Tech
Why building big AIs costs billions – and how Chinese startup DeepSeek dramatically changed the calculus
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.
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.
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.
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.
STM Blog
From Hand Signals to Smart Crosswalks: The Evolution of the Modern Pedestrian Signal
Discover the history of the modern pedestrian signal, from Garrett A. Morgan’s groundbreaking traffic signal to today’s smart, accessible crosswalks.
Last Updated on July 12, 2026 by Daily News Staff
Every day, millions of people rely on pedestrian signals to cross busy street safely. A glowing white walking figure, an orange-red hand, and a countdown timer have become familiar sights around the world. While these signals may seem like simple pieces of infrastructure, they are the result of more than a century of innovation, engineering, and public safety improvements.
The modern pedestrian signal did not appear overnight. Instead, it evolved through the contributions of inventors, engineers, city planners, and transportation officials who continually refined traffic control systems as cities grew and automobiles became more common.
The Early Days of Traffic Control
Before electric traffic signals, intersections were controlled by police officers, railway-style semaphores, or even hand signals. As horse-drawn wagons gave way to automobiles in the early 1900s, traffic congestion and accidents increased dramatically, creating an urgent need for better traffic management.
One of the earliest electric traffic lights was installed in Cleveland, Ohio, in 1914. It used red and green lights and was manually operated. While it improved vehicle movement, pedestrians still had to judge for themselves when it was safe to cross.
Garrett A. Morgan’s Breakthrough
One of the most important milestones came in 1923 when inventor and entrepreneur Garrett Augustus Morgan received U.S. Patent No. 1,475,024 for an improved traffic signal.
Morgan’s design introduced a third position in addition to “Stop” and “Go.” This intermediate phase temporarily stopped traffic in every direction before allowing vehicles to proceed. The brief pause reduced confusion at intersections and provided additional time for pedestrians to cross safely.
Morgan reportedly developed his design after witnessing a serious traffic accident. His invention demonstrated how thoughtful engineering could improve public safety while making increasingly busy streets more efficient.
Although Morgan did not invent the illuminated “WALK” and “DON’T WALK” pedestrian signal used today, his three-position signal became a foundational step in the evolution of modern traffic control.
The Birth of Dedicated Pedestrian Signals
As cities expanded after World War II, pedestrian safety became an even greater concern. More people were walking in increasingly crowded downtown districts, and separating pedestrian movements from vehicle traffic became a priority.
During the early 1950s, several American cities began experimenting with dedicated pedestrian signals. New York City became one of the first major municipalities to install illuminated “WALK” and “DON’T WALK” signs at busy intersections.
These early systems gave pedestrians their own designated crossing phase, reducing conflicts with turning vehicles and improving safety at some of the nation’s busiest intersections.
Standardization Across America
By the 1960s and 1970s, traffic engineers recognized the importance of creating consistent traffic control devices nationwide.
The Manual on Uniform Traffic Control Devices (MUTCD) established national standards for traffic signs, pavement markings, and pedestrian signals. Standardized designs helped ensure that pedestrians could understand crossing signals regardless of where they traveled in the United States.
Eventually, words gave way to internationally recognized symbols—a walking person to indicate it was safe to cross and an upraised hand to indicate pedestrians should wait. These symbols transcended language barriers and improved accessibility for visitors and non-English speakers.
The Countdown Era
One of the most significant modern improvements arrived with pedestrian countdown timers.
Rather than simply flashing a warning, countdown displays show exactly how many seconds remain before the crossing phase ends. Research has shown that countdown timers help pedestrians make better crossing decisions and improve compliance with traffic signals.
Today, countdown timers have become standard equipment at intersections across much of the United States.
Accessibility Takes Center Stage
Modern pedestrian signals are designed to serve everyone.
Accessible Pedestrian Signals (APS) now provide audible tones, spoken messages, vibrating push buttons, and locator sounds that assist pedestrians who are blind or have low vision. These features allow more people to navigate intersections independently and safely.
The continued development of accessible technology reflects a broader commitment to making transportation systems inclusive for all users.
The Future of Pedestrian Safety
Pedestrian signals continue to evolve.
Many cities now use smart traffic systems that detect pedestrians waiting to cross, automatically adjust signal timing based on traffic conditions, and prioritize people walking during busy periods.
Researchers are exploring artificial intelligence, connected vehicle technology, and sensor-based systems capable of communicating directly with autonomous vehicles. Future pedestrian crossings may adapt in real time to weather conditions, crowd sizes, emergency vehicles, and even the needs of older adults or individuals with disabilities.
A Legacy Built by Many Innovators
The pedestrian signal we know today is the product of more than a century of collaboration and innovation.
