Every year, as the amount of data we create grows and software becomes increasingly more complex, it is more crucial to improve the efficiency of computer systems. However, in this complex ecosystem, dependability, such as ensuring software contains fewer bugs and achieves greater security, is often considered an afterthought, said Allen School professor Baris Kasikci. Software and hardware have been plagued by bugs that can lead to data loss, security vulnerabilities and costly critical infrastructure failures.
In his research, Kasikci focuses on developing techniques for building systems that are both efficient and have a strong foundation of dependability, with an emphasis on real-world technical and societal impact. At the 55th Annual IEEE/IFIP International Conference on Dependable Systems and Networks (DSN 2025) in June, Kasikci was recognized with the Rising Star in Dependability Award for “his impressive track-record and contributions in the field of dependable systems, including multiple publications in highly regarded venues, and influence on current industry practice.”
“Building systems that are simultaneously efficient and dependable is challenging because there is a strong tension between techniques aimed to achieve these properties,” said Kasikci. “This tension is due to the different trade-offs involved in achieving efficiency and dependability (e.g., performance optimizations versus defenses against vulnerabilities that cause slowdown). To rise to this challenge, my work draws insights from a broad set of disciplines such as systems, computer architecture, programming languages, machine learning and security.”
Kasikci’s work helps improve society’s trust in computer systems with new secure hardware systems and bug detection techniques. He was part of the team that discovered Foreshadow, a speculative attack on Intel processors which enables an attacker to steal sensitive information that is stored inside personal computers and third-party clouds. Their work has influenced the redesign of Intel processors and motivated all major cloud vendors to deploy mitigations against the vulnerability. He and his collaborators also developed REPT, which is one of the most widely-deployed failure analysis systems in the world and is in use across all modern Microsoft Windows platforms. It has allowed Microsoft engineers to tackle bugs that have been open for many years, and the techniques behind the system have been adopted by both Intel and Facebook.
He is also interested in developing methods for optimizing the entire computing stack, from hardware to operating systems. At the hardware level, Kasikci helped introduce techniques to improve code, such as Ripple, a software-only technique that uses program context to inform the placement of cache eviction instructions. More recently, he developed new methods for making systems that support large language models more efficient including NanoFlow, a novel LLM-serving framework with close to optimal throughput.
In future work, Kasikci is interested in advancing reliability techniques such as debugging and testing by making production execution data such as logs and core dumps more useful. For example, he envisions using this information to quickly find bugs that production systems may suffer from, while also managing how much storage and resources these logs can consume.
Networks have become one of the foundational pillars for modern society. When they go down, we cannot fly airplanes, operate bank accounts or even scroll through social media. One of the long-standing challenges in networking, however, is the difficulty of accurately predicting how changes to device configurations will impact real-world traffic flows — especially when a single code line revision can easily break a network, explained Allen School professor and alum Ratul Mahajan (Ph.D., ‘05).
To help network engineers and operators ensure that their networks operate exactly as they intend, Mahajan and his collaborators introduced Batfish, an open source network configuration analysis tool that can find errors and ensure the accuracy of planned network configurations, helping to prevent costly outages. At the ACM Special Interest Group on Data Communication (SIGCOMM) conference last month, Batfish was recognized with the SIGCOMM Networking Systems Award for its “significant impact on the world of computer networking.”
“Over the years, networks have become super complicated and it has gotten to the point where humans cannot assure that changes they make to the network are not going to bring it down,” said Mahajan, who is one of the co-directors for the UW Center for the Future of Cloud Infrastructure (FOCI). “With Batfish, we focus on what we call proactive validation. Instead of, after the fact, discovering that something bad has happened to the network, Batfish takes your planned change to the network and makes a model of how the network will behave if this change were to go through.”
The platform was first developed in 2015 by Mahajan and colleagues at Microsoft Research; University of California, Los Angeles; and University of Southern California. It was later commercialized by Intentionet, where Mahajan was the CEO and co-founder. Today, Batfish is managed by Amazon Web Services (AWS), and more than 75 companies rely on the tool to help design and test their networks.
Batfish uses an offline “snapshot” of the network to build a model and infer if there are any issues present within the configuration. The platform takes in device configurations from various vendors including Cisco, Juniper and Arista, and it then converts these configurations into a unified and vendor-independent model of the network. Once the model is built, engineers can query Batfish about topics such as the reachability between various network parts, potential routing loops, access control list (ACL) configurations such as incorrectly assigned permissions, or other policy and security constraints. Batfish then provides specific information needed to find and fix the misconfigurations.
While Batfish’s main architecture and original goal has stood the test of time, many of its underlying techniques have been revamped and enhanced to tackle scalability and usability challenges that complex, real-world networks face. For example, for each violated property, Batfish originally only provided one counterexample packet that was randomly picked by the SMT solver from violating headspace, however, these counterexamples lacked context and could be confusing. To help engineers understand what went wrong, Batfish now provides a positive example, or a packet that does not violate the property, alongside a counterexample that engineers can compare to pinpoint the issue.
As one of the earliest and most widely-adopted network verification platforms, it has helped shape key areas of research such as control-plane and data-plane modeling and network automation. From tech giants to small startups, multiple organizations rely on Batfish every day to both validate network configurations and drive innovations in network designs and operations.
“The main lasting impact of Batfish, beyond the code itself, would be changing the practice of networking to use these types of tools,” Mahajan said. “It was one of the first pieces of technology that made automated reasoning for networks a reality.”
