenAI o1—a new series of AI models designed to spend more time thinking before they respond.
These models can reason through complex tasks and solve harder problems than previous models in science, coding, and math. openai.com/index/introducing…
Elie Georges retweeted
excited to announce Structured Outputs -- our newest feature in the api. model outputs will now reliably follow your exact json schemas, matching the parameters and types accurately.
schema reliability is now 100% with this feature enabled, up from 86% on our past gpt-4o model
Elie Georges retweeted
Today we’re rolling out advanced Voice Mode in ChatGPT to small group of Plus users, all Plus users in fall. Richer and more natural real-time conversations make the technology less rigid — we’ve found it more collaborative and helpful and think you will as well.
Elie Georges retweeted
Our SearchGPT prototype is now live. We’re looking for feedback as we prepare to integrate the experience in ChatGPT
We’re testing SearchGPT, a temporary prototype of new AI search features that give you fast and timely answers with clear and relevant sources.
We’re launching with a small group of users for feedback and plan to integrate the experience into ChatGPT. openai.com/index/searchgpt-p…
Elie Georges retweeted
Cranking up the accuracy and safety of our models, while cutting down on the need for a lot of human-labeled data.
Rule-based rewards (RBRs) use model to provide RL signals based on a set of safety rubrics, making it easier to adapt to changing safety policies wo/ heavy dependency on human data. It also enables us to look at safety and capability in a more unified lens as a more capable grader model gives us higher quality RL signals.
Congrats to @lilianweng and team!
We’ve developed Rule-Based Rewards (RBRs) to align AI behavior safely without needing extensive human data collection, making our systems safer and more reliable for everyday use. openai.com/index/improving-m…
Elie Georges retweeted
Our new mini model is significantly pushing the cost-intelligence frontier, and I can't wait to see all the new applications that it will enable! openai.com/index/gpt-4o-mini…
Elie Georges retweeted
Launching GPT-4o mini: the most intelligent and cost-efficient small model yet!
Smarter and cheaper than GPT-3.5 Turbo, it’s ideal for function calling, large contexts, real-time interactions—and has vision capabilities. Can’t wait to see what you build!
openai.com/index/gpt-4o-mini…
Elie Georges retweeted
1/ Excited to share our new paper: "Prover-Verifier Games Improve Legibility of LLM Outputs"! We explore how to make large language model (LLM) outputs easier to check and thereby more legible. 🧵
Elie Georges retweeted
Excited to share our new work “Prover-Verifier Games improve legibility of language model outputs”! We trained strong language models to produce text that is more checkable by weak language models, and found that this also made it more legible to humans. openai.com/index/prover-veri…
Elie Georges retweeted
We’ve trained a model, CriticGPT, to catch bugs in GPT-4’s code. We’re starting to integrate such models into our RLHF alignment pipeline to help humans supervise AI on difficult tasks: openai.com/index/finding-gpt…
Elie Georges retweeted
At OpenAI, we’re working to advance scientific understanding to help improve human well-being. The AI tools we are building, like Sora, GPT-4o, DALL·E and ChatGPT, are impressive from a technical standpoint. But what really matters is how they’re starting to change the way we interact with information and ideas.
A few years ago, in my essay "Language & Coding Creativity", I wrote about how these systems represent a big shift in our relationship with language and creativity. As we keep improving these tools, our mission stays the same: to make them helpful, safe, easy to use, and available to as many people as possible. We want to help reduce the obstacles that have traditionally kept people from expressing their unique ideas and perspectives.
By carefully designing these technologies to collaborate with human creators, I think we can build wonderful tools to help artists have more control, be more innovative, and explore new frontiers of possibility. When we made DALL·E, we worked closely with artists, designers, and storytellers, trying to build a tool that fits into their creative process and helps bring their visions to life. Moving forward, I believe AI has the potential to democratize creativity on an unprecedented scale. A person’s creative potential should not be limited by their access to resources, education, or industry connections. AI tools could lower the barriers and allow anyone with an idea to create.
At the same time, we must be honest and acknowledge that AI will automate certain tasks. Just like spreadsheets changed things for accountants and bookkeepers, AI tools can do things like writing online ads or making generic images and templates. But it's important to recognize the difference between temporary creative tasks and the kind that add lasting meaning and value to society. With AI tools taking on more repetitive or mechanistic aspects of the creative process, like generating SEO metadata, we can free up human creators to focus on higher-level creative thinking and choices. This lets artists stay in control of their vision and focus their energy on the most important parts of their work.
To make sure these technologies are developed and used in a way that does the most good and the least harm, we work closely with red-teaming experts from early stages of research. We also use an iterative approach, gradually releasing tools and carefully studying how they impact the real world to guide future development. Protecting and strengthening the most valuable aspects of creativity is fundamental to our human experience.
Realizing the potential of AI is not guaranteed. It takes carefully building tools and using them responsibly, in close partnership with creators and communities they’re intended to benefit. This means putting strong safeguards in place, reducing harmful biases, and proactively dealing with potential negative effects. At OpenAI, this is at the core of how we work, and we’ve never wavered in our commitment to this as we've released new tools.
Elie Georges retweeted
What @OpenAI's new GPT-4o model means for developers
“My bet is that GPT-4o is going to usher in a wave of new applications and products that are truly audio-first.” @oliviergodement
venturebeat.com/ai/what-open…
by @carlfranzen @VentureBeat