Early traffic engineers created the first electric traffic lights. Garrett A. Morgan improved intersection safety with his groundbreaking three-position traffic signal. Transportation agencies standardized traffic control devices, while engineers continued refining pedestrian technology through countdown timers, accessible features, and intelligent traffic systems.
Every safe crossing today reflects the work of countless inventors, planners, researchers, and public officials dedicated to protecting lives.
As cities continue to grow and transportation technology advances, the humble pedestrian signal remains one of the most effective—and often overlooked—public safety innovations ever developed.
At STM Daily News, we celebrate the inventors, engineers, and visionaries whose everyday innovations quietly improve life for millions of people. Sometimes the most important inventions aren’t the ones that grab headlines—they’re the ones we depend on every single day without giving them a second thought.
Related Reading
- Federal Highway Administration – Manual on Uniform Traffic Control Devices (MUTCD)
- National Museum of African American History and Culture – Garrett Augustus Morgan
- United States Patent and Trademark Office
- Federal Highway Administration – Accessible Pedestrian Signals
- National Highway Traffic Safety Administration (NHTSA)
🧠 Discover the remarkable innovators, inventors, and trailblazers who helped shape our world but rarely receive the recognition they deserve. Share your thoughts in the comments and subscribe to the STM Daily News newsletter to catch every new Forgotten Genius Friday feature and more inspiring stories delivered to your inbox.
Rotem Rozental, University of Southern California
Film photography is experiencing a resurrection, summoned by unlikely conjurers: Gen Z.
It wasn’t too long ago that analog photography – which uses photographic film and chemical processing – was declared all but dead, relegated to the province of niche hobbyists and professional artists.
Digital cameras had taken over nearly all areas of photographic production. Film industry titans like Polaroid and Kodak had shrunk dramatically from their heyday, becoming shells of their former selves. Darkrooms, where students learned how to manually develop and print film, shuttered at high schools and college campuses across the country, replaced by digital labs. For most people, the spirit of analog photography was mainly channeled through Instagram filters.
But within the past five years, younger people have been increasingly drawn to the old way of doing photography.
In 2025, 35% of the 42 million active film camera users worldwide were reported to be between the ages of 18 and 30. The year prior, online searches for analog photography saw a 41% rise.
Disposable camera sales have been steadily increasing since 2023. The photography journal PetaPixel went a step further and announced 2024 as “film’s best year in decades,” as major brands have introduced new cameras in response to renewed demand and revived classic models. More than 30% of respondents to a 2024 Ilford Photo survey on film photography were in the 25-34 age group.
As I’ve witnessed more and more of my undergraduate art and design students embrace analog photography, I’m not seeing this as a trend rooted in a nostalgic yearning for the past. Instead, I’m seeing it as young people rejecting algorithms, breaking free from the alienation of social media and reacting to childhoods spent on Zoom and TikTok – a deliberate move to redefine the future of art, social connection and engagement with the world.
Pining for a ‘third place’
In my work as a historian of photography and lecturer at the University of Southern California, I’ll often ask my students about how they take photos – whether they’re using digital cameras their smartphones or analog devices.
This year, for the first time, some of my students discussed images they’d printed and the physical photography albums they’d put together of their friends and family. They talked about how they’d also been sending postcards, writing letters and tacking photographs to their bedroom walls.
I couldn’t help but think about how so much of the language tied to early social media seemed to refashion physical gestures for a virtual world – “posting” on a “wall,” “poking,” “tagging” and “bookmarking,” not to mention “friending.”
This was a rhetorical move by social media companies, likely designed to help people feel as though they were in a familiar terrain of social connection. Yet the underlying business model of these platforms depended more on maximizing engagement and advertising revenue than on nurturing authentic relationships.
Everyone knows what happened next: The more connected young people became online, the more isolated and detached they started to feel. The COVID-19 lockdown pushed social life online even further, and researchers are only now starting to see how the combination of increased screen time and isolation negatively affected adolescents’ mental health. By 2023, 51% of American teenagers reported they spend at least four hours a day on social media.
I see the attraction of analog photography as a response to life lived through screens, a pathway toward community engagement and the desire for what sociologists call “a third place.”
Coined by sociologist Ray Oldenburg in his 1989 book “The Great Good Place,” third places are meant as a space separate from home and work. They offer a reprieve for the in-between, generating the conditions needed for creative cross-pollination. They might include a local cafe, a neighborhood writing group, a weekly Magic: The Gathering game or a college fraternity – any space that allows for social interaction and personal growth.
These spaces also combat loneliness. They get people out of their heads and into a community. Oldenburg also referred to them as “havens of sociability,” places or gatherings where people can arrive alone to join others, and the atmosphere is “democratic and festive.”
