AI Daily: NVIDIA Open Sources Earth-2 Weather Model, OpenAI Hosts Developer Town Hall, ChatGPT Ad Prices Surpass Traditional TV

NVIDIA officially open sources the Earth-2 weather forecasting model, with institutions including Taiwan’s Central Weather Administration being among the first adopters. Meanwhile, OpenAI held a developer town hall, revealing new tools and the GPT-5 roadmap. On the other hand, ChatGPT’s ad pricing has leaked, with a CPM of up to $60 shocking the market. This article will analyze these three major AI stories for you.

The pace of the tech world is always breathtaking, especially when two giants, NVIDIA and OpenAI, make major moves almost simultaneously. Have you ever imagined that future weather forecasts could be accurate to your doorstep without waiting hours for supercomputer calculations? Or, have you wondered what commercial value lies behind ChatGPT’s powerful conversational abilities?

Today, we’re discussing three major AI events that just happened. First is the bombshell from NVIDIA: they have fully open-sourced the powerful Earth-2 weather model, which is not only a boon for scientists but also closely related to our daily lives. Next is OpenAI’s just-concluded developer town hall and the heavy-hitting news it brought. Finally, there’s the jaw-dropping ad pricing of ChatGPT.

Let’s break down the implications behind these stories one by one.

NVIDIA Earth-2 Fully Open Sourced: A Revolution in Democratizing Weather Forecasting

In the past, when weather forecasting was mentioned, people usually thought of expensive supercomputers and complex physical equations. This has always been a high-barrier field. But NVIDIA has just done something quite remarkable: they announced the launch of the NVIDIA Earth-2 Open Model Family. What does this mean? Simply put, it places top-tier AI weather forecasting technology into the hands of researchers and developers worldwide.

This tool is not just a simple model, but a complete suite of accelerated climate AI software stacks. It covers everything from processing initial observational data to generating global forecasts for up to 15 days, and even local storm predictions. What’s most exciting is that NVIDIA has already put these resources on NVIDIA Earth-2 Studio and Hugging Face, where anyone can download and study them.

Why Is This More Powerful Than Traditional Methods?

Traditional weather forecasting relies on physical simulations, which are computationally intensive and time-consuming. NVIDIA’s new method uses AI for inference. Take the Earth-2 Medium Range (Atlas) model as an example: it uses a completely new architecture capable of predicting over 70 weather variables for the next 15 days, including temperature, pressure, wind speed, and humidity. In standard benchmarks, its performance has already surpassed the leading open models currently on the market.

Another highlight is Earth-2 Nowcasting (StormScope). This is a model specifically designed for “nowcasting.” It uses generative AI to convert satellite and radar data into high-resolution local storm forecasts. This is crucial for disaster prevention, providing warnings for dangerous weather in minutes, far faster than traditional physical models.

Taiwan’s Central Weather Administration Among First Adopters

This technology is not far from us. In fact, the Central Weather Administration (CWA) is already one of the partners for this technology. In addition to CWA, the Israel Meteorological Service and the US National Weather Service are also evaluating or using these models. This shows that Earth-2 is not just a laboratory product but a tool with combat-ready capabilities. Through these open-source models, meteorological units can provide more accurate forecasts at a lower cost and faster speed, directly benefiting the general public.

This is not just a display of technology, but the establishment of an ecosystem. When developers can easily access these tools, we may soon see various innovative applications for agriculture, renewable energy, or extreme climate defense.

OpenAI Developer Town Hall Concludes: GPT-5 Roadmap and “Sign in with ChatGPT” Revealed

Just as NVIDIA’s news shook the market, OpenAI CEO Sam Altman just finished an online Town Hall for “AI builders.” This event, defined as “experimental,” was not a traditional product launch but a high-density two-way dialogue. In the meeting, Sam Altman responded positively to developers’ concerns about the model roadmap, the future of Agents, and new features.

Admitting Current Models are “Spiky”, GPT-5 Will Pursue Balance

The most attention-grabbing focus of the meeting was the discussion on the next generation of models. Sam Altman admitted that current model development has become “Spiky”—the pursuit of powerful reasoning and coding capabilities has led to a decline in writing and user experience. He promised that the future GPT-5 version will be dedicated to correcting this issue, making all capability metrics equally outstanding. In addition, he threw out a concrete vision: the goal is to provide models with GPT-5.2 level intelligence by the end of 2027, while reducing costs by 100 times.

“Sign in with ChatGPT” Coming Soon

For developers, another piece of good news is that OpenAI confirmed it is developing a “Sign in with ChatGPT” feature. This means that in the future, users can directly log in to third-party applications with their ChatGPT accounts, token budgets, and even “long-term memory”. This will greatly lower the barrier for AI applications, allowing developers to focus more on building features rather than repeatedly training user preferences.

