Google Expands Access to Imagen 3:
Advanced AI Image Generator Now Available to More Users
Google has announced that Imagen 3, its cutting-edge AI image generator, is now accessible to a wider audience. This latest version introduces enhanced capabilities such as improved image quality, refined style transfer, and higher resolution outputs. Initially available to a limited user base, the expanded rollout aims to leverage AI's potential in creative and professional applications, making advanced image synthesis technology more broadly available.
Advanced AI Image Generator Now Available to More Users
Google has announced that Imagen 3, its cutting-edge AI image generator, is now accessible to a wider audience. This latest version introduces enhanced capabilities such as improved image quality, refined style transfer, and higher resolution outputs. Initially available to a limited user base, the expanded rollout aims to leverage AI's potential in creative and professional applications, making advanced image synthesis technology more broadly available.
ChatGPT now has 200 million weekly active users
It's hardly surprising that ChatGPT has managed to double its weekly active users since last November, given how much of a buzzword AI has been over the past year and the fact that ChatGPT is the most well known generative AI chatbot.
The OpenAI service now boasts 200 million active users every week, up from 100 million in 2023. OpenAI also says 92% of Fortune 500 companies are using its products, while usage of its automated API has doubled since the release of GPT-4o Mini last month.
It's hardly surprising that ChatGPT has managed to double its weekly active users since last November, given how much of a buzzword AI has been over the past year and the fact that ChatGPT is the most well known generative AI chatbot.
The OpenAI service now boasts 200 million active users every week, up from 100 million in 2023. OpenAI also says 92% of Fortune 500 companies are using its products, while usage of its automated API has doubled since the release of GPT-4o Mini last month.
Upwork puts AI to work, uniting team members, operations and product development
Upwork is the world's largest work marketplace, connecting businesses and freelance professionals worldwide. The platform enables companies of all sizes, from startups to Fortune 500 enterprises, to hire highly skilled freelance professionals to scale their workforces and solve business problems, big and small.
Upwork saw the potential for AI to transform every facet of its business. After evaluating many leading AI systems, OpenAI stood out for its ideal combination of models and applications that Upwork could immediately use to improve its product portfolio, reduce operational overhead, and surge productivity.
“We saw ourselves becoming an ‘OpenAI shop.’ We knew we had to move quickly with our AI adoption, and OpenAI offered solutions for everything we needed. Now we’re able to leverage AI to empower our workforce, and infuse it into our product development process”
Source-Link: OPEN AI
Upwork is the world's largest work marketplace, connecting businesses and freelance professionals worldwide. The platform enables companies of all sizes, from startups to Fortune 500 enterprises, to hire highly skilled freelance professionals to scale their workforces and solve business problems, big and small.
Upwork saw the potential for AI to transform every facet of its business. After evaluating many leading AI systems, OpenAI stood out for its ideal combination of models and applications that Upwork could immediately use to improve its product portfolio, reduce operational overhead, and surge productivity.
“We saw ourselves becoming an ‘OpenAI shop.’ We knew we had to move quickly with our AI adoption, and OpenAI offered solutions for everything we needed. Now we’re able to leverage AI to empower our workforce, and infuse it into our product development process”
Source-Link: OPEN AI
Prover-Verifier Games improve legibility of language model outputs
Making sure that language models produce understandable text is crucial to making them helpful for people, especially when dealing with complex tasks like solving math problems.
We found that when we optimize the problem-solving process of strong models solely for getting the correct answer, the resulting solutions can become harder to understand. In fact, when we asked human evaluators with limited time to assess these highly optimized solutions, they made nearly twice as many errors compared to when they evaluated less optimized solutions. This finding highlights the importance of not just correctness, but also clarity and ease of verification in AI-generated text.
By training advanced language models to create text that weaker models can easily verify, we found that humans could also evaluate these texts more effectively – a process we call improving legibility.
This is where prover-verifier games come into play.
Making sure that language models produce understandable text is crucial to making them helpful for people, especially when dealing with complex tasks like solving math problems.