Akari Asai (Ph.D., ‘25), research scientist at the Allen Institute for AI (Ai2) and incoming faculty at Carnegie Mellon University, is interested in tackling one of the core challenges with today’s large language models (LLMs). Despite their increasing potential and popularity, LLMs can often get facts wrong or even combine tidbits of information into a nonsensical response, also known as hallucinations. This can be especially concerning when these LLMs are used for scientific literature or software development, where accuracy is vital.
For Asai, the solution is developing retrieval-augmented language models, a new class of LLMs that pull relevant information from an external datastore using a query that the LLM generates. Her research has helped establish the foundations for retrieval-augmented generation (RAG) and showcase its effectiveness at reducing hallucinations. Since then, she has gone on to add adaptive and self-improvement capabilities and apply these innovations to practical applications such as multilingual natural language processing (NLP).
Asai was recently named one of MIT Technology Review’s Innovators Under 35 2025 for her pioneering research improving artificial intelligence. The TR35 award recognizes scientists and entrepreneurs from around the world who “stood out for their early accomplishments and the ways they’re using their skills and expertise to tackle important problems.”
“With the rapid adoption of LLMs, the need to investigate their limitations, develop more powerful models and apply them in safety-critical domains has never been more urgent,” said Asai.
Traditional LLMs generate responses to user inputs based solely on their training data. In comparison, RAG enhances the LLM with an additional information retrieval component that utilizes the user input to first pull information from a new, external datastore. This allows the model to generate responses that incorporate up-to-date information without needing additional training data. By checking this datastore, an LLM can better detect when it is generating a falsehood, which it can then verify and correct using the retrieved information.
Asai took that research a step further and she and her collaborators introduced Self-reflective RAG, or Self-RAG, that improves LLMs’ quality and factual accuracy with retrieval and self-reflection. With Self-RAG, a model uses reflection tokens to decide when to retrieve relevant external information and critique the quality of its own generations. While RAG can only retrieve relevant information a fixed number of time steps, Self-RAG can retrieve multiple times — making it useful for diverse downstream queries including instruction following.
She is interested in utilizing these retrieval-augmented language models to solve real-world problems. In 2024, Asai introduced OpenScholar, a new model that can help scientists more effectively and efficiently navigate and synthesize scientific literature. She has also investigated how retrieval-augmented language models can be useful for code generation, and helped develop frameworks that can improve information access across multiple languages such as AfriQA, the first cross-lingual question answering dataset focused on African languages.
“Akari is among the pioneers in advancing retrieval-augmented language models, introducing several paradigm shifts in this area of research,” said Allen School professor Hannaneh Hajishirzi, Asai’s Ph.D. advisor and also senior director at Ai2. “Akari’s work not only provides a foundational framework but also highlights practical applications, particularly in synthesizing scientific literature.”
This award comes on the heels of another MIT Technology Review recognition. Last year, Asai was named one of the publication’s Innovators Under 35 Japan. She has also received an IBM Ph.D. Fellowship and was selected as one of this year’s Forbes 30 Under 30 Asia in the Healthcare and Science category.
At the 63rd Annual Meeting of the Association for Computational Linguistics (ACL 2025), Allen School researchers brought home multiple awards for their work that is moving natural language processing research forward. Their projects ranged from laying the foundation for how artificial intelligence systems understand and follow human instructions to exploring how large language models (LLMs) pull responses from their training data — and more.
TACL Test of Time Award: Weakly Supervised Learning of Semantic Parsers for Mapping Instructions to Actions
From robotics to voice assistants, today’s AI systems rely on their ability to understand human language and interact naturally with it.
Luke Zettlemoyer
In their 2013 paper “Weakly Supervised Learning of Semantic Parsers for Mapping Instructions to Actions,” Allen School professor Luke Zettlemoyer and then student Yoav Artzi (Ph.D., ‘15), now a professor at Cornell Tech, set the groundwork for this capability with an approach for teaching models to follow human instructions without the need for detailed manual explanations. For their lasting contributions to the field, the researchers received the Test of Time Award as part of the inaugural Transactions of the Association for Computational Linguistics (TACL) Paper Awards presented at ACL 2025.
“This was the first paper in a line of work on learning semantic parsers from easily gathered interactions with an external world, instead of requiring supervised training data,” Zettlemoyer said. “Although the form of the models has changed significantly over time, this style of learning is still relevant today as an early precursor to techniques such as Reinforcement Learning with Verifiable Rewards (RLVR).”
Zettlemoyer and Artzi developed the first comprehensive model that tackles common issues that arise with learning and understanding unrestricted natural language. For example, if you told a robot to follow the navigation instructions “move forward twice to the chair and at the corner, turn left to face the blue hall,” it would need to solve multiple subproblems to interpret the instructions correctly. It would have to resolve references to specific objects in the environment such as “the chair,” clarify words based on context, and also understand implicit requests like “at the corner” that provide goals without specific steps.
To address these challenges without the need for intensive engineering effort, the duo developed a grounded learning approach that can jointly reason about meaning and context, and then continue to learn from their interplay. It uses Combinatory Categorial Grammar (CCG), which is a framework that assigns words to different syntactic categories such as noun or prepositional phrase, efficiently parsing complex instructional language for meaning by mapping them into logical expressions. Then, the weighted CCG ranks possible meanings for each instruction.
