OpenAI Unveils Major Upgrade to ChatGPT’s Image-Generation Capabilities

New Feature Leverages GPT-4o Model for Native Image and Photo Creation

During a livestream on Tuesday, OpenAI CEO Sam Altman announced the first major upgrade to ChatGPT’s image-generation capabilities in over a year. The company’s AI-powered chatbot platform has long been known for its ability to generate and edit text, but now, it can also natively create and modify images and photos.

How it Works

The new feature leverages OpenAI’s GPT-4o model, which has been used to power the company’s AI video-generation product, Sora. GPT-4o is capable of generating more accurate and detailed images than its predecessor, DALL-E 3, which it replaces. The model can also edit existing images, including those with people in them, by transforming or "inpainting" details like foreground and background objects.

Training Data

To power the new image feature, OpenAI trained GPT-4o on publicly available data, as well as proprietary data from its partnerships with companies like Shutterstock. This training data is a key component of the model’s ability to generate high-quality images, but it’s also a sensitive topic for many companies, who often keep it close to the chest to avoid potential IP-related lawsuits.

Policies and Guidelines

OpenAI has implemented policies to ensure the responsible use of its image-generation capabilities. The company respects the rights of artists and has an opt-out form for creators to request that their works be removed from its training datasets. Additionally, OpenAI respects requests to disallow its web-scraping bots from collecting training data, including images, from websites.

Competition and Controversy

The new feature follows on the heels of Google’s experimental native image output for Gemini 2.0 Flash, one of the company’s flagship models. While the feature went viral on social media, it was met with controversy due to a lack of guardrails, allowing users to remove watermarks and create images depicting copyrighted characters.

Conclusion

The upgrade to ChatGPT’s image-generation capabilities is a significant step forward for the company, offering users new creative possibilities and applications. As the technology continues to evolve, it’s essential for companies like OpenAI to prioritize responsible use and respect for the rights of artists and creators.

Frequently Asked Questions

• Q: What is GPT-4o?
  A: GPT-4o is a model used by OpenAI to generate and modify images and photos.
• Q: How does GPT-4o work?
  A: GPT-4o uses publicly available data and proprietary data from OpenAI’s partnerships to generate and modify images and photos.
• Q: How does OpenAI ensure responsible use of its image-generation capabilities?
  A: OpenAI has implemented policies to ensure responsible use, including an opt-out form for creators and a prohibition on web-scraping bots collecting training data from websites.

Headphone and Earbud Deals

Just as flower buds are beginning to bloom to mark the seasonal shift toward warmer months, Amazon is hosting its first big sales event of 2025. The Big Spring Sale, as it is known, is live, and runs through Monday, March 31st, bringing with it a slew of discounts on gardening supplies, bedding, and other necessities.

Speaker and Soundbar Deals

We’ve dug into all the deals and discounts available so far, the best of which you’ll find below. By and large, most of the deals will last until the end of the event. And, while having a Prime subscription means your packages will certainly arrive more quickly, you don’t necessarily need to be a member to get in on these tried-and-tested Verge favorites. We’ll also be keeping an eye on deals at competing retailers, including Best Buy, Walmart, and Target.

Tablet and E-reader Deals

Smartwatch and Fitness Tracker Deals

Update, March 25th: Updated to reflect current pricing/availability and several new deals, including those for the Glocusent book light, the Hisense U8N, and the AeroPress Go coffee maker.

Conclusion

The Big Spring Sale has something for everyone, from gardening supplies to tech gadgets. Whether you’re looking for a new pair of headphones or a smartwatch to track your fitness goals, you can find it at a discounted price on Amazon. Don’t miss out on these deals, which will be available until March 31st. Be sure to check back for updates, as new deals are being added daily.

Frequently Asked Questions

Q: Do I need to be an Amazon Prime member to participate in the Big Spring Sale?
A: No, you don’t need to be an Amazon Prime member to take advantage of the deals.

Q: How long will the Big Spring Sale last?
A: The Big Spring Sale will run through Monday, March 31st.

Q: Can I find deals at other retailers besides Amazon?
A: Yes, we’ll be keeping an eye on deals at competing retailers, including Best Buy, Walmart, and Target.

The Urgent Need for Global Cooperation on Artificial General Intelligence

The Growing Concerns About AI

There is a lot of talk in Beijing this week over when President Trump and President Xi Jinping of China will meet face to face. Some Chinese experts say the two leaders need to wait a few months until Trump decides exactly what tariffs he is going to impose on China — and sees what China will do in response.

The Birth of Artificial General Intelligence

Can I just butt in and say: “Excuse me, Mr. Presidents, but you two need to get together, like, tomorrow. But it’s not to discuss the golden oldies — tariffs, trade, and Taiwan. There is an earthshaking event coming — the birth of artificial general intelligence. The United States and China are the two superpowers closing in on A.G.I. — systems that will be as smart or smarter than the smartest human and able to learn and act on their own. Whatever you both may think you’ll be judged on by history, I assure you that whether you collaborate to create a global architecture of trust and governance over these emerging superintelligent computers, so humanity gets the best out of them and cushions their worst, will be at the top.”

The Importance of Collaboration

I realize many will consider this wasted breath with all the turmoil unleashed by the new administration in Washington, but that will not deter me from making the point as loudly as I can. Because what Soviet-American nuclear arms control was to world stability since the 1970s, U.S.-Chinese A.I. collaboration to make sure we effectively control these rapidly advancing A.I. systems will be for the stability of tomorrow’s world.

