Ambuj Tewari, University of Michigan

State-of-the-art artificial intelligence systems like OpenAI’s ChatGPT, Google’s Gemini and Anthropic’s Claude have captured the public imagination by producing fluent text in multiple languages in response to user prompts. Those companies have also captured headlines with the huge sums they’ve invested to build ever more powerful models.

An AI startup from China, DeepSeek, has upset expectations about how much money is needed to build the latest and greatest AIs. In the process, they’ve cast doubt on the billions of dollars of investment by the big AI players.

I study machine learning. DeepSeek’s disruptive debut comes down not to any stunning technological breakthrough but to a time-honored practice: finding efficiencies. In a field that consumes vast computing resources, that has proved to be significant.

Where the costs are

Developing such powerful AI systems begins with building a large language model. A large language model predicts the next word given previous words. For example, if the beginning of a sentence is “The theory of relativity was discovered by Albert,” a large language model might predict that the next word is “Einstein.” Large language models are trained to become good at such predictions in a process called pretraining.

Pretraining requires a lot of data and computing power. The companies collect data by crawling the web and scanning books. Computing is usually powered by graphics processing units, or GPUs. Why graphics? It turns out that both computer graphics and the artificial neural networks that underlie large language models rely on the same area of mathematics known as linear algebra. Large language models internally store hundreds of billions of numbers called parameters or weights. It is these weights that are modified during pretraining. https://www.youtube.com/embed/MJQIQJYxey4?wmode=transparent&start=0 Large language models consume huge amounts of computing resources, which in turn means lots of energy.

Pretraining is, however, not enough to yield a consumer product like ChatGPT. A pretrained large language model is usually not good at following human instructions. It might also not be aligned with human preferences. For example, it might output harmful or abusive language, both of which are present in text on the web.

The pretrained model therefore usually goes through additional stages of training. One such stage is instruction tuning where the model is shown examples of human instructions and expected responses. After instruction tuning comes a stage called reinforcement learning from human feedback. In this stage, human annotators are shown multiple large language model responses to the same prompt. The annotators are then asked to point out which response they prefer.

Advertisement

It is easy to see how costs add up when building an AI model: hiring top-quality AI talent, building a data center with thousands of GPUs, collecting data for pretraining, and running pretraining on GPUs. Additionally, there are costs involved in data collection and computation in the instruction tuning and reinforcement learning from human feedback stages.

All included, costs for building a cutting edge AI model can soar up to US$100 million. GPU training is a significant component of the total cost.

The expenditure does not stop when the model is ready. When the model is deployed and responds to user prompts, it uses more computation known as test time or inference time compute. Test time compute also needs GPUs. In December 2024, OpenAI announced a new phenomenon they saw with their latest model o1: as test time compute increased, the model got better at logical reasoning tasks such as math olympiad and competitive coding problems.

Slimming down resource consumption

Thus it seemed that the path to building the best AI models in the world was to invest in more computation during both training and inference. But then DeepSeek entered the fray and bucked this trend.

Their V-series models, culminating in the V3 model, used a series of optimizations to make training cutting edge AI models significantly more economical. Their technical report states that it took them less than $6 million dollars to train V3. They admit that this cost does not include costs of hiring the team, doing the research, trying out various ideas and data collection. But $6 million is still an impressively small figure for training a model that rivals leading AI models developed with much higher costs.

The reduction in costs was not due to a single magic bullet. It was a combination of many smart engineering choices including using fewer bits to represent model weights, innovation in the neural network architecture, and reducing communication overhead as data is passed around between GPUs.

It is interesting to note that due to U.S. export restrictions on China, the DeepSeek team did not have access to high performance GPUs like the Nvidia H100. Instead they used Nvidia H800 GPUs, which Nvidia designed to be lower performance so that they comply with U.S. export restrictions. Working with this limitation seems to have unleashed even more ingenuity from the DeepSeek team.

Advertisement

DeepSeek also innovated to make inference cheaper, reducing the cost of running the model. Moreover, they released a model called R1 that is comparable to OpenAI’s o1 model on reasoning tasks.

They released all the model weights for V3 and R1 publicly. Anyone can download and further improve or customize their models. Furthermore, DeepSeek released their models under the permissive MIT license, which allows others to use the models for personal, academic or commercial purposes with minimal restrictions.