This joint model of meaning and context allows for the system to continue to learn from situated cues, such as the visible objects in the environment. For example, in the earlier set of navigation instructions, “the chair” can refer to multiple different objects such as chairs, barstools or even recliners. While the CCG framework would include a lexical item for each meaning of “the chair,” the execution of the task might fail depending on what objects are in the world. It allows for the system to learn by watching examples play out, and then following if the actions lead to successful outcomes such as completing the task or reaching a destination.
The researchers tested their method using a benchmark navigational instructions dataset and found that their joint approach successfully completed 60% more instruction sets compared to the previous state-of-the-art methods.
Outstanding Paper Award: Byte Latent Transformer: Patches Scale Better Than Tokens
Allen School Ph.D. students Artidoro Pagnoni and Margaret Li earned an ACL Outstanding Paper Award for research done at Meta with their advisor Zettlemoyer, who is also the senior research director at Meta FAIR.
Alongside their collaborators, they introduced the Byte Latent Transformer (BLT), a new byte-level LLM architecture that is the first to be able to match the more standard tokenization-based LLM performance at scale. At the same time, BLT is also able to improve efficiency and increase robustness to noisy data.
Many existing LLMs are trained using tokenization, where raw text is broken down into more manageable tokens which then serve as the model’s vocabulary. This process was essential because training LLMs directly with bytes was cost prohibitive at scale. However, these tokens can influence how string is compressed and lead to issues such as domain sensitivity.
Instead, BLT groups bytes into dynamically-sized patches, serving as the primary units of computation. These patches are then segmented based on the entropy of the next byte, allowing the system to allocate more model capacity where needed. For example, higher entropy indicates a more complex sequence which can then prompt a new, shorter patch. In the first Floating-Point Operations (FLOP) controlled scaling study of byte-level models, the team found that BLT’s performance was on par or superior to models such as Llama 3. With its efficiency and adaptability, the researchers position BLT as a promising alternative to the traditional token-based models available.
As LLMs become increasingly popular in higher-stake scenarios, it is important to understand why they generate certain responses and where they get their answers from. Fully open language models such as OLMo have been trained on trillions of tokens that everyone can access, but current behavior tracing methods are not scaled to work within this multi-trillion-token setting.
To address this, a team of researchers at the Allen School and the Allen Institute for AI (Ai2) introduced OLMoTrace, the first system that allows users in real time to explore how LLM outputs connect back to their training data. For the research’s innovation and practical application, the team received the ACL 2025 Best Demo Paper Award.
Jiacheng Liu
“Today’s large language models are so complex that we barely understand anything about how they generate the responses we see,” said lead author and Allen School Ph.D. student Jiacheng Liu. “OLMoTrace is powered by a technology I previously developed at UW, ‘infini-gram,’ with numerous system optimizations enabling us to deliver instant insights to how LLMs likely have learned certain phrases and sentences.”
The OLMoTrace inference pipeline works by scanning the LLM’s output and identifying long, unique and relevant text spans that appear verbatim in the model’s training data. For each span, the system retrieves up to 10 snippets from the training data that contain the span, prioritizing the most relevant documents. Finally, the system does some post-processing on the spans and document snippets, and presents them to the user through the chat interface. OLMoTrace is publicly available in the Ai2 model playground with the OLMo 2 family of models.
The researchers proposed multiple practical applications for OLMoTrace. For example, if the model generates a fact, users can look back to the training data to fact check the statement. It can also reveal the potential source of seemingly creative and novel LLM-generated expressions. In addition, OLMoTrace can help debug erratic LLM behaviors, such as hallucinations or incorrect self-knowledge, which are crucial to address as LLMs become increasingly more commonplace, Liu explained.
ACL Dissertation Award: Rethinking Data Use in Large Language Models
For her Allen School Ph.D. dissertation titled “Rethinking Data Use in Large Language Models,” Sewon Min, now faculty at University of California, Berkeley, received the inaugural ACL Dissertation Award. In her work, Min tackled fundamental issues that current language models face, such as factuality and privacy, by introducing a new class of language models and alternative approaches for training such models.
This new class of models, called nonparametric language models, is able to identify and reason with relevant text from its datastore during inference. Compared to conventional models that have to remember every applicable detail from their training set, models with a datastore available at inference time have the potential to be more efficient and flexible.
Nonparametric language models can also help address the legal constraints that traditional models often face. Language models are commonly trained using all available online data, which can lead to concerns with copyright infringement and crediting data creators. Min developed a new approach where language models are trained solely on public domain data. Copyrighted or other high-risk data is then kept in a datastore that the model can only access during inference and which can be modified at any time.
Each year, Qualcomm recognizes exceptional Ph.D. students whose research proposals promote the company’s core values of innovation, execution and teamwork with the Qualcomm Innovation Fellowship. With the goal of enabling “students to pursue their futuristic innovative ideas,” the winning teams receive a one-year fellowship as well as mentorship from Qualcomm engineers to help their projects succeed.
Two of this year’s winning teams from North America feature Allen School students. Andrew Alex and Megan Frisella received support for their proposed project “Productive Programming of Multi-Threaded Hardware Accelerators.” Fellow Allen School Ph.D. student Zixian Ma and Yushi Hu, a Ph.D. student in the University of Washington’s Department of Electrical & Computer Engineering (ECE), also earned a fellowship for their project “Learning Multi-modal Agents to Reason and Act for Complex Multi-modal Tasks.”
