The Download: attempting to track AI, and the next generation of nuclear power
The Most Misunderstood Graph in AI
Every time OpenAI, Google, or Anthropic drops a new frontier large language model, the AI community holds its breath. It doesn't exhale until METR, an AI research nonprofit whose name stands for "Model Evaluation & Threat Research," updates a now-iconic graph that has played a major role in the AI discourse since it was first released in March of last year.
The graph suggests that certain AI capabilities are developing at an exponential rate, and more recent model releases have outperformed that already impressive trend. That was certainly the case for Claude Opus 4.5, the latest version of Anthropic's most powerful model, which was released in late November. In December, METR announced that Opus 4.5 appeared to be capable of independently completing a task that would have taken a human about five hours—a vast improvement over what even the exponential trend would have predicted.
But the truth is more complicated than those dramatic responses would suggest. While the graph has undoubtedly captured the imagination of the AI community, it has also been subject to misinterpretation and oversimplification. In this article, we'll delve into the nuances of the graph and explore what it really means for the future of AI.
Why the Graph Matters
The graph in question is a plot of the performance of various AI models over time, with the x-axis representing the model's performance and the y-axis representing the model's age. The graph shows a clear exponential trend, with each new model release outperforming the previous one by a significant margin.
At first glance, this might seem like a straightforward indication of progress in AI research. However, there are several reasons why the graph is more complex than it initially appears.
The Problem with Exponential Growth
One of the key issues with the graph is that it assumes a linear relationship between model performance and age. In other words, it assumes that each new model release will improve performance by a fixed amount, regardless of the model's age.
However, this is not necessarily the case. As models become more complex and sophisticated, the rate of improvement may slow down or even reverse. This is known as the "law of diminishing returns," and it can have significant implications for the future of AI research.
The Role of Data and Resources
Another important factor to consider is the role of data and resources in AI research. As models become more complex, they require increasingly large amounts of data and computational resources to train and run.
This can create a bottleneck in AI research, as the availability of data and resources may limit the rate of progress. In other words, even if a new model release is theoretically possible, it may not be feasible to train and run it due to the lack of resources.
The Future of AI Research
So what does the graph really mean for the future of AI research? While it's difficult to predict exactly what will happen, there are several possible scenarios.
One possibility is that the exponential trend will continue, with each new model release outperforming the previous one by a significant margin. This would suggest that AI research is making rapid progress, and that we can expect significant improvements in the near future.
However, this scenario is unlikely to occur in the absence of significant advances in data and resources. As models become more complex, they will require increasingly large amounts of data and computational resources to train and run.
Another possibility is that the law of diminishing returns will set in, and that the rate of improvement will slow down or even reverse. This would suggest that AI research is reaching a plateau, and that further progress will be difficult to achieve.
Conclusion
In conclusion, the graph in question is a complex and multifaceted representation of the progress of AI research. While it's difficult to predict exactly what will happen, there are several possible scenarios.
One thing is certain, however: the future of AI research will be shaped by a combination of technological, economic, and societal factors. As we move forward, it's essential to consider the implications of these factors and to develop strategies that will help us navigate the challenges and opportunities that lie ahead.
The Next Generation of Nuclear Power
Nuclear power continues to be one of the hottest topics in energy today, and in our recent online Roundtables discussion about next-generation nuclear power, hyperscale AI data centers, and the grid, we got dozens of great audience questions.
These ran the gamut, and while we answered quite a few (and I'm keeping some in mind for future reporting), there were a bunch we couldn't get to, at least not in the depth I would have liked. So let's answer a few of your questions about advanced nuclear power.
What is Next-Generation Nuclear Power?
Next-generation nuclear power refers to the development of new nuclear reactors and technologies that are designed to be safer, more efficient, and more cost-effective than traditional nuclear power plants.
These new reactors and technologies are being developed in response to the challenges facing traditional nuclear power, including high construction costs, long construction times, and concerns about safety and waste disposal.
What are the Key Features of Next-Generation Nuclear Power?
Next-generation nuclear power plants are designed to have several key features that differentiate them from traditional nuclear power plants.