The Risks of Unchecked AI

A.I. systems and humanoid robots offer so much potential benefit to humanity, but they could be hugely destructive and destabilizing if not embedded with the right values and controls. In addition, this new age must be defined by a lot of planning about what humans will do for work, and how to preserve the dignity they derive from work, when machines will be able to do so many things better than people. Millions of people possibly losing their jobs and dignity at the same time is a prescription for disorder.

China’s Perspective

A veteran Chinese economist made clear to me that China is very alive to these risks: “Today, a lot of Chinese cannot find jobs. With A.I. they will not be able to find jobs forever. What happens if they cannot find appropriate jobs” because “70 percent of civil servants are robots? That will be super risky.”

The Necessity of Action

There is no time to lose in thinking about how we adapt, and yet we can be so nearsighted when it comes to the signs and warnings. A decade from now, what will journalists say was the most important news story in the fall of 2024 that should have received more attention, given the long-term consequences? Will they say it was the second election of Donald Trump as president in November 2024? Or will they say it was Uber’s decision in September 2024 to go beyond its pilot project in Phoenix and start offering driverless, all-electric Waymo cars on its ride-hailing app in Austin and Atlanta — replacing human Uber drivers?

Conclusion

The advances that China has made on A.I. in just the past year have made it absolutely clear that Beijing and Washington are now the world’s two A.I. superpowers. If you thought otherwise, China’s premier, Li Qiang, opened the China Development Forum, the event that drew me to Beijing, by proudly noting how China’s recently unveiled DeepSeek A.I. system “burst onto the scene,” highlighting “the huge power of innovation and creativity of the Chinese people.”

FAQs

Q: What is Artificial General Intelligence (A.G.I.)?
A: A.G.I. is the holy grail of A.I. — single systems that can master math, physics, biology, chemistry, material science, Shakespeare, poetry, and literature as well as the smartest humans but that can also reason across all of them and see connections no human polymath ever could.

Q: How will A.G.I. impact society?
A: A.G.I. will revolutionize the way we live and work, but it also poses significant risks, including job displacement, social unrest, and the potential for misuse.

Q: Can A.I. be used for good or evil?
A: A.I. can be used for both good and evil, depending on how it is designed and used. It is crucial that we develop ethical guidelines and regulations to ensure that A.I. is used responsibly.

Q: Can the United States and China collaborate on A.I.?
A: Yes, the United States and China can collaborate on A.I. to ensure that it is developed and used in a way that benefits humanity. This requires a commitment to transparency, trust, and cooperation.

Boosting AI Performance without Clean Labeled Data

The Challenge of Dirty Data

Jonathan Frankle, chief AI scientist at Databricks, has been talking to customers about the key challenges they face in getting AI to work reliably. The problem, Frankle says, is dirty data. "Everybody has some data, and has an idea of what they want to do," but the lack of clean data makes it challenging to fine-tune a model to perform a specific task.

The Solution: Test-time Adaptive Optimization (TAO)

Databricks’ model offers a rare look at some of the key tricks that engineers are using to improve the abilities of advanced AI models, especially when good data is hard to come by. The method leverages ideas that have helped produce advanced reasoning models by combining reinforcement learning, a way for AI models to improve through practice, with "synthetic," or AI-generated training data.

How it Works

The Databricks method exploits the fact that, given enough tries, even a weak model can score well on a given task or benchmark. Researchers call this method of boosting a model’s performance "best-of-N". Databricks trained a model to predict which best-of-N result human testers would prefer, based on examples. The Databricks reward model, or DBRM, can then be used to improve the performance of other models without the need for further labelled data.

The DBRM Process

DBRM is then used to select the best outputs from a given model. This creates synthetic training data for further fine-tuning the model so that it produces a better output first time. Databricks calls its new approach Test-time Adaptive Optimization (TAO). "This method we’re talking about uses some relatively lightweight reinforcement learning to basically bake the benefits of best-of-N into the model itself," Frankle says.

Scalability and Future Development

The research done by Databricks shows that the TAO method improves as it is scaled up to larger, more capable models. Reinforcement learning and synthetic data are already widely used but combining them in order to improve language models is a relatively new and technically challenging technique.

Conclusion

Databricks’ new approach to AI model development offers a promising solution to the problem of dirty data. By combining reinforcement learning with synthetic data, the company has developed a method that can boost the performance of AI models without the need for clean labelled data. This could have significant implications for businesses and organizations that struggle to fine-tune their AI models.

FAQs

Q: What is the main challenge in getting AI to work reliably?
A: The main challenge is dirty data, or the lack of clean data, which makes it difficult to fine-tune a model to perform a specific task.

Q: What is Databricks’ solution to this challenge?
A: Databricks’ solution is to combine reinforcement learning with synthetic data to boost the performance of AI models without the need for clean labelled data.

Q: How does the Databricks method work?
A: The method uses a model to predict which best-of-N result human testers would prefer, based on examples. The Databricks reward model, or DBRM, can then be used to improve the performance of other models without the need for further labelled data.

Q: How does the DBRM process work?
A: DBRM is used to select the best outputs from a given model, creating synthetic training data for further fine-tuning the model so that it produces a better output first time.