Productive Programming of Multi-Threaded Hardware Accelerators
Andrew Alex
The hardware landscape of modern systems-on-chips (SoCs), which are made of a variety of digital signal processors (DSP), graphic processing units (GPU) and differing sizes of general purpose cores, is constantly evolving. Because each new generation of hardware requires an optimized kernel library to ensure that their machine learning and signal processing workloads run smoothly and efficiently, it can be difficult for performance engineers to keep up.
Performance engineers are turning to user-scheduled languages (USL), an emerging class of programming languages designed for heterogeneous hardware. They work by dividing the algorithm that specifies the program’s functional behavior from the schedule, which defines how that computation is carried out. Alex and Frisella aim to build a language system that is an extension of Exo, a popular USL that can optimize high-performance computing kernels on to new hardware accelerators, but lacks support for asynchronous parallelism or concurrency. Their proposed language system is capable of scheduling programs to exploit asynchronous and concurrent SoC targets, while also ensuring that the program’s behavior is preserved
Megan Frisella
“The tools for producing the highly-performant code that is important for fields like machine learning and signal processing have not kept pace with the ever-expanding capabilities of the hardware that runs the code,” said Alex, who is advised by Allen School professor Gilbert Bernstein. “Our project aims to remedy this gap by enabling a programmer to express this highly-performant, concurrent code as a sequence of equivalence-preserving transformations of a sequential program.”
Using such a system, performance engineers will not be limited to choices made by the optimizing compiler to transform their code using a cost model that may not reflect all the newly available features in the hardware it is targeting. Instead, engineers can apply their own domain and hardware-specific knowledge to the problem without existing tools such as compilers getting in the way — helping them write code faster and with less effort.
“Extending the power of user-scheduling to asynchronous and concurrent SoCs will unlock productivity in programming emerging hardware,” said Frisella, who is co-advised by Bernstein and faculty colleague Stephanie Wang.
Learning Multi-modal Agents to Reason and Act for Complex Multi-modal Tasks
Zixian Ma
Real-world multi-modal foundation models can help with various tasks ranging from answering simple visual questions about objects in daily life to solving more difficult problems about travel planning. Although these state-of-the-art models can answer generic and straightforward questions well, they struggle with complex questions and with generalizing about new tasks. For example, a user may take a picture of a panel showing different gas prices and ask the model how many gallons they can buy within a certain budget, but the model will have trouble answering.
To address these challenges, Ma and Hu propose to develop multi-modal agents that can explicitly reason about and act on these complex tasks using chains-of-thought-and-action. They aim to curate a new dataset with images from across various domains — such as daily life images, web screenshots and medical images — and pair it with a novel learning method.
“Our work aims to enhance open-source foundation multi-modal models’ capabilities to not only perform complex tasks through reasoning and actions but also do so in a more interpretable manner,” said Ma, who is co-advised by Allen School professor Ranjay Krishna and professor emeritus Daniel Weld.
With the large-scale dataset, Ma and Hu plan to train generalizable multi-modal agents using heterogeneous pretraining and domain-specific supervised finetuning and reinforcement learning. The researchers will build a similar architecture to that of heterogeneous pretrained transformers, which are able to combine a huge amount of data from multiple sources into one system to teach a robot an array of tasks using stems, trunks and heads.
Yushi Hu
In their proposed system, each stem features a domain-specific vision encoder that maps the visual data from various domains to visual features, or the numerical representations of an image’s visual content. The shared trunk is a transformer encoder block which connects these domain-specific visual features to shared representations in the same dimension of the text embeddings. Then the shared head, which is a decoder-only language model, takes both the visual tokens from the shared encoder as well as the text tokens of the input query, and generates the next set of text tokens following the inputs.
“This research focuses on developing artificial intelligence that can seamlessly understand, reason and generate across vision, language, audio — mirroring the way people interact with the world. By unifying these diverse streams, we aim to move beyond passive chatbots toward truly helpful agents that collaborate with humans and complete real-world tasks,” said Hu, who is co-advised by Allen School professor Noah A. Smith and ECE professor Mari Ostendorf.
The Allen School-affiliated teams are among three UW winners of the Qualcomm Innovation Fellowship this year. They are joined by ECE Ph.D. students Marziyeh Rezaei and Pengyu Zeng, who earned a fellowship to pursue their research proposal titled “Ultra-low Power Coherent Front-haul Optical Links to enable multi-Tb/s Capacity for 6G Massive MIMOs and Edge AI Datacenters.”
Rules are vital for building a safe and healthy functioning online community, and Reddit is no exception. For community moderators, however, it can be difficult to make data-driven decisions on what rules are best for their community.
A team of researchers in the Allen School’s Social Futures Lab and Behavioral Data Science Lab conducted the largest-to-date analysis of rules on Reddit looking at over 67,000 rules and their evolution across more than 5,000 communities over a period of five years — accounting for almost 70% of all content on the platform. This study is the first to connect Reddit rules to community outcomes. They found that rules on who participates, how content is formatted and tagged as well as rules about commercial activities were the most strongly associated with community members speaking positively about how their community is governed.