The Future of Software Engineers: From “Writing Code” to “Conducting”

Regarding the anxiety about whether AI will replace engineers, Sam Altman cited the “Jevons paradox” to explain. He believes that when AI makes coding cheaper and faster, the market demand for software will actually skyrocket. The role of future engineers will undergo a huge transformation, shifting from “spending time typing code” to “conducting computers to create value.” OpenAI itself is also adjusting its recruitment strategy, and future interviews will value candidates who “know how to use AI tools to complete work that used to take two weeks in a very short time.”

Although this event did not have flashy sound and light effects, it pointed out the direction for developers in the coming years: adapt to high-intensity AI collaboration and get ready for an era of smarter, lower-cost models.

ChatGPT Ad Pricing Exposed: An Expensive Battle for Attention

The last piece of news is about “money,” and big money at that. According to a report by The Information and Peter Gostev’s summary, ChatGPT’s advertising pricing strategy has surfaced, and the price is surprisingly high.

Current information shows that ChatGPT’s CPM (Cost Per Mille) pricing is as high as $60. What does this mean? Let’s compare it with other digital advertising rates:

• ChatGPT: $60
• Broadcast TV: Approximately $43.5
• Streaming/CTV: Approximately $27.3
• Social Platforms (e.g., Meta, TikTok): Usually between $5 and $10

Why Does ChatGPT Dare to Ask for a Sky-High Price?

You might ask, why is advertising on ChatGPT so expensive? The logic behind this lies in “Intent.” When users talk to ChatGPT, they usually come with specific questions or needs. This high interactivity and precise context understanding allow ad delivery to achieve unprecedented precision.

Compared to passively receiving information when scrolling through Instagram or TikTok, ChatGPT users are actively seeking answers. For brands, the potential conversion rate of intervening when users are “looking for answers” is huge. Although the price of $60 is far higher than TV and streaming media, it also reflects the market’s high expectations for “conversational search ads.” Whether this will change the landscape of digital advertising is worth our continued observation.

Frequently Asked Questions (FAQ)

Q1: Can the general public download NVIDIA Earth-2 models? Yes. NVIDIA has announced Earth-2 as a fully open model family. These models (including Earth-2 Medium Range and Nowcasting) are publicly available via NVIDIA Earth-2 Studio, Hugging Face, and GitHub. While these are tools designed primarily for scientists, developers, and meteorological agencies, any individual with the relevant technical skills can indeed access and study these resources.

Q2: Did OpenAI reveal news about GPT-5 at this event? Yes, Sam Altman revealed important clues. He admitted that current models (like GPT-4.5/o1) perform “Spikily,” sacrificing writing experience for reasoning and coding capabilities. The future GPT-5 series will strive to balance all capabilities (writing, coding, reasoning). In addition, he mentioned a specific goal: hoping to provide models with GPT-5.2 level intelligence by the end of 2027, with costs reduced by 100 times.

Q3: Why is ChatGPT’s ad pricing more expensive than TV? What is the specific form? According to data from The Information, ChatGPT’s ad CPM (Cost Per Mille) is as high as $60, which is indeed far higher than broadcast TV (about $43.5) and streaming media (about $27.3). Although current sources have not disclosed the specific ad presentation format (such as whether it is conversational placement), such high pricing reflects the market’s view of it as an extremely high-value digital advertising channel.

Q4: How much faster is NVIDIA Earth-2 specifically? Depending on the model, the acceleration effect is very significant. For example, the Earth-2 CorrDiff model is 500 times faster than traditional methods when downscaling predictions to high-resolution regional weather. Another model, Earth-2 FourCastNet3, generates forecasts 60 times faster than traditional methods. Earth-2 Nowcasting, aimed at immediate forecasting, can even generate high-resolution storm predictions in “minutes.”

Q5: What role does the Central Weather Administration (CWA) play in this project? The Central Weather Administration (CWA) is listed by NVIDIA as one of the partners for Earth-2, alongside agencies like the US National Weather Service (NWS) and the Israel Meteorological Service. These agencies use Earth-2 models to forecast weather and gain actionable insights. For example, the Israel Meteorological Service uses this technology to cope with extreme weather and significantly reduce computing costs, showing that this technology already has practical value for disaster prevention and forecasting applications.

The evolution of technology is always interconnected. NVIDIA provides powerful infrastructure and models to make predicting the Earth easier; OpenAI continues to explore new frontiers of human-machine interaction and attempts to build a new developer ecosystem; and ChatGPT’s commercialization attempts mark a clear value for this AI revolution. These developments may seem independent, but they jointly depict a smarter, more real-time future picture.