These features include:
- Advanced reactor designs that are more efficient and safer than traditional reactors
- Improved fuel cycles that reduce waste production and improve fuel utilization
- Enhanced safety features that reduce the risk of accidents and improve public acceptance
- Increased efficiency and reduced costs through the use of advanced materials and technologies
What are the Benefits of Next-Generation Nuclear Power?
Next-generation nuclear power has several benefits that make it an attractive option for the future of energy production.
These benefits include:
- Reduced greenhouse gas emissions and other pollutants
- Improved energy security and reduced dependence on fossil fuels
- Increased energy efficiency and reduced costs
- Improved safety and reduced risk of accidents
What are the Challenges Facing Next-Generation Nuclear Power?
Despite the benefits of next-generation nuclear power, there are several challenges that must be addressed in order to make it a reality.
These challenges include:
- High upfront costs and long construction times
- Concerns about safety and waste disposal
- Public acceptance and regulatory hurdles
- Competition from other forms of energy production
Conclusion
In conclusion, next-generation nuclear power is a promising option for the future of energy production. While there are several challenges that must be addressed, the benefits of next-generation nuclear power make it an attractive option for reducing greenhouse gas emissions and improving energy security.
The Must-Reads
Here are some of the most interesting and important stories about technology that we've come across recently:
- Anthropic's new coding tools are rattling the markets. Fields as diverse as publishing and coding to law and advertising are paying attention.
- This Apple setting prevented the FBI from accessing a reporter's iPhone. Lockdown Mode has proved remarkably effective—for now.
- Last month's data center outage disrupted all TikTok categories. Not just the political content that some users claimed.
- Big Tech is pouring billions into AI in India. A newly-announced 20-year tax break should help to speed things along.
- YouTubers are harassing women using body cams. They're abusing freedom of information laws to humiliate their targets.
- Jokers have created a working version of Jeffrey Epstein's inbox. Complete with notable starred threads.
- What's the last thing you see before you die? A new model might help to explain near-death experiences—but not all researchers are on board.
- A new app is essentially TikTok for vibe-coded apps. Words which would have made no sense 15 years ago.
- Rogue TV boxes are all the rage. Viewers are sick of the soaring prices of streaming services, and are embracing less legal means of watching their favorite shows.
- Climate change is threatening the future of the Winter Olympics. Artificial snow is one (short term) solution.
Quote of the Day
"We've heard from many who want nothing to do with AI."
—Ajit Varma, head of Mozilla's web browser Firefox, explains why the company is reversing its previous decision to transform Firefox into an "AI browser," PC Gamer reports.
One More Thing
A major AI training data set contains millions of examples of personal data. Millions of images of passports, credit cards, birth certificates, and other documents containing personally identifiable information are likely included in one of the biggest open-source AI training sets, new research has found.
We Can Still Have Nice Things
A place for comfort, fun and distraction to brighten up your day. (Got any ideas? Drop me a line or skeet 'em at me.)
- If you're crazy enough to be training for a marathon right now, here's how to beat boredom on those long, long runs.
- Mark Cohen's intimate street photography is a fascinating window into humanity.
- A seriously dedicated gamer has spent days painstakingly recreating a Fallout vault inside the Sims 4.
- Here's what music's most stylish men are wearing right now—from leather pants to khaki parkas.
Deep Dive
The Download
The Download: China's dying EV batteries, and why AI doomers are doubling down
Plus: TikTok has finally signed a deal to keep operating in the US
By Rhiannon Williamsarchive page
The Download: sodium-ion batteries and China's bright tech future
Plus: This company is developing gene therapies for muscle growth, erectile dysfunction, and "radical longevity"
By Charlotte Jeearchive page
The Download: the worst technology of 2025, and Sam Altman's AI hype
Plus: China has built a major chip-making machine
By Rhiannon Williamsarchive page
The Download: cut through AI coding hype, and biotech trends to watch
Plus: read our predictions for the five hottest AI trends to watch
By Charlotte Jeearchive page
Stay connected
Illustration by Rose Wong
Get the latest updates fromMIT Technology Review
Discover special offers, top stories, upcoming events, and more.
Enter your email
Privacy PolicyThank you for submitting your email!
Explore more newsletters
It looks like something went wrong.
We’re having trouble saving your preferences.
Try refreshing this page and updating them one
more time. If you continue to get this message,
reach out to us at
[email protected] with a list of newsletters you’d like to receive.