Analog communities IRL
In April 2026, the inaugural AnalogCon took place in Los Angeles. Organized by the Los Angeles Center of Photography, where I serve as executive director and chief curator, it was a festival for all things analog photography. It didn’t just serve as a third place for photography enthusiasts; it also showed how analog photography – as a practice, ritual and community – is flourishing.
Vendors, industry leaders, artists and teachers participated in the two-day event, which included exhibitions, panels, demonstrations and guided photography tours around Little Tokyo. The excitement and thirst for similar events was palpable.
Photography now joins a broader trend of a generational preoccupation with physical cultural objects and media. Although music streaming represents 82% of revenues generated in the music industry, vinyl records sales have been rising for over a decade, crossing the US$1 billion threshold in the U.S. in 2025.
Nearly 60% of Gen Z are now purchasing records. VHS tapes and VCR players are also making a strange comeback, with stores like Be Kind Video and Videotheque in California offering VHS, DVDs and Blu-ray rentals.
But beyond that, record stores and video rental shops have become third places in their own right. There’s a big difference between selecting a film to stream from your bed and getting out of the house, going to a store and talking about movies with a clerk and fellow film enthusiasts.
Think about the sound a tape cassette makes when you open and close it, or the vibrant graphics on the covers of DVDs or VHS tapes. Think about rewinding or making a mixtape for your recent crush. These are objects of belonging that signal specific cultural moments, rituals and aesthetics, and many young people today are starting to experience them for the first time.
Now, think about gently inserting a roll of film into a camera. Think about choosing an angle carefully when snapping a photo, because the number of frames is limited and you want to make them count. Think about the thrill of discovery when the pictures finally emerge as objects on paper.
To me, these are more than fleeting trends. They signal a push against a digital culture that is designed to cultivate envy and reward outrage, insults and humiliation.
Instead, armed with rolls of film, more and more Gen Zers appear to be opting out of their algorithmic feeds in favor of experiencing life in ways that feel more deliberate, personal and tangible.
Rotem Rozental, Lecturer in Critical Studies, Roski School of Art and Design, University of Southern California
This article is republished from The Conversation under a Creative Commons license. Read the original article.
📰 Enjoying STM Daily News? Join the conversation!
💬 Leave a comment, share your thoughts, and subscribe to our newsletter for the latest stories, updates, and “News You Can Use This Moment!” delivered to your inbox.
Stay connected with STM Daily News!
How to Protect Yourself from a Smartphone Scam
(Feature Impact) The first sign is unexpectedly losing access to your cell phone. Soon after, when you connect to Wi-Fi, the gravity of the situation sinks in: a criminal has gained access to your cell phone number and is trying to siphon money from your credit cards and bank accounts.
The scam is called SIM swapping, or SIM hijacking, and it’s a concern for law enforcement in the United States and abroad as more than 5,000 people have reported SIM swapping scams to the FBI since 2022. Older adults, caregivers and families can benefit from understanding the warning signs of SIM swapping and taking simple security steps to prevent it from happening.
How SIM swapping works
A SIM card, or its digital version known as an eSIM, helps connect a phone number to a carrier network. In a SIM swapping scam, a criminal collects basic information about their victim, such as their name, birthdate and address, to try to move the victim’s phone number to a SIM card or eSIM profile the criminal controls.
Once complete, the scammer gains access to accounts you may be logged into on your phone, such as bank accounts or credit card apps, without touching your phone or being near you.
How to protect yourself from SIM swapping scams
Preparation is the best protection against SIM swapping. Cell phone users should use strong, unique passwords for each online account – password managers are a helpful tool in creating complex and randomized passwords. Use two-factor authentication where it’s offered; this adds an extra layer of security when accessing sensitive accounts.
Next, consumers should protect personal information they share online, whether on social media or in texts or emails asking for identifying data, such as PIN numbers, birthdates or one-time security codes. Be wary of anyone pushing you to share personal information, particularly if they’re pushy with their request or make it sound urgent.
Check your mobile carrier to see if it offers SIM protection. For example, Verizon customers can toggle on a protection feature on the carrier’s website or app to lock lines on their account to help prevent SIM changes.
If you get an unprompted notification that your SIM has been changed, or otherwise suspect you’ve been targeted in a SIM swapping scam, contact your banks immediately and have them freeze your accounts, including ones the criminals may not have targeted yet. Next, work with your cell phone provider to help regain access to your mobile device. If you’re able, share as much information as possible with law enforcement so they can investigate, or at least document trends, in how often this scam occurs.
To find more advice to protect against smartphone scams, visit Verizon.com.
Photo courtesy of Shutterstock
SOURCE:
📰 Enjoying STM Daily News? Join the conversation!
💬 Leave a comment, share your thoughts, and subscribe to our newsletter for the latest stories, updates, and “News You Can Use This Moment!” delivered to your inbox.
Stay connected with STM Daily News!