We found that when we optimize the problem-solving process of strong models solely for getting the correct answer, the resulting solutions can become harder to understand. In fact, when we asked human evaluators with limited time to assess these highly optimized solutions, they made nearly twice as many errors compared to when they evaluated less optimized solutions. This finding highlights the importance of not just correctness, but also clarity and ease of verification in AI-generated text.
By training advanced language models to create text that weaker models can easily verify, we found that humans could also evaluate these texts more effectively – a process we call improving legibility.
This is where prover-verifier games come into play.
New compliance and administrative tools for ChatGPT Enterprise
The new Enterprise Compliance API and eight integrations developed by leading eDiscovery and Data Loss Prevention (DLP) companies help ChatGPT Enterprise customers in regulated industries such as finance, healthcare, legal services, and government comply with logging and audit requirements.
With the API, workspace owners(opens in a new window) can efficiently audit and take action on their ChatGPT Enterprise workspace data. The API provides a record of time-stamped interactions, including conversations, uploaded files, workspace GPT configuration and metadata, memories, and workspace users. You can see the full list of permissions in our help center(opens in a new window).
Enterprise workspace owners can access the Enterprise Compliance API directly or can choose to use a third-party compliance integration to simplify the process of syncing ChatGPT Enterprise data.
The new Enterprise Compliance API and eight integrations developed by leading eDiscovery and Data Loss Prevention (DLP) companies help ChatGPT Enterprise customers in regulated industries such as finance, healthcare, legal services, and government comply with logging and audit requirements.
With the API, workspace owners(opens in a new window) can efficiently audit and take action on their ChatGPT Enterprise workspace data. The API provides a record of time-stamped interactions, including conversations, uploaded files, workspace GPT configuration and metadata, memories, and workspace users. You can see the full list of permissions in our help center(opens in a new window).
Enterprise workspace owners can access the Enterprise Compliance API directly or can choose to use a third-party compliance integration to simplify the process of syncing ChatGPT Enterprise data.
👍1
GPT-4o mini: advancing cost-efficient intelligence
GPT-4o mini surpasses GPT-3.5 Turbo and other small models on academic benchmarks across both textual intelligence and multimodal reasoning, and supports the same range of languages as GPT-4o. It also demonstrates strong performance in function calling, which can enable developers to build applications that fetch data or take actions with external systems, and improved long-context performance compared to GPT-3.5 Turbo.
GPT-4o mini has been evaluated across several key benchmarks2.
Reasoning tasks: GPT-4o mini is better than other small models at reasoning tasks involving both text and vision, scoring 82.0% on MMLU, a textual intelligence and reasoning benchmark, as compared to 77.9% for Gemini Flash and 73.8% for Claude Haiku.
GPT-4o mini surpasses GPT-3.5 Turbo and other small models on academic benchmarks across both textual intelligence and multimodal reasoning, and supports the same range of languages as GPT-4o. It also demonstrates strong performance in function calling, which can enable developers to build applications that fetch data or take actions with external systems, and improved long-context performance compared to GPT-3.5 Turbo.
GPT-4o mini has been evaluated across several key benchmarks2.
Reasoning tasks: GPT-4o mini is better than other small models at reasoning tasks involving both text and vision, scoring 82.0% on MMLU, a textual intelligence and reasoning benchmark, as compared to 77.9% for Gemini Flash and 73.8% for Claude Haiku.
Improving Model Safety Behavior with Rule-Based Rewards
To ensure AI systems behave safely and align with human values, we define desired behaviors and collect human feedback to train a "reward model." This model guides the AI by signaling desirable actions. However, collecting this human feedback for routine and repetitive tasks is often inefficient. Additionally, if our safety policies change, the feedback we've already collected might become outdated, requiring new data.
Thus, we introduce Rule-Based Rewards (RBRs) as a key component of OpenAI’s safety stack to align model behavior with desired safe behavior. Unlike human feedback, RBRs uses clear, simple, and step-by-step rules to evaluate if the model's outputs meet safety standards. When plugged into the standard RLHF pipeline, it helps maintain a good balance between being helpful while preventing harm, to ensure the model behaves safely and effectively without the inefficiencies of recurrent human inputs.