“This was my first paper, and I am extremely grateful for it to be named best paper of ICWSM 2025,” said lead author Leon Liebmann (B.S., ‘25), now at the online privacy company Westbold. “The work was difficult at times, and my advisors and co-authors Allen School Ph.D. student Galen Weld and professor Tim Althoff provided me with the direction and methods I needed to get it done. These people shaped my time in the Allen School and gave me a love for research I’d love to revisit.”
To better understand the rules on Reddit, the team first had to map out which communities had what rules and when. The researchers developed a retrieval-augmented GPT-4o model to classify rules into different categories based on their target, tone and topic. They then assessed the rules based on how common they were and how they varied across different communities, and also collected timelines on how the communities’ rules changed over time. At the same time, the researchers used a classification pipeline to identify posts and comments discussing community governance.
Taken together, this study can help inform Reddit moderators and community leaders on what rules their communities should have. The researchers found that the most common rules across communities covered post content, spam and low quality content, and respect for others — guidelines that platforms could use to create “starter packs” for new communities. They also found that how moderators word rules could influence how positively or negatively communities view their governance. For example, prescriptive rules, or those that describe what community members should do, are viewed more favorably than restrictive rules that focus on what community members should not do. By choosing to phrase rules prescriptively, moderators can help communities have a positive view of their governance.
In addition to Leibmann, Weld and Althoff, Allen School professor Amy X. Zhang was also a co-author on the paper.
“The industry will continue to need smart, creative software engineers who understand how to build and harness the latest tools — including AI.” Allen School leaders Magdalena Balazinska and Dan Grossman examine some of the myths surrounding computer science careers in the era of artificial intelligence. Photo by Yevhen Smyk, Vecteezy
There has been a lot of chatter lately, online and in various media outlets, about the supposed dwindling prospects for new computer science graduates in the artificial intelligence era. Recent layoffs in the technology sector have students, parents and educators worried that a degree in computing, once seen as a sure path to a fulfilling career, is no longer a reliable bet.
“The alarmist doom and gloom prevalent in the news is not consistent with the experiences of the vast majority of our graduates,” said Magdalena Balazinska, professor and director of the Allen School. “The industry will continue to need smart, creative software engineers who understand how to build and harness the latest tools — including AI. And the fact remains that a computer science degree is great preparation for a broad range of fields within and beyond technology, including the natural sciences, finance, medicine and law.”
In our own version of MythBusters, we asked Balazinska and professor Dan Grossman, vice director of the Allen School, to examine the myths and realities surrounding AI and the prospects for current and future Allen School majors. Their answers indicate that the rumored demise of software engineering as a career path has been greatly exaggerated — and that no matter what path computer science majors choose after graduation, they can use their education to change the world.
Let’s start with the question on everyone’s mind: What’s the job market like for computer science graduates these days?
Dan Grossman: Individuals’ mileage may vary depending on a range of factors, but what’s being reported in some media outlets doesn’t reflect what we’re seeing here at the University of Washington. More than 120 different companies hired this year’s Allen School graduates into software engineering roles. Amazon alone hired more than 100 graduates from the Allen School’s 2024-25 class. Google and Meta didn’t hire at that scale, but still more than last year — they each hired 20 graduates this year. Microsoft hired more than two dozen graduates from this latest class. I expect these numbers to grow as we hear from more graduates.
So, while the job market is tighter now than it was a few years ago, the sky is not falling. It’s important to remember that the Allen School is one of the top computer science programs in the nation, with a cutting-edge curriculum that evolves alongside the industry while remaining grounded in the fundamentals of our field. Our graduates are highly sought after, so their experience in the job market doesn’t necessarily reflect the experience of others. That has always been the case, even before this latest handwringing over AI. A B.S. in CS is not a uniform credential. The Allen School has always produced highly competitive graduates.
Magdalena Balazinska: In addition to those who found employment after graduation, more than 100 of our recent graduates opted to continue their education by enrolling in a master’s or Ph.D. program, which, of course, also makes us immensely proud!
How is AI impacting software engineering jobs?
Magdalena Balazinska: “The industry will continue to need smart, creative software engineers who understand how to build and harness the latest tools — including AI.”
MB: There are two factors: (1) AI’s impact on the work of a software engineer, and (2) AI’s impact on the job market for software engineers. Regarding the latter, it’s not so much that AI is taking the jobs, but that companies are having to devote tremendous resources to the infrastructure behind AI, which is very expensive. Also, many companies over-hired during COVID, and now they’re doing a course-correction for the AI era. I look at this as more of a reset. There’s no question that AI is affecting many areas of computing, just as it’s affecting just about every other sector of the economy. Companies will continue to invest in the people who know how to build and leverage these and other tools.
To my first point: With AI, we should expect the work of a software engineer to change, but to change in a really exciting way! The task of coding, or the translation of a very precise design into software instructions, can largely be handled by AI. But that’s not the most exciting or challenging part of software engineering. Understanding the requirements, figuring out an appropriate design, and articulating it as a precise specification are the hard parts. Going forward, software engineers will spend more time imagining what systems to build and how to organize the implementation of those systems, and then let AI handle many of the details of converting those ideas into code.
DG: One of our former faculty colleagues, Oren Etzioni, said, “You won’t be replaced by an AI system, but you might be replaced by a person who uses AI better than you.” I think that’s the direction we’re headed. Not AI as a replacement for people, but as a differentiator. Here at the Allen School, one of our goals is to enable students to differentiate themselves in this rapidly evolving landscape. For example, we are introducing a course on AI-aided software development, which will teach students how to effectively harness these tools.
