Connect with us

The Conversation

Why building big AIs costs billions – and how Chinese startup DeepSeek dramatically changed the calculus

Published

on

theconversation.com – Ambuj Tewari, Professor of Statistics, University of Michigan – 2025-01-29 08:08:00

Why building big AIs costs billions – and how Chinese startup DeepSeek dramatically changed the calculus

DeepSeek burst on the scene – and may be bursting some bubbles.
AP Photo/Andy Wong

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.

YouTube video
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.

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.

YouTube video
DeepSeek sent shockwaves through the tech financial ecosystem.

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.

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.The Conversation

Ambuj Tewari, Professor of Statistics, University of Michigan

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

Read More

The post Why building big AIs costs billions – and how Chinese startup DeepSeek dramatically changed the calculus appeared first on theconversation.com

The Conversation

Who’s who at the Vatican?

Published

on

theconversation.com – Daniel Speed Thompson, Associate Professor of Religious Studies, University of Dayton – 2025-03-03 07:18:00

Who’s who at the Vatican?

Deacons take part in a mass in St. Peter’s Basilica that was supposed to be presided over by Pope Francis.
AP Photo/Alessandra Tarantino

Daniel Speed Thompson, University of Dayton

For more than two weeks, eyes have been on the Vatican, awaiting news about Pope Francis’ health. The pope has been at Rome’s Gemelli Hospital since Feb. 14, 2025, being treated for double pneumonia and other complications.

When a pope is ill, resigns or passes away, who steps in? And who else helps lead the Holy See? The Conversation U.S. asked Daniel Speed Thompson, a theologian at the University of Dayton, for some insight into Vatican City.

Who are the most powerful people at the Vatican, besides the pope?

The Vatican houses the central government of the Catholic Church and is also an independent city-state. The pope is both the head of the Catholic Church and head of state.

In order to govern both, he has the Roman Curia, meaning “court.” In modern terms, the Curia is the papal bureaucracy. It is an extension of the pope’s authority.

In Catholic doctrine, the pope has the highest authority in the church. He can exercise it alone or with the College of Bishops, made up of all the bishops in the world. Bishops named by the pope to the office of “cardinal” can, if under 80 years old, vote to elect a new pope. Some cardinals, but by no means all, serve in the papal Curia in Rome.

Besides the pope, curial officials who oversee important aspects of the church’s political and religious life are often powerful figures. For example, the secretariat of state, headed by Cardinal Pietro Parolin, oversees relations with other countries and international organizations. It also oversees the Vatican’s diplomatic corps.

Two men in black robes with red skullcaps and red sashes walk on a paved road, flanking a man in white robes.
Pope Francis smiles as he walks alongside Vatican Secretary of State Pietro Parolin, left, and Cardinal Giuseppe Versaldi at the Vatican in 2014.
AP Photo/Gregorio Borgia

The Dicastery – “department” – for the Doctrine of the Faith, led by Cardinal Víctor Manuel Fernández, addresses questions about correct Catholic teaching on faith and morals. The Dicastery of Bishops, headed by Cardinal Robert Prevost, coordinates the nominations of new bishops around the world.

All these officials work under the authority of the pope, advocating for and implementing his agenda. For example, Prevost has suggested that all Catholics should be involved in the selection of bishops. This idea is linked with Francis’ call for a more “synodal” church: one that is less hierarchical and shaped by lay Catholics’ concerns and challenges.

If a pope can’t fulfill his duties, who steps in?

When a pope dies – or resigns, like Benedict XVI did in 2013 – the governance of the Catholic Church formally falls to the College of Cardinals. However, the authority of the college is very limited. On their own, cardinals cannot make any significant decisions concerning faith, morals and worship. Nor can they undo previous papal decisions or change church laws about electing a new pope.

All the heads of the dicasteries lose their office upon the death or resignation of a pope. The College of Cardinals serves as a caretaker government whose primary purpose is to prepare for the election of the new pope and oversee day-to-day workings of the Vatican.