To ensure AI systems behave safely and align with human values, we define desired behaviors and collect human feedback to train a "reward model." This model guides the AI by signaling desirable actions. However, collecting this human feedback for routine and repetitive tasks is often inefficient. Additionally, if our safety policies change, the feedback we've already collected might become outdated, requiring new data.
Thus, we introduce Rule-Based Rewards (RBRs) as a key component of OpenAI’s safety stack to align model behavior with desired safe behavior. Unlike human feedback, RBRs uses clear, simple, and step-by-step rules to evaluate if the model's outputs meet safety standards. When plugged into the standard RLHF pipeline, it helps maintain a good balance between being helpful while preventing harm, to ensure the model behaves safely and effectively without the inefficiencies of recurrent human inputs.
SearchGPT Prototype
We’re testing SearchGPT, a prototype of new search features designed to combine the strength of our AI models with information from the web to give you fast and timely answers with clear and relevant sources. We’re launching to a small group of users and publishers to get feedback. While this prototype is temporary, we plan to integrate the best of these features directly into ChatGPT in the future.
A new way to search
Getting answers on the web can take a lot of effort, often requiring multiple attempts to get relevant results. We believe that by enhancing the conversational capabilities of our models with real-time information from the web, finding what you’re looking for can be faster and easier.
We’re testing SearchGPT, a prototype of new search features designed to combine the strength of our AI models with information from the web to give you fast and timely answers with clear and relevant sources. We’re launching to a small group of users and publishers to get feedback. While this prototype is temporary, we plan to integrate the best of these features directly into ChatGPT in the future.
A new way to search
Getting answers on the web can take a lot of effort, often requiring multiple attempts to get relevant results. We believe that by enhancing the conversational capabilities of our models with real-time information from the web, finding what you’re looking for can be faster and easier.
Introducing Structured Outputs in the API
Generating structured data from unstructured inputs is one of the core use cases for AI in today’s applications. Developers use the OpenAI API to build powerful assistants that have the ability to fetch data and answer questions via function calling(opens in a new window), extract structured data for data entry, and build multi-step agentic workflows that allow LLMs to take actions. Developers have long been working around the limitations of LLMs in this area via open source tooling, prompting, and retrying requests repeatedly to ensure that model outputs match the formats needed to interoperate with their systems. Structured Outputs solves this problem by constraining OpenAI models to match developer-supplied schemas and by training our models to better understand complicated schemas.
Generating structured data from unstructured inputs is one of the core use cases for AI in today’s applications. Developers use the OpenAI API to build powerful assistants that have the ability to fetch data and answer questions via function calling(opens in a new window), extract structured data for data entry, and build multi-step agentic workflows that allow LLMs to take actions. Developers have long been working around the limitations of LLMs in this area via open source tooling, prompting, and retrying requests repeatedly to ensure that model outputs match the formats needed to interoperate with their systems. Structured Outputs solves this problem by constraining OpenAI models to match developer-supplied schemas and by training our models to better understand complicated schemas.
Rakuten pairs data with AI to unlock customer insights and value
As Rakuten seeks to provide ever-increasing value to its customers, it’s hard to overstate the importance of data. This includes the vast amounts of transactional and behavioral data from Rakuten’s ecosystem of services, down to internal business data preserved in PDFs or Word documents. “For us, the data asset is the key corporate asset,” said Yusuke Kaji, General Manager of AI for Business at Rakuten Group, Inc.
With AI, Rakuten is finding innovative ways to leverage this asset: “We believe AI models, such as those developed by OpenAI, are the way to amplify the impact we can make on top of the data,” Kaji said. In partnership with OpenAI, Tokyo-headquartered Rakuten is setting a standard for how companies around the world can approach generative AI initiatives with security and privacy at the forefront.