Resetting expectations

DeepSeek has fundamentally altered the landscape of large AI models. An open weights model trained economically is now on par with more expensive and closed models that require paid subscription plans.

The research community and the stock market will need some time to adjust to this new reality.

Ambuj Tewari, Professor of Statistics, University of Michigan

This article is republished from The Conversation under a Creative Commons license. Read the original article.

Advertisement

STM Daily News is a vibrant news blog dedicated to sharing the brighter side of human experiences. Emphasizing positive, uplifting stories, the site focuses on delivering inspiring, informative, and well-researched content. With a commitment to accurate, fair, and responsible journalism, STM Daily News aims to foster a community of readers passionate about positive change and engaged in meaningful conversations. Join the movement and explore stories that celebrate the positive impacts shaping our world.

https://stmdailynews.com/

Eric Smalley, The Conversation and Matt Williams, The Conversation

The U.S. Supreme Court on Jan. 17, 2025, upheld a law requiring TikTok’s China-based parent company, ByteDance, to sell the video app by Jan. 19, 2025, or face a nationwide ban on the app. In a unanimous decision, the court rejected TikTok’s claim that the law violates its First Amendment rights.

The court’s ruling is the latest development in a lengthy saga over the fate of an app that is widely popular, especially among young Americans, but that many politicians in Washington say is a security risk.

The ruling is unlikely to be the end of the story. President Joe Biden said that he will not enforce the law in the waning hours of his administration. President-elect Donald Trump said he will reverse the ban and is reportedly considering an executive order to do so.

But why is TikTok controversial? Are the claims of it being a national security risk valid? And what will the case mean for free speech? The Conversation’s contributors have been on hand to answer these questions.

1. An agent of the Chinese state?

Politicians who wanted to ban TikTok, or at least sever its links to China, fear that the app provides a way for the Chinese Communist Party to influence Americans or use their data for malicious purposes. But how much influence does the Chinese government have on TikTok? That question is addressed by Shaomin Li, a scholar of China’s political economy and business at Old Dominion University.

Li explains that the relationship between TikTok, ByteDance and the Chinese Communist Party is nuanced – it isn’t simply a matter of officials in Beijing telling ByteDance to jump and the parent company dictating how high its subsidiary will leap. Rather, as with all companies in China, employees are under certain obligations when it comes to advancing national interests. In China, private enterprises, such as ByteDance, operate as joint ventures with the state.

“Regardless of whether ByteDance has formal ties with the party, there will be the tacit understanding that the management is working for two bosses: the investors of the company and, more importantly, their political overseers that represent the party,” Li writes. “But most importantly, when the interests of the two bosses conflict, the party trumps.”

Advertisement

2. Exploiting user data

The risks TikTok poses to U.S. users are similar to the risks posed by many popular apps, principally that the app collects data about you. That data, which includes contact information and website tracking, plus all data you post and messages you send through the app, is available to use or misuse by ByteDance and any other entity that has or gains access to it.

Iowa State University cybersecurity researcher Doug Jacobson writes that U.S. officials and lawmakers are concerned that the Chinese government could exploit TikTok user data to spy on U.S. citizens. Government hackers could use the TikTok data to trick users into revealing more personal information.

But if the goal is to counter Chinese hackers, banning TikTok is likely to prove too little, too late. “By some estimates, the Chinese government has already collected personal information on at least 80% of the U.S. population via various means,” Jacobson writes. “The Chinese government – along with anyone else with money – also has access to the large market for personal data.”

3. The security risks of a ban

Banning TikTok could also make U.S. users more vulnerable to hackers of all stripes. Rochester Institute of Technology computer security expert Robert Olson writes that many of the 170 million U.S. TikTok users could try to get around a ban on the app, with negative consequences for their digital safety.

If TikTok ends up banned from Apple’s and Google’s app stores, users could try to access the app elsewhere via sideloading. This practice of getting around the Apple and Google app stores leaves users vulnerable to malware posing as the TikTok app. TikTok users might also be motivated to circumvent Apple and Google security controls in order to keep the app installed, a move that would make users’ phones more vulnerable.

“I find it unlikely that a TikTok ban (is) technologically enforceable,” Olson writes. “This … legislation – aimed at improving cybersecurity – could motivate users to engage in riskier digital behavior.”