One cardinal, known as the “camerlengo,” is responsible for confirming the pope’s death or resignation. He then assumes control over the pope’s residence and coordinates the funeral, if needed. The camerlengo also takes custody of the Vatican’s property in Rome and supervises details for the upcoming conclave.

A man wearing a priest's collar gestures as he speaks, sitting in front of a framed portrait of Pope Francis.
Cardinal Camerlengo Kevin Farrell talks with The Associated Press in his office in Rome in 2018.
AP Photo/Paolo Santalucia

The day-to-day business of the Catholic Church continues, but no big decisions can be made in the absence of a pope. The church cannot appoint new bishops, and the Vatican cannot start new diplomatic efforts.

Are officials at the Vatican often nominated to be pope?

Sometimes. Francis was a cardinal from Argentina before his election as pope and had not served in the Roman Curia. However, Benedict XVI, Francis’ predecessor, did serve as the prefect of the Congregation – now called Dicastery – for the Doctrine of the Faith. Some recent popes served in the Curia earlier in their career but not immediately before their election.

What do you wish more people understood about the Vatican?

Three things. First, the Vatican is unlike any organization in the world. Its religious mission and political status rest on nearly 2,000 years of history. This complicated story provides a unique tradition that anchors the institution of the Catholic Church, but can also block the church from critical self-examination and renewal.

Second, the Vatican is like every organization in the world. Vatican officials can be faithful to the highest standards of their religion, truly wishing to serve the church and the common good of humanity. But they can also be flagrantly immoral, even criminals, and careerist seekers of status or luxury. Francis has consistently called out priests and bishops who see themselves as somehow superior by virtue of their office or their ordination.

Finally, compared with the massive bureaucracies of modern governments and corporations, the Vatican is relatively small and not as wealthy as it is often portrayed.

Although the Curia manages a vast international organization, its resources are far closer to my own midsize Catholic university than to the U.S. government or Apple. Vatican City and the Holy See employ about 2,000 people, with an operating budget of about US$835 million.

Yes, the Catholic Church has wealth – and the ongoing problem of deficits and financial corruption. But the Vatican’s resources pale in comparison with what a modern state or large company can muster.The Conversation

Daniel Speed Thompson, Associate Professor of Religious Studies, University of Dayton

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

Read More

The post Who’s who at the Vatican? appeared first on theconversation.com

Continue Reading

The Conversation

Coastal economies rely on NOAA, from Maine to Florida, Texas and Alaska – even if they don’t realize it

Published

on

theconversation.com – Christine Keiner, Chair, Department of Science, Technology, and Society, Rochester Institute of Technology – 2025-02-28 07:46:00

Coastal economies rely on NOAA, from Maine to Florida, Texas and Alaska – even if they don’t realize it

U.S. fishing industries, both commercial and recreational, rely on healthy coastal areas.
Wolfgang Kaehler/LightRocket via Getty Images

Christine Keiner, Rochester Institute of Technology

Healthy coastal ecosystems play crucial roles in the U.S. economy, from supporting multibillion-dollar fisheries and tourism industries to protecting coastlines from storms.

They’re also difficult to manage, requiring specialized knowledge and technology.

That’s why the National Oceanic and Atmospheric Administration – the federal agency best known for collecting and analyzing the data that make weather forecasts and warnings possible – leads most of the government’s work on ocean and coastal health, as well as research into the growing risks posed by climate change.

The government estimates that NOAA’s projects and services support more than one-third of the nation’s gross domestic product. Yet, this is one of the agencies that the Trump administration has targeted, with discussions of trying to privatize NOAA’s forecasting operations and disband its crucial climate change research.

As a marine environmental historian who studies relationships among scientists, fishermen and environmentalists, I have seen how NOAA’s work affects American livelihoods, coastal health and the U.S. economy.

Here are a few examples from just NOAA’s coastal work, and what it means to fishing industries and coastal states.

