ARENA 7.0 Participant (see Dani’s LinkedIn, GitHub, and website)

I have a background in ML and signal processing, and after many years doing non-technical business work in the music industry, I spent the last two years pivoting back towards technical ML work, with ARENA serving as a culmination of that transition. I was advised by many people that the pivot would be too hard and competitive, but I found technical AI safety both socially meaningful and intellectually stimulating, which gave me the motivation to keep going. Beyond the highly enjoyable in-depth curriculum, the capstone project offered a taste of what working on open-ended research problems is like, which gave me a lot more confidence in my ability to do real research. The talks and informal conversations at LISA also challenged some assumptions I'd arrived with about AI safety, and broadened my sense of what impactful work can look like. ARENA gave me both the credibility and the clarity to pursue technical AI safety roles seriously, and I'm hoping to turn that into a full-time position soon.

ARENA 7.0 Participant (see Tyler’s LinkedIn, GitHub and website)

I came to ARENA from a software engineering background (mostly web applications) and began shifting toward ML through an MS in CS. I applied because I wanted a faster path to strong fundamentals in mechanistic interpretability and research engineering, plus the tight feedback loop you get from building alongside other highly-motivated researchers. The most valuable parts were the structured exercises, the emphasis on “understand it by implementing it,” and the cohort environment. There is lots of rapid iteration, debugging, and idea exchange. Participating in the programme was so fun! ARENA helped me build confidence in my ability to read papers, translate concepts into working code, communicate results clearly, and it has also expanded my network in the interpretability/alignment space. Post-ARENA, I’m continuing to deepen my interpretability work through research projects and coursework, with the goal of moving into a research engineering/ML research role focused on mechanistic interpretability and AI safety.

ARENA 7.0 Participant (see Kat’s LinkedIn and website)

As a Ph.D. student in Computer Science, I realized that the most critical problems facing my field today are in making AI systems safe and aligned, and I wanted to direct my research toward addressing them. Coming from a different line of research, I wasn't sure how to make that transition, but ARENA provided a clear pathway. While the curriculum and learning materials were incredibly valuable, the biggest impact came from the connections I made at LISA, the collaborations that formed, and the conversations that followed. I'm leaving with both the confidence to pursue research in this space and a renewed sense of inspiration to tackle these challenges. Currently, I'm wrapping up my dissertation while applying to research positions in the field. Because of ARENA, I feel much better positioned to find a role that I am genuinely passionate about.

ARENA 7.0 Participant (see Alex’s LinkedIn, GitHub and website)

After ARENA, I have a firm grasp on my knowledge gaps and how to fill them, and on how to quickly go from research idea to testing. Most importantly, I have direction in my research. What I enjoyed most was the people: the participants, the ARENA staff, and the other AI safety people at LISA. Listening to how everyone thinks about research was deeply inspiring.

This programme has given me strong confidence in my ability to contribute to AI safety research, a clear sense of direction, and a community of friends I can talk to about research. I'm now actively pursuing the transition: I've applied for a career transition grant through Coefficient Giving, I'm receiving mentored guidance through Successif, and I'm applying to positions in AI safety.

ARENA 7.0 Participant (see Giuseppe’s LessWrong profile here)

I’m a tech CTO with a background in cultural anthropology, and I’m pivoting into alignment science with a bias toward mechanistic interpretability. I applied to ARENA to stop engaging with AI safety at a distance and instead learn inside the community, in person, at full pace. The content was dense and hands-on, but what made it work was the people: TAs and peers who were generous with feedback, blunt in the useful way, and serious about turning ideas into experiments. That environment made the field feel real and livable, and it left me with stronger technical foundations and a clearer research direction. I’m currently CTO at Orma Lab and also doing independent research, building interpretability tooling and writing about alignment and model organisms.  