How has AI affected student interest in the Allen School?
DG: Student interest remains strong — we received roughly 7,000 applications for Fall 2025 first-year admission.
MB: That may sound like a daunting number. However, we were able to offer admission to 37% of the Washington high school students who applied. That’s not as high as we would like it to be, but it’s far higher than public perception. We achieve this by heavily favoring applicants from Washington. For Fall 2025, we offered admission to only 4% of applicants from outside Washington.
If AI can write code, why should students major in computer science?
MB: Because computer science is so much more than coding! Creating a new system or application, perhaps a system to help the elderly take care of their daily tasks and manage their paperwork or a new approach for doctors to perform long-distance tele-operations, isn’t just a matter of “writing code.” A software engineer begins by clearly understanding the requirements — what the system needs to provide. Then the software engineer will decompose the problem into pieces, understand how those pieces will fit together, and anticipate failures. What happens if there is a power or network failure, or someone tries to hack the system? This gets progressively more challenging with the complexity and scale of systems that software engineers build, typically on teams with many people working together. Coding is the relatively easier part.
DG: In that spirit, here in the Allen School, we do teach students how to code, but as a component of how to envision, design and build systems and applications that solve complex problems and touch people’s lives. The principles, precision, and reasoning gained from reading and writing code is a necessary foundation that serves our students very well — including graduates who now use AI in industry. It is the software engineers with the deepest knowledge who will be most effective at using AI to write their code, because they will know how and where AI can go wrong and how to steer it toward producing a correct output.
MB:Engineers have always used tools, and their tools have always advanced with time and opened the door to innovation. Thanks to developments like modern coding libraries and languages, online repositories like StackOverflow, GitHub, automated testing, cloud computing, and more, software engineers today are far more efficient and can develop applications more quickly than ever before. And this was before AI for coding had really taken off. And yet, there are more software engineers doing more interesting and important things than ever before!
How does the Allen School prepare students for a workplace — and a world — being transformed by AI?
Dan Grossman: “One of our goals is to enable students to differentiate themselves in this rapidly evolving landscape.”
MB: As a leader in AI research, the Allen School is ideally positioned to help students learn how to use AI, how to build AI, and how to move the field of AI forward to benefit humanity. We give students multiple opportunities to explore AI topics and tools as part of our curriculum. Dan mentioned our AI-assisted software development course, and many of our other courses allow for using AI assistance in well-prescribed ways. This enables students to focus on core course concepts, generate more complex projects, and so on. Gaining experience with any AI tool can give a sense of what the technology can help with — along with its limitations. That said, we will continue in some courses to expect students to build, design, test, and document software without AI assistance.
DG: Our courses sometimes use the same cutting-edge tools used in industry, and other times will provide a simpler setting for pedagogical purposes. Software engineering tools change rapidly, so we tend not to get into the weeds on any one particular tool but give students the confidence to pick up future tools. Importantly, we don’t just teach students how to build and use AI. We also help them to think critically about the ethics and societal impacts of these technologies, such as their potential to reinforce bias or be used as a surveillance tool, and ways to mitigate those impacts.
MB: Another advantage we have at the Allen School is that we are a leading research institution, and our faculty are among the foremost experts in the field. This gives us the ability to incorporate new concepts and techniques into our coursework quickly. We also have a program devoted to supporting undergraduates in engaging in hands-on research in our labs alongside those very same faculty and our amazing graduate students and postdocs. Many students choose to get involved in research during their undergraduate studies.
What if a student is interested in computing, but not AI?
DG: Great! There are many open challenges across computing, from systems development, to human-facing interactive software design, to hardware design, to data management, and many others. Even if you are not using or developing AI itself, building systems that can run AI efficiently is driving a lot of exciting work in the field these days. While the big breakthroughs that have been driving rapid change over the last few years are AI-centered, computing remains a broad field.
MB: A student can major in computer science and follow their passion wherever it takes them. A subset of students will choose to study AI and build the next AI technologies, but the vast majority will use AI as a tool while building systems for medicine, education, transportation, the environment, and other important purposes. Or they will build back-end infrastructure at global companies like Google, Amazon, or Microsoft, or tackle other challenges like those that Dan mentioned. The more we advance computing, the more we open new opportunities. I think that’s why the number of software engineers just keeps growing. There is always more to do. The job is never done.
What is your advice to current and aspiring computer science majors who worry about their career prospects with the rise of AI?
MB: First, if you think you want to be a computer scientist or a computer engineer, pursue that! If you choose a major that you are excited about, you will not mind spending hours deepening your knowledge and sharpening your skills, which will help you to become an expert and to enjoy your chosen profession even more. My advice to every student is to take a broad range of challenging courses. Learn how to use the current tools, with the understanding that the tools you use today will not be the ones you use tomorrow. This field moves fast, which is what makes it exciting.
When it’s time to start your job search, whether for an internship or a full-time job, apply broadly. Apply to large companies, small companies, companies in various sectors, non-profits, and so on. Many organizations need software engineers! And not all interesting technical jobs that use a computing degree have the title of software engineer. Pick the position where you will learn the most. It’s important to optimize for learning and for growth, especially early on in one’s career.