As Rakuten seeks to provide ever-increasing value to its customers, it’s hard to overstate the importance of data. This includes the vast amounts of transactional and behavioral data from Rakuten’s ecosystem of services, down to internal business data preserved in PDFs or Word documents. “For us, the data asset is the key corporate asset,” said Yusuke Kaji, General Manager of AI for Business at Rakuten Group, Inc.
With AI, Rakuten is finding innovative ways to leverage this asset: “We believe AI models, such as those developed by OpenAI, are the way to amplify the impact we can make on top of the data,” Kaji said. In partnership with OpenAI, Tokyo-headquartered Rakuten is setting a standard for how companies around the world can approach generative AI initiatives with security and privacy at the forefront.
GPT-4o System Card
We thoroughly evaluate new models for potential risks and build in appropriate safeguards before deploying them in ChatGPT or the API. We’re publishing the model System Card together with the Preparedness Framework scorecard to provide an end-to-end safety assessment of GPT-4o, including what we’ve done to track and address today’s safety challenges as well as frontier risks.
Building on the safety evaluations and mitigations we developed for GPT-4, and GPT-4V, we’ve focused additional efforts on GPT-4o's audio capabilities which present novel risks, while also evaluating its text and vision capabilities.
We thoroughly evaluate new models for potential risks and build in appropriate safeguards before deploying them in ChatGPT or the API. We’re publishing the model System Card together with the Preparedness Framework scorecard to provide an end-to-end safety assessment of GPT-4o, including what we’ve done to track and address today’s safety challenges as well as frontier risks.
Building on the safety evaluations and mitigations we developed for GPT-4, and GPT-4V, we’ve focused additional efforts on GPT-4o's audio capabilities which present novel risks, while also evaluating its text and vision capabilities.
👍1
Zico Kolter Joins OpenAI’s Board of Directors
We’re strengthening our governance with expertise in AI safety and alignment. Zico will also join the Safety & Security Committee.
We’re announcing the appointment of Zico Kolter to OpenAI’s Board of Directors. As a professor and Director of the Machine Learning Department at Carnegie Mellon University, Zico’s work predominantly focuses on AI safety, alignment, and the robustness of machine learning classifiers. His research and expertise spans new deep network architectures, innovative methodologies for understanding the influence of data on models, and automated methods for evaluating AI model robustness, making him an invaluable technical director for our governance.
We’re strengthening our governance with expertise in AI safety and alignment. Zico will also join the Safety & Security Committee.
We’re announcing the appointment of Zico Kolter to OpenAI’s Board of Directors. As a professor and Director of the Machine Learning Department at Carnegie Mellon University, Zico’s work predominantly focuses on AI safety, alignment, and the robustness of machine learning classifiers. His research and expertise spans new deep network architectures, innovative methodologies for understanding the influence of data on models, and automated methods for evaluating AI model robustness, making him an invaluable technical director for our governance.
Introducing SWE-bench Verified
We’re releasing a human-validated subset of SWE-bench that more reliably evaluates AI models’ ability to solve real-world software issues.
One of the most popular evaluation suites for software engineering is SWE-bench(opens in a new window)1—a benchmark for evaluating large language models’ (LLMs’) abilities to solve real-world software issues sourced from GitHub. The benchmark involves giving agents a code repository and issue description, and challenging them to generate a patch that resolves the problem described by the issue. Coding agents have made impressive progress on SWE-bench, with top scoring agents scoring 20% on SWE-bench and 43% on SWE-bench Lite according to the SWE-bench leaderboard(opens in a new window) as of August 5, 2024.
Our testing identified some SWE-bench tasks which may be hard or impossible to solve, leading to SWE-bench systematically underestimating models’ autonomous software engineering capabilities.
We’re releasing a human-validated subset of SWE-bench that more reliably evaluates AI models’ ability to solve real-world software issues.
One of the most popular evaluation suites for software engineering is SWE-bench(opens in a new window)1—a benchmark for evaluating large language models’ (LLMs’) abilities to solve real-world software issues sourced from GitHub. The benchmark involves giving agents a code repository and issue description, and challenging them to generate a patch that resolves the problem described by the issue. Coding agents have made impressive progress on SWE-bench, with top scoring agents scoring 20% on SWE-bench and 43% on SWE-bench Lite according to the SWE-bench leaderboard(opens in a new window) as of August 5, 2024.