Preventing fisheries from collapsing

One of the oldest divisions within NOAA is the National Marine Fisheries Service, known as NOAA Fisheries. It dates to 1871, when Congress created the U.S. Commission of Fish and Fisheries. At that time, the first generation of conservationists started to worry that America’s natural resources were finite.

By conducting surveys and interviewing fishermen and seafood dealers, the fish commissioners discovered that freshwater and saltwater fisheries across the country were declining.

YouTube video
Looking back on 150 years of NOAA’s fisheries history.

Oil spills and raw sewage were polluting waterways. Fishermen were using high-tech gear, such as pound nets, to catch more and more of the most valuable fish. In some areas, overfishing was putting the future of the fisheries in jeopardy.

One solution was to promote aquaculture, also known as fish or shellfish farming. Scientists and entrepreneurs reared baby fish in hatcheries and transferred them to rivers, lakes or bays. The Fish Commission even used refrigerated railroad cars to ship fish eggs across the country.

Today, U.S. aquaculture is a US$1.5 billion industry and the world’s fastest-growing food sector. Much of the salmon you see in grocery stores started as farm-raised hatchlings. NOAA provides training, grants and regional data to support the industry.

Men stand in line with pails to deposit them in a train car on a siding.
Men carry pails of fish specimens to a U.S. Fish Commission ‘fish car’ – a train car designed specifically for transporting fish or fish eggs to stock U.S. rivers, lakes and coastal waters – in this historical photo.
Smithsonian Institution Archives

NOAA Fisheries also helps to regulate commercial and recreational fishing to keep fish populations healthy and prevent them from crashing.

The 1976 Magnuson-Stevens Fishery Conservation and Management Act and other laws implemented catch limits to prevent overfishing. To develop fair regulations and combat illegal practices, NOAA and its predecessors have worked with fishing organizations through regional fishery management councils for decades.

These industries generate $321 billion in sales and support 2.3 million jobs.

Restoring coral reefs to help marine life thrive

NOAA also benefits U.S. coastal communities by restoring coral reefs.

Corals build up reefs over centuries, creating “cities of the sea.” When they’re healthy, they provide nurseries that protect valuable fish species, like snapper, from predators. Reefs also attract tourism and protect coastlines by breaking up waves that cause storm-driven flooding and erosion.

The corals of Hawaii, Florida, Puerto Rico and other tropical areas provide over $3 billion a year in benefits – from sustaining marine ecosystems to recreation, including sport fishing.

However, reefs are vulnerable to pollution, acidification, heat stress and other damage. Warming water can cause coral bleaching events, as the world saw in 2023 and 2024.

NOAA monitors reef health. It also works with innovative restoration strategies, such as breeding strains of coral that resist bleaching, so reefs have a better chance of surviving as the planet warms.

Battling invasive species in the Great Lakes

A third important aspect of NOAA’s coastal work involves controlling invasive species in America’s waters, including those that have menaced the Great Lakes.

Zebra and quagga mussels, spiny water flea and dozens of other Eurasian organisms colonized the Great Lakes starting in the late 1900s after arriving in ballast water from transoceanic ships. These invaders have disrupted the Great Lakes food web and clogged cities’ water intake systems, causing at least $138 million in damage per year.

Hoses on a boat covered in zebra mussels.
Zebra mussels found attached to this boat at an inspection station in Oregon show how easily invasive species can be moved. The boat had come from Texas and was on its way to Canada.
Oregon Department of Fish & Wildlife, CC BY-SA

In the Northwest Atlantic, Caribbean and Gulf of Mexico, invasive lionfish, native to Asia and Australia, have spread, preying on native fish essential to coral reefs. Lionfish have become one of the world’s most damaging marine fish invasions.

NOAA works with the Coast Guard, U.S. Geological Survey and other organizations to prevent the spread of invasive aquatic species. Stronger ballast water regulations developed through the agency’s research have helped prevent new invasions in the Great Lakes.