DG: I would also remind students that a UW education is not about vocational training; our goal is that students graduate with the knowledge and skills to succeed in their chosen career, yes, but also to be engaged citizens of the world. While you’re here, make the most of your education — take a range of challenging courses and put in the time to learn the material. After all, it is a multi-year investment on your part, and the faculty have invested a lot of time and effort into creating a challenging, coherent curriculum for you. Take the hardest classes that you think are also the most exciting ones, and then focus on learning as much as you can.
Any final thoughts?
DG: Don’t choose a major solely because it’s popular. Choose a major that you’re passionate about. If that’s computer science or computer engineering, we’d love to see you at the Allen School. If it’s something else, we’d still love to see you in some of our classes.
MB: Everyone can benefit from learning at least a little computer science, especially now in the AI era!
Back in 2011, a team of University of Washington and University of California San Diego researchers published a paper detailing how they could remotely hack into a pair of 2009 Chevy Impalas. By targeting a range of attack vectors including CD players, Bluetooth and cellular radio, the researchers were able to control multiple vehicle functions, from the windshield wipers to the brakes.
Since its publication, the team’s research has helped lead to new standards for motor vehicle security and put the brakes on automobile cyberattacks. For their lasting contributions, their paper titled “Comprehensive Experimental Analyses of Automotive Attack Surfaces” received the Test of Time Award at the 34th USENIX Security Symposium in Seattle earlier this month.
Franziska Roesner
“I was only a first-year graduate student when we started this project, and I had just switched my focus to security. It was such a privilege to be able to help out on such an important and impactful project, and to learn from all of the other members of the team about how to do this kind of research,” said co-author Franziska Roesner (Ph.D., ‘14), Brett Helsel Professor and co-director of the Security and Privacy Research Lab in the Allen School.
Modern automobiles are made up of independent computers called electronic control units (ECUs), typically connected through the Controller Area Network (CAN), that oversee different motor functions. In a previous paper, the team found that if an attacker physically connected to the car’s internal network could override critical safety systems. Building off of that work, the researchers analyzed the modern automobile’s external attack surface and found that an adversary could hack into a car from miles away.
The team identified three categories of components that were vulnerable to cyberattacks. An attacker could use an indirect physical channel such as tools that connect to the OBD-II port, which can access all CAN buses in the car, or through the media player. For example, the researchers compromised the car’s radio and then used a doctored CD to upload custom firmware. If an attacker is able to place a wireless transmitter in proximity to the car’s receiver, they can gain access to the ECU via Bluetooth or even remote keyless entry, the team found. Attackers do not have to be nearby to wreak havoc. Using long-range communication channels such as cellular, it is possible to exploit vulnerabilities in how the car’s telematics unit uses the aqLink code to remotely control the vehicle.
“More than 10 years ago, we saw that devices in our world were becoming incredibly computerized, and we wanted to understand what the risks might be if they continued to evolve without thought toward security and privacy,” said senior author Tadayoshi Kohno, who was then a professor at the Allen School, now faculty at Georgetown University, in a UW News release.
The impact of the team’s work can still be felt today. As a result of the research, car manufacturers including GM have hired entire security teams. The work has influenced the development of guidelines for original equipment manufacturers (OEMs) and also led to the creation of the Electronic Systems Safety Research Division at the National Highway Traffic Safety Administration. As cars grow increasingly more connected and autonomous, the insights from the UW and UCSD collaboration will continue to inform the automotive industry against emerging threats.
“Beyond the practical impact of the work, that experience has also made for great stories to tell in the computer security courses I teach now — for example, the time that we accidentally set the car’s horn to a permanent ‘on’ state while experimenting outside the Allen Center,” Roesner said.
Joining Roesner and Kohno at UW at the time of the original paper were Karl Koscher (Ph.D. ‘14), now a postdoc at UCSD, and Alexei Czeskis (Ph.D., ‘13), currently at LinkedIn. The original University of California San Diego group was made up of UCSD faculty members Stefan Savage (Ph.D., ‘02) and Hovav Shacham; Stephen Checkoway (B.S., ‘05), now faculty at Oberlin College; Damon McCoy, faculty at New York University; Danny Anderson, who runs a software consulting company; and late researcher Brian Kantor.
The Allen School at the University of Washington is working with Ai2 and other partners on a new initiative to advance open AI for science and the science of AI, with support from the U.S. National Science Foundation and NVIDIA.
The University of Washington’s Paul G. Allen School of Computer Science & Engineering has teamed up with the Allen Institute for AI (Ai2) on a new project aimed at developing the first fully open set of artificial intelligence tools to accelerate scientific discovery and enhance the United States’ leadership in AI innovation. Today the U.S. National Science Foundation (NSF) and NVIDIA announced a combined investment of $152 million in this effort, including $75 million awarded through the NSF’s Mid-Scale Research Infrastructure program.
Ai2 will lead the Open Multimodal AI Infrastructure to Accelerate Science (OMAI) project. The principal investigator is Ai2 Senior Director of NLP Research Noah A. Smith, who is also Amazon Professor of Machine Learning at the Allen School. Smith’s faculty colleague Hanna Hajishirzi, Torode Family Professor at the Allen School, is co-principal investigator on behalf of UW and also Ai2’s senior director of AI.
“OMAI is a terrific opportunity to leverage the longstanding partnership between Ai2 and the Allen School, which has yielded some of the most exciting developments in building truly open AI models and trained some of the most promising young scientists working in AI today,” said Hajishirzi. “This is a pivotal moment for us to form the foundation for scientific discovery and innovation across a variety of domains — and also, importantly, advance the science of AI itself.”
