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/

Transit Equity Day

Los Angeles – Mark your calendars! On February 4, 2025, Southern Californians will have the unique opportunity to ride Metrolink free of charge, as the region’s six-county passenger rail provider celebrates Transit Equity Day. This annual event honors the legacy of civil rights pioneer Rosa Parks and underscores the essential role public transportation plays in fostering access and equity for all communities.

On this special day, riders can board at any of Metrolink’s 67 stations without needing a ticket and can travel systemwide, including the Arrow service. To further celebrate Transit Equity Day, numerous local transit carriers will also waive fares, providing additional options for those seeking to explore the region.

A Statement of Purpose

Metrolink Board Chair and Orange Orange County Fourth District Supervisor Doug Chaffee expressed the significance of Transit Equity Day, stating, “This day honors Rosa Parks’ enduring legacy, while highlighting the vital role public transportation plays in ensuring people from all communities have access.” He continued, “When Metrolink introduced the largest service expansion in our history last fall, it was to create a more equitable system that works for all Southern Californians. Offering free rides on Transit Equity Day helps fulfill that promise.”

Exciting Changes in Service

In October 2023, Metrolink increased weekday service by nearly 23%, adding more midday and late-evening options for those needing to travel outside traditional work hours. Thanks to this expansion and recent improvements, Metrolink has seen a surge in ridership, setting multiple post-pandemic, single-day records. Moreover, a new San Bernardino Line schedule implemented on January 27 aims to enhance on-time performance along this high-volume corridor, allowing riders to experience efficient travel firsthand on Transit Equity Day.

A Commitment to Accessibility

Metrolink is not just about free rides on special occasions. The agency is dedicated to making public transit accessible to everyone year-round. In addition to offering three free-ride days each year, Metrolink provides everyday discounts for children, seniors, riders with disabilities, low-income riders, and active military members. For students, the Student Adventure Pass program allows K-12, college, and trade school students to ride free with a valid student ID, an initiative extended through June 30, 2025.

Metrolink ticketholders also enjoy free transfers to many connecting local transit providers, further enhancing accessibility across the region.

Advertisement

Explore the Region

If you’re looking to discover new places, Metrolink has curated a special repository of destinations that are easily accessible from its stations. Check out these exciting travel options at metrolinktrains.com/explore and start planning your adventure on Transit Equity Day!

Join Metrolink, LA Metro, Omnitrans, the Orange County Transportation Authority, the Riverside County Transportation Commission, the San Bernardino County Transportation Authority, the Ventura County Transportation Commission, and other agencies in taking part in this celebratory day of equitable access to public transportation.

Ride Free, Celebrate Equity!

Transit Equity Day is more than just a free ride; it’s a reminder of the importance of public transport in promoting equity, access, and community connection. We invite all Southern Californians to take advantage of this opportunity, explore the region, and honor the legacy of Rosa Parks. This February 4th, hop aboard Metrolink and experience the freedom of transit—at no cost!

Metrolink website: Travelers looking for inspiration can view Metrolink’s specially curated repository of destinations easily accessible from Metrolink stations at metrolinktrains.com/explore.

https://stmdailynews.com/groundbreaking-for-a-sustainable-future-la-metros-southeast-gateway-line-light-rail-project/

Advertisement

Paul F. Clark, Penn State

Whole Foods workers at the Philadelphia flagship store in the city’s Art Museum area voted to unionize on Jan. 27, 2025. They are the first store in the Amazon-owned grocery chain to do so.

Paul Clark, a professor of labor and employment relations at Penn State University, talked to Kate Kilpatrick, The Conversation U.S. Philadelphia editor, about why this is happening – and why in Philly.

The Whole Foods workers in Philadelphia voted 130-100 in favor of unionizing. What do we know about their grievances?

From what I understand, these workers have felt that compensation, benefits and work conditions were not what they should be. Some are long-standing employees and say they struggle to afford their basic necessities.

Why did the union drive effort succeed now, and in Philly?

In the last five years, there has been a surge in union organizing. There are a number of reasons for this. First is the labor market. Low unemployment emboldens workers to take the risk of organizing a union. If workers feel their employer can’t replace them or that they can easily get a similar job, they are less fearful of angering the employer by trying to organize.

The second reason is that the Biden administration was a labor-friendly administration – perhaps the most in history. The U.S. president appoints a majority of members to the National Labor Relations Board, which interprets and enforces the labor law that governs organizing. Under Biden, the NLRB regularly issued decisions that provided greater protection to workers and held employers accountable when they violated workers’ rights. During Republican administrations, the board’s decisions are generally pro-business and provide less protection to workers. So workers had the wind at their back in that regard.

Advertisement

Also recent polling shows that 70% of Americans approve of unions, compared with less than half of Americans just 15 years ago. The generally favorable view of unions creates a more supportive environment for organizing.