ARENA 7.0 Participant (see Ahmed’s LinkedIn here)

Having a technical background but spending most of my time setting up projects to address infrastructure gaps in the broader social impact space, I came to ARENA for two reasons: To better understand and address bottlenecks around AI risk mitigation and to upskill myself to contribute directly to alignment research. ARENA, while intense, has been a great space for rapid technical upskilling, making many meaningful connections, and improving my world model for mitigating AI risks. I am currently back in Australia, where I have taken up a teaching assistant role for the Technical Alignment Research Accelerator, am setting up the Queensland AI Safety Initiative, and pursuing research on adversarial robustness of AI systems for cyber defence at the University of Queensland.

ARENA 7.0 Participant (see Johannes’ GitHub here)

My background is in physics and applied mathematics. In the final year of my master's, I discovered AI safety through a local student group, then did the ML4Good bootcamp to test my fit/enthusiasm/commitment (it worked). ARENA's hands-on work (implementing probes, RL, building evals, and especially the capstone project) was invaluable, and I believe it directly contributed to landing an AI safety PhD position in Sahar Abdelnabi's group (I talked a lot about the capstone project in my interview). Until the PhD starts, I'm facilitating technical AI safety courses at BlueDot and doing a SPAR project on emergent misalignment.

ARENA 7.0 Participant (see Davide’s LinkedIn here)

Coming from a background in AI and mathematics, I joined ARENA during the first year of my PhD at Mila to deepen my general understanding of LLM internals and alignment techniques. The program was an incredible opportunity to get hands-on experience with the engineering realities of safety research, but the highlight was undoubtedly the stimulating environment; collaborating with peers in London and debating ideas over lunch was invaluable. I made many new friends, gained the confidence to tackle more ambitious technical problems and directly informed my current research direction. I am now continuing my PhD work on language-model-based agents, specifically focusing on private working memory and the faithfulness of self-explanations.

ARENA 7.0 Participant

Before ARENA, I was doing independent research on understanding toy transformer models (trained on self-avoiding walks on a grid); how a model can use extra compute on tokens unrelated to the task to improve performance, and information flow methods for testing faithfulness (SPAR project with Noah Siegel). I was also finishing my PhD in geometry and topology in Saint Petersburg. I applied to ARENA to strengthen my technical skills, meet great researchers, ask them questions I am interested in, find job opportunities and more. The part I enjoyed most turned out to be evaluations: I liked observing how models behave under different conditions, and carefully adjusting setups to isolate the behaviour I wanted to study. ARENA gave me a great big-picture view – grounded in concrete details – of how the AI safety field works in practice, what matters most right now and how I can successfully re-approach fundamental questions that I asked before without success. More than that, I am now in contact with great researchers with whom I can discuss my ideas!

ARENA 6.0 Participant (see Hugo’s LinkedIn here)

Coming from a background in software development and mathematics with an interest in AI Safety, I applied to ARENA to gain hands-on experience with the skills necessary for technical AI Safety work. What I loved most about the programme was the exceptionally dedicated learning environment it provided, from the LISA office space and provided meals to the stimulating discussions with peers and other AI Safety researchers in the co-working space. ARENA significantly boosted my confidence in presenting existing research and developing ideas at the forefront of AI Safety research. I'm now doing an internship as a Quantitative AI Safety Researcher on AI evaluations at the Existential Risk Observatory, where I'm applying the skills learned from ARENA's evaluation week.

ARENA 6.0 Participant (see Mo’s LinkedIn here)

I was a founding software engineer of a tech startup for 3 years. After rapid gains in AI capabilities, I was concerned about their safety and pivoted full-time to AI research, starting by pursuing a master's in AI. I applied to ARENA because it was the perfect way to gain a comprehensive overview and hands-on skills needed for doing safety research. I am glad that I got to attend the programme in person and loved every aspect of it – from meeting the fellow participants and the vibrant LISA community to the comprehensiveness of the material, to making it more manageable and engaging by pairing up with fellow participants and getting continuous assistance from the TAs. After ARENA, I feel more confident and connected than ever and feel ready to make impactful research and contributions. I am currently working as an independent researcher, funded by Open Philanthropy, to advance our understanding of sleeper agents and backdoored LLMs.