Our testing identified some SWE-bench tasks which may be hard or impossible to solve, leading to SWE-bench systematically underestimating models’ autonomous software engineering capabilities.
Collaborating with The Met to Awaken “Sleeping Beauties” with AI
At OpenAI, we believe AI can enrich our lives by making them more creative and beautiful. Our recent collaboration with the Metropolitan Museum of Art's Costume Institute for their exhibit "Sleeping Beauties: Reawakening Fashion" showcases this potential.
Through this partnership, we created a custom chat experience to bring the world of Natalie Potter, a New York socialite from the early 20th century, to life. Visitors can explore her 1931 wedding dress and engage with an AI representation of Natalie to learn about her life, wedding, and era.
To build the experience, we worked with the museum’s digital team to curate a dataset of letters, newspaper articles, and historical documents. Leveraging OpenAI’s most advanced language model with custom instructions based on these sources, we created a custom chat experience that responds authentically to Natalie’s character and time period. This collaborative effort with the museum’s historians and curators ensures the AI is both helpful and respectful of the historical content, while giving the museum-goers a more active role in the exhibit. The "Chat with Natalie" experience incorporates the same safety mechanisms as ChatGPT, ensuring safe and appropriate interactions for all.
While there is still work to do to achieve a world where everyone can benefit from AI, collaborations like this push us towards that ideal. Examples like this highlight the potential for AI to drive human progress, and showcase how AI can be a tool to enhance how we think, create and experience the world around us.
At OpenAI, we believe AI can enrich our lives by making them more creative and beautiful. Our recent collaboration with the Metropolitan Museum of Art's Costume Institute for their exhibit "Sleeping Beauties: Reawakening Fashion" showcases this potential.
Through this partnership, we created a custom chat experience to bring the world of Natalie Potter, a New York socialite from the early 20th century, to life. Visitors can explore her 1931 wedding dress and engage with an AI representation of Natalie to learn about her life, wedding, and era.
To build the experience, we worked with the museum’s digital team to curate a dataset of letters, newspaper articles, and historical documents. Leveraging OpenAI’s most advanced language model with custom instructions based on these sources, we created a custom chat experience that responds authentically to Natalie’s character and time period. This collaborative effort with the museum’s historians and curators ensures the AI is both helpful and respectful of the historical content, while giving the museum-goers a more active role in the exhibit. The "Chat with Natalie" experience incorporates the same safety mechanisms as ChatGPT, ensuring safe and appropriate interactions for all.
While there is still work to do to achieve a world where everyone can benefit from AI, collaborations like this push us towards that ideal. Examples like this highlight the potential for AI to drive human progress, and showcase how AI can be a tool to enhance how we think, create and experience the world around us.
Indeed uses OpenAI to deliver contextual job matching to millions of job seekers
Since Indeed’s inception, AI has powered the millions of connections between job seekers and employers on the platform, through features such as ‘Invite to Apply’ which sends AI-based job recommendations to job seekers based on their resume, Indeed Profile, and other qualifications. Improvements in AI—specifically generative AI—are helping match job seekers to jobs in new and exciting ways. Using OpenAI's GPT models and fine-tuning capabilities, Indeed enhanced the personalized language in the ‘Invite to Apply’ feature to better explain why a candidate’s background or previous work experience makes a job a good fit.
Since Indeed’s inception, AI has powered the millions of connections between job seekers and employers on the platform, through features such as ‘Invite to Apply’ which sends AI-based job recommendations to job seekers based on their resume, Indeed Profile, and other qualifications. Improvements in AI—specifically generative AI—are helping match job seekers to jobs in new and exciting ways. Using OpenAI's GPT models and fine-tuning capabilities, Indeed enhanced the personalized language in the ‘Invite to Apply’ feature to better explain why a candidate’s background or previous work experience makes a job a good fit.
Fine-tuning now available for GPT-4o
Fine-tune custom versions of GPT-4o to increase performance and accuracy for your applications.