And the last factor is that Generation Z, the youngest group of workers, clearly wants more out of their work and employment than previous generations. So we see a lot of young workers across the country organizing at Starbucks, Trader Joe’s, Apple and now at Whole Foods and other stores.

Why Philadelphia? Philadelphia is a relatively strong union town. The percentage of the workforce that is represented by a union is higher in Philadelphia than in most cities and areas of the country. So when workers express interest in organizing in Philadelphia they get a lot of support. Other unions might turn out members for their rallies, pressure the company to not oppose the organizing drive and offer other aid and assistance.

The starting wage at the Philadelphia Whole Foods store is US$16 an hour. Is that considered low when the city’s minimum wage is just $7.25 an hour?

The minimum wage in Philadelphia is $7.25 because that is the federal minimum wage. States can institute a higher minimum wage if they choose to, but Pennsylvania is one of the few Northeast states that hasn’t adopted a minimum wage higher than the federal minimum. The minimum wages in New Jersey, New York and Massachusetts, for example, are $15 or above.

But the minimum wage in Pennsylvania is almost irrelevant because of today’s labor market. Unemployment is low, and many employers have to offer significantly more than the minimum wage to get workers.

And the minimum wage is supposed to be a starting wage for workers with little experience or seniority. What workers want is a living wage. According to the MIT Living Wage Calculator, a single person in Philadelphia needs to earn around $24 per hour to cover the basic costs of living. And Whole Foods is a profitable business. It’s part of Amazon, one of the most profitable, largest companies in the world. I think workers at these companies believe that they play an important role in generating those profits because of the work they do. And they think they should get a fair share of those profits.

Advertisement

How might the Whole Foods workers expect the company to fight back?

When employees win an organizing election as the Whole Food workers have, they have won a battle but not the war. The purpose of forming a union is to improve wages and benefits and working conditions, and you do that by negotiating a contract with the company. That is the next step in the process. But the law only requires employers to bargain with employees – to meet at reasonable times and exchange proposals. It doesn’t compel them to agree to anything.

The typical strategy of companies that aggressively oppose their workers having a union is to drag their feet in bargaining and not sign a contract. That is technically illegal, but labor law in the U.S. is relatively weak, and with good legal advice you can drag out bargaining for a very long time.

We’ve seen this with the Starbucks campaign. The first Starbucks store unionized in 2021. Over 540 stores have organized since then. And Starbucks workers at those stores still do not have a contract.

Could the new Trump administration have any impact on how this plays out in Philly?

The fact that the Trump administration has taken over gives companies more confidence that the standard delay strategy will work.

On Jan. 28, 2025, President Donald Trump fired Jennifer Abruzzo, the general counsel of the NLRB. The general counsel is the official at the board who basically enforces the National Labor Relations Act. Abruzzo was very aggressive in holding employers accountable if they violated the act and in protecting the rights of workers who tried to organize.

Advertisement

Trump’s approach to labor law in his first four years in office was at the other extreme. He appointed as general counsel Peter Robb, who was seen as far less aggressive in protecting workers’ rights and his interpretations of the law were much more pro-business.

Under the Biden administration, if a company was coming to the bargaining table month after month and not agreeing to anything, the NLRB would eventually step in and cite the employer for not bargaining in good faith. The NLRB could find the employer guilty of unfair labor practices and genuinely put pressure on it to bargain a contract.

Based on the board’s actions during the first Trump administration, the board in the next few years will be more likely to allow companies to delay and delay in reaching a contract.

What leverage do the Whole Foods employees have?

They can go on strike. But Amazon has the resources to put up with a strike at one Whole Foods store forever.

Other Whole Foods stores may be considering union drives. The more stores that organize, the more momentum the Philadelphia store will have. But for now, these workers in Philly are going to have their work cut out for them.

That said, they won’t be alone. The Whole Foods workers organized with the UFCW Local 1776, which is basically a statewide union that’s been around for decades. It has a lot of resources and experienced and knowledgeable leaders, plus the resources of the national UFCW. So it’s going to lean into this fight, and these workers will also have a lot of support from the rest of the labor community in Philadelphia.

Advertisement

Earlier this month, three Congressional representatives from Pennsylvania wrote a letter to Jason Buechel, the Whole Foods CEO, and to Jeff Bezos, the Amazon founder, that expressed their concerns about efforts to suppress the union drive. Is that support typical?

It’s not unusual. But there is no legal basis for elected officials to intervene in a labor-management dispute. I’d put that under the heading of community support.

You have a lot of progressive elected officials in Philadelphia who are supportive of unions, and that’s true in Pennsylvania right up to the governor.

Paul F. Clark, Professor of Labor and Employment Relations, Penn State

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

Advertisement