ARENA 6.0 Participant (see Lenz’s LinkedIn here)

I finished my undergrad degree in humanities right before ARENA, although I was already involved in AI safety field-building as an operations contractor. I applied to ARENA 6.0 since in the months prior, I started to transition towards value alignment research, but realised I lack a lot of technical training aside from just knowing how to code because of Leetcode. I would definitely say participating in the full live cohort helped me level up my programming skills and connect with a deeply curious community who helped me solidify a theory of change for myself. Since the programme, I've been doing independent consulting for AI policy organisations while applying to other AI safety fellowships with the intention of exploring interpretability-based evals.

ARENA 6.0 Participant (see Nitzan’s LinkedIn here)

Before ARENA, I worked in cybersecurity (reverse engineering and system security). I joined ARENA to pivot into AI safety and apply those skills where they can have the most impact. I loved the high-quality content, the pace, and the excellent staff. ARENA gave me clarity, confidence, and a network that translated into concrete opportunities. Today I’m Head of Cyber at Heron Security, helping bridge cyber talent to AI security while deepening my foundation in advanced AI systems and pursuing my own research using skills I learned in the programme.

ARENA 6.0 Participant (see Alexander’s Google Scholar page here)

I joined ARENA as a second-year PhD student working on LLM jailbreaks and red-teaming. What I loved most about the programme was its strong focus on interpretability and the deep intuition it builds around LLM internals. Working from LISA was an amazing experience: being surrounded by like-minded people, learning from others’ projects, and sharing ideas over lunch. I built lasting connections with Anthropic/MATS fellows working from there, and I’m excited to collaborate with them. I’m now back to my PhD, with newly adjusted focus on white-box mitigation techniques for alignment.

ARENA 6.0 Participant (see Luca’s GitHub here)

Coming from a (PhD) research background in computational neuroscience, participating in ARENA was a fantastic opportunity to deepen my understanding of the technical AI safety landscape while gaining hands-on experience implementing and experimenting with different approaches. Having first become interested in AI safety while living far from the main hubs, being physically present at LISA made me feel genuinely connected to the field and its community: I built valuable connections with peers and engaged in conversations with more experienced researchers. After ARENA I feel more confident in my AI safety research skills and have greater clarity about my next steps.

ARENA 6.0 Participant (see Sam’s LinkedIn here)

Before ARENA, I completed my Ph.D. in physics in 2023; my graduate research was on machine learning and data science techniques in x-ray spectroscopy. After that, while applying to jobs and learning more about LLMs, I discovered AI safety and was immediately captivated. Since then, I have been upskilling in AI safety research. I applied to ARENA to learn the content in a collaborative and immersive setting, but the experience was more than I expected. Not only were the TAs phenomenal, but pair programming on the content with the other participants was by far my favorite part, making the content more fun and accessible than working through it alone. Additionally, both the mealtime lectures at LISA and conversations with others in the space were invaluable in helping me learn from and connect with the community. I’m currently participating in SPAR — my team is working on multi-signal AI control — but I am also looking for full-time research experiences. I am excited to see where I end up next!

ARENA 5.0 Participant (see Vili’s LinkedIn here)

I'm an applied math PhD student at Aalto University researching high-performance computing. Nearing the end of my doctorate, I had started to focus more on AI safety. ARENA was perfectly timed for me, and I loved the support throughout the programme – it made the intensive learning experience both manageable and really enjoyable. The programme equipped me with the hands-on skills I was missing to contribute meaningfully to AI safety work. Now on top of finishing my PhD I'm running another bootcamp based on ARENA's materials, helping to get about fifteen people in my home country up to speed on technical safety.

ARENA 5.0 Participant (see Matteo’s website here)

Before ARENA, I was diving deep into AI safety during my Ph.D., especially around mechanistic interpretability and machine unlearning. I joined the programme to get more hands-on experience and connect with others in the field, and it absolutely delivered! It was intense, challenging, and super fun. I learned a lot, but what stood out the most was the incredible community: the feedback, support, and curiosity from everyone made the experience truly special. I’m now back to research, feeling more confident and connected than ever.