Today, we’re launching fine-tuning for GPT-4o, one of the most requested features from developers. We are also offering 1M training tokens per day for free for every organization through September 23.
Developers can now fine-tune GPT-4o with custom datasets to get higher performance at a lower cost for their specific use cases. Fine-tuning enables the model to customize structure and tone of responses, or to follow complex domain-specific instructions. Developers can already produce strong results for their applications with as little as a few dozen examples in their training data set.
From coding to creative writing, fine-tuning can have a large impact on model performance across a variety of domains. This is just the start—we’ll continue to invest in expanding our model customization options for developers.
Fine-tune custom versions of GPT-4o to increase performance and accuracy for your applications.
Today, we’re launching fine-tuning for GPT-4o, one of the most requested features from developers. We are also offering 1M training tokens per day for free for every organization through September 23.
Developers can now fine-tune GPT-4o with custom datasets to get higher performance at a lower cost for their specific use cases. Fine-tuning enables the model to customize structure and tone of responses, or to follow complex domain-specific instructions. Developers can already produce strong results for their applications with as little as a few dozen examples in their training data set.
From coding to creative writing, fine-tuning can have a large impact on model performance across a variety of domains. This is just the start—we’ll continue to invest in expanding our model customization options for developers.
👍1
We are making progress on our mission to ensure that artificial general intelligence benefits all of humanity. Every week, over 250 million people around the world use ChatGPT to enhance their work, creativity, and learning. Across industries, businesses are improving productivity and operations, and developers are leveraging our platform to create a new generation of applications. And we’re only getting started.
We’ve raised $6.6B in new funding at a $157B post-money valuation to accelerate progress on our mission. The new funding will allow us to double down on our leadership in frontier AI research, increase compute capacity, and continue building tools that help people solve hard problems.
We aim to make advanced intelligence a widely accessible resource. We’re grateful to our investors for their trust in us, and we look forward to working with our partners, developers, and the broader community to shape an AI-powered ecosystem and future that benefits everyone.
We’ve raised $6.6B in new funding at a $157B post-money valuation to accelerate progress on our mission. The new funding will allow us to double down on our leadership in frontier AI research, increase compute capacity, and continue building tools that help people solve hard problems.
We aim to make advanced intelligence a widely accessible resource. We’re grateful to our investors for their trust in us, and we look forward to working with our partners, developers, and the broader community to shape an AI-powered ecosystem and future that benefits everyone.
👍1
Learning to Reason with LLMs
We are introducing OpenAI o1, a new large language model trained with reinforcement learning to perform complex reasoning. o1 thinks before it answers—it can produce a long internal chain of thought before responding to the user.
OpenAI o1 ranks in the 89th percentile on competitive programming questions (Codeforces), places among the top 500 students in the US in a qualifier for the USA Math Olympiad (AIME), and exceeds human PhD-level accuracy on a benchmark of physics, biology, and chemistry problems (GPQA). While the work needed to make this new model as easy to use as current models is still ongoing, we are releasing an early version of this model, OpenAI o1-preview, for immediate use in ChatGPT and to trusted API users(opens in a new window).
Our large-scale reinforcement learning algorithm teaches the model how to think productively using its chain of thought in a highly data-efficient training process.
We are introducing OpenAI o1, a new large language model trained with reinforcement learning to perform complex reasoning. o1 thinks before it answers—it can produce a long internal chain of thought before responding to the user.
OpenAI o1 ranks in the 89th percentile on competitive programming questions (Codeforces), places among the top 500 students in the US in a qualifier for the USA Math Olympiad (AIME), and exceeds human PhD-level accuracy on a benchmark of physics, biology, and chemistry problems (GPQA). While the work needed to make this new model as easy to use as current models is still ongoing, we are releasing an early version of this model, OpenAI o1-preview, for immediate use in ChatGPT and to trusted API users(opens in a new window).
Our large-scale reinforcement learning algorithm teaches the model how to think productively using its chain of thought in a highly data-efficient training process.