Understanding climate change

One of NOAA’s most crucial roles is its leadership in global research into understanding the causes and effects of climate change.

The oil industry has known for decades that greenhouse gases released into the atmosphere from burning fossil fuels would raise global temperatures.

Evidence and research from around the world have connected greenhouse gas emissions from human activities to climate change. The data have shown how rising temperatures have increased risks for coastal areas, including worsening heat waves and ocean acidification that harm marine life; raising sea levels, which threaten coastal communities with tidal flooding and higher storm surges; and contributing to more extreme storms.

NOAA conducts U.S. climate research and coordinates international climate research efforts, as well as producing the data and analysis for weather forecasting that coastal states rely on.

Why tear apart an irreplaceable resource?

When Republican President Richard Nixon proposed consolidating several different agencies into NOAA in 1970, he told Congress that doing so would promote “better protection of life and property from natural hazards,” “better understanding of the total environment” and “exploration and development leading to the intelligent use of our marine resources.”

The Trump administration is instead discussing tearing down NOAA. The administration has been erasing mentions of climate change from government research, websites and policies – despite the rising risks to communities across the nation. The next federal budget is likely to slash NOAA’s funding.

Commercial meteorologists argue that much of NOAA’s weather data and forecasting, also crucial to coastal areas, couldn’t be duplicated by the private sector.

As NOAA marks its 55th year, I believe it’s in the nation’s and the U.S. economy’s best interest to strengthen rather than dismantle this vital agency.The Conversation

Christine Keiner, Chair, Department of Science, Technology, and Society, Rochester Institute of Technology

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

Read More

The post Coastal economies rely on NOAA, from Maine to Florida, Texas and Alaska – even if they don’t realize it appeared first on theconversation.com

Continue Reading

The Conversation

As flu cases break records this year, vaccine rates are declining, particularly for children and 65+ adults

Published

on

theconversation.com – Annette Regan, Adjunct Associate Professor of Epidemiology, University of California, Los Angeles – 2025-02-28 07:46:00

As flu cases break records this year, vaccine rates are declining, particularly for children and 65+ adults

It’s not too late to get a flu shot.
Fat Camera/E+ via Getty Images

Annette Regan, University of California, Los Angeles

In February 2025, flu rates spiked to the highest levels seen in at least 15 years, with flu outpacing COVID-19 infections and hospitalizations for the first time since the beginning of the COVID-19 pandemic. The Centers for Disease Control and Prevention has classified this flu season as having “high” severity across the U.S.

The Conversation asked epidemiologist Annette Regan to explain why this flu season is different from last year’s and what people can do to help reduce the spread.

How do flu cases and hospitalizations this year compare with previous years?

Beginning in late January and extending through February 2025, flu hospitalizations have been higher than any other week since before 2009.

Most flu cases appear to be from influenza A strains, with a split between influenza A/H3N2 and influenza A/H1N1. These are two different subtypes of the influenza A virus.

Researchers believe that historically seasons that are predominated by influenza A/H3N2 infections tend to be more severe, but infections from influenza A/H1N1 can still be very severe.

This year’s season is also peaking “late” compared with the past three flu seasons, which peaked in early or late December.

Unfortunately, there have been a number of deaths from flu too this season. Since Jan. 1, 2025, alone, over 4,000 people, including 68 children, have died from flu. While the number of deaths do not mark a record number, it shows that flu can be a serious illness, even in children.

YouTube video
Unless directed otherwise, everyone ages 6 months and older should get a flu shot.

Why are flu cases so high this year?

There are a number of factors behind any severe season, including poor community protection from low immunization rates and low natural immunity, virus characteristics, vaccine effectiveness and increased human contact via travel, office work or schools.

Unfortunately, flu vaccination rates have declined since the COVID-19 pandemic. At the end of the 2023-24 flu season, 9.2 million fewer doses were administered in pharmacies and doctors’ offices compared with an average year before the pandemic.