ARENA 5.0 Participant (see Nadja’s LinkedIn here)

I decided to do ARENA because I’ve been adjacent to the AIS field for a while (I work in forecasting x-risks), and felt that I needed a deeper technical understanding of the field. What I enjoyed most about ARENA was building and understanding GPT-2, seeing that intelligence is essentially just matrix multiplication. ARENA has given me a much better understanding of models, evals and their behaviour. My work in forecasting is going to benefit from this enormously, because I am now much more equipped to think critically about threat models involving AI. 

ARENA 5.0 Participant (see Anton’s LinkedIn here)

I have a background in Physics and Data Science, working with ML and AI, and the recent advances in capabilities worried me. I wanted to contribute to AI safety but I had only a high-level understanding of the different fields within AI safety, so I applied to ARENA in order to get hands on experience on a wide range of technical AI safety approaches. The curriculum is brilliant, and I'm so glad I was offered the opportunity to attend in person, having support from other students and teaching staff was invaluable for deeply understanding the content. Meeting other people who are passionate about AI safety helped me immensely in developing my thoughts, and in solidifying my motivation to pursue a career in this field.

ARENA 5.0 Participant (see Jasper’s LinkedIn here)

I came to ARENA from a backend web-engineering background, eager to move beyond black-box API work into building, interpreting and fine-tuning language models with PyTorch and RLHF. I found the challenging curriculum and pair-programming setup invaluable, especially the week-long capstone research sprint that solidified my practical skills. ARENA boosted my confidence in running local experiments and testing bespoke models, and serves as a strong signal of ML expertise to prospective employers. Since graduating, I’ve co-authored a NeurIPS submission with FAR.AI collaborators and am now pursuing positions at leading AI safety research organisations.

ARENA 5.0 Participant (see Leon’s website here)

I come from a theoretical background with research experience in multivariate information theory and the theory of reward learning. Now that I'm at the end of my PhD, ARENA was a great opportunity to complement my knowledge with engineering skills. I really enjoyed ARENA's structured approach to learning difficult materials, with 6 hours of concentrated pair-programming per day that guided us quickly through a range of topics, from implementing a transformer to PPO and evals with Inspect. I'm now much more confident that I can reproduce interesting papers if I want to, and I expect these skills will help me in my upcoming job search for research positions in AI Safety and Alignment.

ARENA 5.0 Participant (see Liza’s website here)

I'm a social scientist who entered AI research during my PhD. As such, my training on specific & recent AI methods (RL, mech interp etc.) all came from self-taught efforts, and felt a little shaky even towards the end of the PhD. I thought it would be very useful to brush up on these concepts and to learn to build models such as transformers from scratch. ARENA certainly gave me confidence and reduced my impostor syndrome in both AI Safety and AI research more broadly. It goes without saying that my knowledge and practical implementation skills are now much stronger as a result of the course as well. In addition, the connections I made on the course have been fun and I'm sure will open doors in the future. My next steps involve contracting roles at Meta and Google DeepMind, and then a full-time role at UK AISI. 

ARENA 5.0 Participant (see Josh’s LinkedIn here)

After time as a data scientist consultant, I started doing entry level courses in AI Safety, quickly realising that I wanted to become a full-time alignment research engineer. I had some disappointing interviews, making it apparent that I was lacking in broad AI/ML safety research knowledge outside of evals, ARENA was a perfect fit for this reason. The programme has been really beneficial – growing my network, enhancing my research skills, and accelerating my understanding of the AI safety landscape. I am now working in model evaluations as a Research Engineer with Equistamp (collaborating with METR/AISI among others), and I hope to receive funding from Open Philanthropy to do independent AI safety research.

ARENA 5.0 Participant (see Michele’s website here)

I studied math (a Bachelor's and Master's degree) with a main focus on mathematical logic. When I discovered the world of AI safety I decided to pursue a PhD on a related topic trying to contribute effectively to the field: ARENA was the most useful thing that could happen to someone with a mathematical background who wants to enter technical AI safety research. But ARENA was more than a way to develop some ML skills: it is a way to challenge some of my points of view (not only about AI safety) and a valid method to meet a lot of interesting and beautiful people. All of these teachings will for sure be helpful in my career.