Noah A. Smith (left) and Hanna Hajishirzi aim to leverage the partnership between Ai2 and the Allen School to benefit science and society.
The cost of building and maintaining today’s AI models is too prohibitive for all but the most well-resourced companies, leaving researchers in academic and not-for-profit labs without ready access to these powerful tools and stifling scientific progress. The goal of the OMAI project is to build out this foundational infrastructure through the creation and evaluation of models trained on open-access scientific literature and informed by the needs of scientists across a range of disciplines. By openly releasing the model weights, training data, code and documentation, the team will provide researchers using its tools with an unprecedented level of transparency, reproducibility and accountability, instilling confidence in both the underlying models and their results.
The concept for OMAI was incubated in an ecosystem of open research and collaboration that the Allen School and Ai2 have built since the latter’s founding in 2014. That ecosystem has enabled dozens of UW students to collaborate with Ai2 on research projects, produced leading-edge open AI artifacts like the Open Language Model (OLMo) and Tulu, and developed tools like OLMoTrace to give anyone full visibility into models’ training data — all of which have helped fuel Seattle’s emergence as a hub of AI innovation.
Smith looks forward to leveraging that longstanding synergy to push technologies that will have a transformational impact on the American scientific enterprise — and even transform the conversation around AI itself.
“There’s been a reaction that seems to be widespread that AI is a thing that is happening to us, as if we are passively subject to this technology and don’t have agency,” Smith said. “But we do have agency. We get to define what the priorities should be for AI and to build tools that scientists will actually be able to use and trust. With OMAI, the UW will be a leader in this new paradigm and push AI in a more responsible direction that will benefit society in a multitude of ways.”
In addition to the UW, academic partners in the OMAI project include the University of Hawai’i at Hilo, the University of New Hampshire and the University of New Mexico.
OMAI represents a landmark NSF investment in the technology infrastructure needed to power AI-driven science — a development that Brian Stone, performing the duties of the agency’s director, described as a “game changer.”
“These investments are not just about enabling innovation; they are about securing U.S. global leadership in science and technology and tackling challenges once thought impossible,” Stone said.
Understanding how different parts of the brain communicate is like trying to follow conversations at a crowded party — many voices overlap, some speakers are far away and others might be hidden entirely. Neuroscientists face a similar challenge: even when they can record signals from multiple brain regions, it is difficult to figure out who is “talking” to whom and what is being said.
Matt Golub
In a recent paper published at the 2025 International Conference on Machine Learning (ICML), a team of researchers led by Allen School professor Matt Golub developed a new machine learning technique to cut through that noise and identify communication between brain regions. The technique, called Multi-Region Latent Factor Analysis via Dynamical Systems (MR-LFADS), uses multi-region neural activity data to decode how different parts of the brain talk to each other — even when some parts can’t be directly observed.
“The many regions within your brain are constantly talking to each other. This communication underlies everything our brains do for us, like sensing our environment, governing our thoughts, and moving our bodies,” said Golub, who directs the Systems Neuroscience & AI Lab (SNAIL) at the University of Washington. “In experiments, we can monitor neural activity within many different brain regions, but these data don’t directly reveal what each region is actually saying — or which other regions are listening. That’s the core challenge we sought to address in this work.”
Unlike existing approaches, MR-LFADS is able to automatically account for unobserved brain regions. For example, neuroscientists can use electrodes to simultaneously monitor the activity of large populations of individual neurons across multiple brain regions. However, this activity may be influenced by neurons and brain regions that are not being recorded, explained Belle Liu, UW Department of Neuroscience Ph.D. student and the study’s lead author.
“Imagine trying to understand a conversation when you’re not able to hear one of the key speakers. You’re only hearing part of the story,” Liu said.
To overcome this, the team devised a custom deep learning architecture to detect when a recorded region reflects an unobserved influence and to infer what the unobserved region was likely saying.
“We wanted to make sure the model can’t just pipe in any unobserved signal that you might need to explain the data,” said co-author and Allen School postdoc Jacob Sacks (Ph.D., ‘23). “Instead, we figured out how to encourage the model to infer input from unobserved sources only when it’s very much needed, because that information can’t be found anywhere else in the recorded neural activity.”
The team tested MR-LFADS using both simulated brain networks and real brain data. First, they designed simulated multi-region brain activity that reflected complicated scenarios for studying brain communication, such as giving each region unique signals from both observed and unobserved sources. For the model, the challenge is to recover those signals and to disentangle the source of those signals and whether they come from observed regions — and if so, which ones — or unobserved regions. The researchers found that their model was able to infer this communication more accurately than existing approaches. When applied to real neural recordings, MR-LFADS could even predict how disrupting one brain region would impact another — effects that it had never seen before.
By helping neuroscientists better map brain activity, this model can help provide insights into treatments for various brain disorders and injuries. For example, different parts of the brain communicate in certain ways in healthy individuals, but “something about that communication gets out of whack in a diseased state,” explained Golub. Understanding when and how that communication breaks down might enable the design of therapies that intervene in just the right way and at just the right time.
“Models and techniques like these are desperately needed for basic neuroscience to understand how distributed circuits in the brain work,” Golub said. “Neuroscientists are rapidly improving our ability to monitor activity in the brain, and these experiments provide tremendous opportunities for computer scientists and engineers to model and understand the intricate flow of computation in the brain.”