4. First Amendment concerns

In its legal challenge to the U.S. government, ByteDance claimed the government is violating its First Amendment rights. Technology law scholars Anupam Chander of Georgetown University and Gautam Hans of Cornell University write that ByteDance had grounds for its claim, and that the implications go beyond this case.

Advertisement

TikTok is a publisher – an online publisher of users’ videos. Forcing ByteDance to divest TikTok is a form of prior restraint – the government preventing speech before it occurs, Chander and Hans write.

“By forcing the sale of TikTok to an entity without ties to the Chinese Communist Party, Congress’ intent with the law is to change the nature of the platform,” they write. “That kind of government action implicates the core concerns that the First Amendment was designed to protect against: government interference in the speech of private parties.”

5. What about the others?

Security and legal issues aside, the forced sale to a U.S.-based company or ban of TikTok in the United States is a questionable approach to solving the problems the law aims to address: potential Chinese government influence in the U.S., harm to teens and data privacy violations, writes Arizona State media scholar Sarah Florini.

The Chinese government – and other U.S. adversaries – have long used social media apps owned by U.S. companies to attempt to influence American public opinion. TikTok is hardly alone in posing harm to teens, as the Facebook whistleblower case amply demonstrated. And vast amounts of Americans’ personal data are already available to any buyer on the open and black markets.

“Concerns about TikTok are not unfounded, but they are also not unique. Each threat posed by TikTok has also been posed by U.S.-based social media for over a decade,” Florini writes.

This is an updated version of an article originally published on Sept. 16, 2024.

Eric Smalley, Science + Technology Editor, The Conversation and Matt Williams, Senior International Editor, The Conversation

Advertisement

This article is republished from The Conversation under a Creative Commons license. Read the original article.

On Thursday, a significant number of Capital One customers reported problems with their direct deposits, causing concern for many who rely on timely access to their funds. As of Thursday morning, more than 2,000 users had expressed frustration over the banking service’s failure to show their expected direct deposits, according to Down Detector, a reputable platform that tracks website and app outages.

By the afternoon, it appeared that some progress had been made, with reports of outages reducing to just over 1,700 by 4:10 p.m. Eastern Time. However, for those affected, the issue created a tense situation, as direct deposits are often critical for managing monthly bills and living expenses.

Real Concerns from Real Customers

Social media was abuzz with complaints, as many customers turned to platforms like X (formerly Twitter) to voice their frustrations. One user summed up the sentiment succinctly: “Some of us have bills to pay and groceries to buy. At least keep us in the loop.” This highlights not only the inconvenience of the technical issue but also the anxiety it causes for individuals who count on their bank to provide reliable services.

Capital One’s Response

In response to the outcry, Capital One acknowledged the issue and assured its customers that it was actively working to resolve the glitch. While specific details about the nature of the technical difficulties have not been disclosed, the bank has expressed its commitment to restoring normal operations as quickly as possible. As financial tech failures can prompt feelings of mistrust, transparency in communication is key during these situations.

What Should Affected Customers Do?

For Capital One customers experiencing issues with their direct deposits, here are a few steps to consider:

1. Check for Updates: Stay informed by checking Capital One’s official social media channels and website for updates regarding the technical issues.
2. Contact Customer Support: If you haven’t received your expected payment, contact Capital One’s customer service for clarification on your account status.
3. Monitor Your Account: Regularly check your bank account for updates on your direct deposit status and any other service announcements.
4. Plan for Contingencies: If you anticipate payments may be delayed, consider adjusting your budget temporarily to accommodate any unexpected expenses.

Conclusion

Banking issues like those experienced by Capital One customers on Thursday serve as a reminder of our reliance on financial institutions for our everyday needs. While banks strive to deliver seamless services, technical hiccups can happen. For now, it’s crucial for customers to remain patient and informed as Capital One works diligently to resolve the situation and restore access to their funds.

If you’ve been affected, share your experience and solutions. Community support during these times can make all the difference!

CBS News Reporting on the issue: https://www.cbsnews.com/news/capital-one-direct-deposit-issues/

STM Daily News is a vibrant news blog dedicated to sharing the brighter side of human experiences. Emphasizing positive, uplifting stories, the site focuses on delivering inspiring, informative, and well-researched content. With a commitment to accurate, fair, and responsible journalism, STM Daily News aims to foster a community of readers passionate about positive change and engaged in meaningful conversations. Join the movement and explore stories that celebrate the positive impacts shaping our world.

Advertisement

https://stmdailynews.com/