In addition, since 2022, fewer and fewer doses of flu vaccine have been distributed by private manufacturers. Flu vaccination rates for adults have historically been in the 30% to 60% range, much lower than the recommended 70%. Before the COVID-19 pandemic, flu vaccination rates were increasing by around 1% to 2% every year.

Flu vaccination rates began dropping after the COVID-19 pandemic, especially in higher-risk groups. Flu vaccination in children has dropped from 59% in 2019-20 to 46% in 2024-25. In adults 65 years and older, the group with the greatest risk of hospitalization and death, flu vaccination rates dropped from 52% in 2019-20 to 43% in 2024-25.

Lower vaccination rates mean a greater portion of the population is not protected by vaccines. Data shows that vaccination reduces the risk of flu hospitalization. Even if a vaccinated person gets infected, they may be less likely to experience severe illness. As a result, low vaccination rates could contribute to higher flu severity this season.

However, low vaccination rates are probably not the only reason for the high rates of flu this season. In previous severe seasons, genetic changes to the viruses have made them better at infecting people and more likely to cause severe illness.

The effectiveness of annual flu vaccines varies depending on how well the vaccine matches the circulating virus. The effectiveness of vaccines ranges from 19% to 60% in any given season. In the 2023-24 flu season, the vaccine was 42% effective.

Similarly, early 2024-25 data from the U.S. shows that the vaccine was 41% to 55% effective against flu hospitalizations in adults and 63% to 78% effective against flu hospitalizations in children.

YouTube video
Something as simple as regular handwashing could keep you from getting the flu.

How do seasonal flu symptoms differ from COVID-19 and other illnesses?

It’s important to remember that people often incorrectly refer to “the flu” when they have a common cold. Flu is caused only by the influenza virus, which tends to be more severe than common colds and more commonly causes a fever.

Many of the signs and symptoms for flu, COVID-19 and other respiratory viruses are the same and can range from mild coldlike symptoms to pneumonia and respiratory distress. Common flu symptoms are fever, cough and fatigue, and may also include shortness of breath, a sore throat, nasal congestion, muscle aches and headache.

Some symptoms, such as changes in or loss of taste and smell, are more common for COVID-19. For both COVID-19 and flu, the symptoms do not start until about one to four days after infection, and symptoms seem to last longer for COVID-19.

The only way to know what virus is causing an infection is to test. This can be done using a rapid test, some of which now test for flu and COVID-19 together, or by seeing a doctor and getting tested using a nasal swab. There are prescription antiviral medications available to treat flu and COVID-19, but they need to be taken near the time that symptoms start.

Some people are at high risk of severe flu and COVID-19, such as those who are immunosuppressed, have diabetes or have chronic heart or lung conditions. In these cases, it is important to seek early care and treatment from a health care professional. Some doctors will also prescribe via telehealth calls, which can help reduce the strain on doctors’ offices, urgent care centers and emergency rooms when infection rates are high.

What can people do now to help steer clear of the flu?

There are a number of ways people can reduce their risk of getting or spreading flu. Since the flu season is still underway, it’s not too late to get a flu vaccine. Even in seasons when the vaccine’s effectiveness is low, it is likely to offer better protection compared with remaining unvaccinated.

Handwashing and disinfecting high-traffic surfaces can help reduce contact with the flu virus. Taking efforts to avoid contact with sick people can also help, including wearing a mask when in health care facilities.

Finally, remember to take care of yourself. Exercising, eating healthy and getting sufficient sleep all help support a healthy immune system, which can help reduce chances of infection.

Those who have been diagnosed with flu or are experiencing flu-like symptoms should avoid contact with other people, especially in crowded spaces. Covering coughs and sneezes can help reduce the amount of virus that is spread.The Conversation

Annette Regan, Adjunct Associate Professor of Epidemiology, University of California, Los Angeles

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

Read More

The post As flu cases break records this year, vaccine rates are declining, particularly for children and 65+ adults appeared first on theconversation.com

Continue Reading

Trending