Are you interested in helping more people transition into AI safety? š
Well, youāre in luck! In this issue:
A Retrospective: some reflections on the week
Context Week Connections: the bios of 10 people from Context Week who are transitioning into AI safety, how you can help them, and their contact information
A Brief Retrospective
Context Week ended up being much more successful than I anticipated, but not necessarily for reasons of āgiving people context.ā Instead, I think the greatest value of Context Week came in the 1-on-1 career guidance we gave participants, and the immersive community experience.
We had participants ranging from 15 to 800 hours of exposure to / involvement in AI safety; for those on the latter end of the spectrum, I was worried they wouldnāt find much value in e.g. reading AI 2040.
However, those folks often came into Context Week with some specific uncertainties they wanted to chat through, e.g. what org would be the best fit for me? Would it be more valuable to work at CAISI or METR? How do I go from āI lurk on LessWrong and read a bunch of AI safety stuffā to āI seriously want to work in the space full timeā?
For these kinds of folks in particular, having 1-on-1 chats with facilitators was a huge value add: we talked through specific cruxes and connected them with relevant people.
With that said, given how much value there was in 1-on-1 advising, Iām sad to see that BlueDot no longer does 1-on-1 calls with people! Their advising page is now solely dedicated to Career Transition Grants. However, there are still orgs that offer this service; Iāll list a few below in case useful:
80,000 Hours ā āfree, personalised advice on your highest-impact careerā
Successif ā mid-career and senior professionals, with a particular focus on AI safety
Probably Good ā for all experience levels
Impact First ā UK, students
Iām really excited by the number of people who, post-Context Week, took concrete steps in the direction of full-time AIS roles! Iāve been having follow-up calls with some of the participants, and itās been lovely to see.
For some reason, we had a bunch of speakers flake at the last minute, rip. I donāt really know why this is, but found it disappointing!
I think the event could have gone better with slightly more prep (my understanding is that this was organized in the span of a couple of weeks, or less?) because there were some points at which the draft syllabus didnāt seem super valuable to me. That said, Iām grateful that Harry gave me executive control over how to steer my group (so I ended up just doing my own thing at some points, which went well I think lol).
A suggestion for next time: Make a site like Praxis and give attendees + speakers access to bios beforehand (maybe also use an AI tool to pair up people who would benefit most from speaking to each other).
Another suggestion: āContext Weekā as a name is fine, but perhaps something like āTransition Weekā would be more accurate? (Def continue workshopping that haha)
Context Week Connections š·
Some lovely people from BlueDotās context week are looking to meet others in AI safety ā for advice, collaborations, roles, etc. If you work at one of the orgs they mention, or know someone who does, it would be really lovely if youād reach out and help guide them as they navigate the field!
Christine Corry
Bio: Christine just graduated from UC Berkeley with degrees in Computer Science and Astrophysics. She is currently a Second Look Research Fellow at UChicagoās XLab, replicating empirical AI safety research (no-CoT evals and emergent misalignment work). Sheās built internal LLM-based tooling for Apple, trained DNA prediction LLMs at the Innovative Genomics Institute, and spent three years studying pilot cognition at NASA. Sheās increasingly convinced that frontier AI policy would be higher leverage and better suited to her skillset as a communicator energized by interpersonal work and focused on translating technical concepts for general audiences.
Looking to meet: Specific orgs: Encode AI, SAIP, CAISI, other think tanks, people actively working in politics, Horizon, GovAI, IAPS (just applied to their special projects role and very excited about my fit for that). Generally: macro-strategy and US-China people.
Roles of interest: Politics directly (think tank, congressional staff, advocacy), strengthening DCāBay Area communications (special projects type role), general field building.
Contact: christinecorry1405@gmail.com
Ishaan Makkar
Bio: I studied physics at Imperial College London and co-founded Hexa, where I built and sold B2B AI products and helped scale the business to $16k MRR. Iām now exploring AI safety research, particularly RL environments for misalignment, evaluation and monitoring. Iād like to work with experienced researchers and find safety problems I could build a startup around.
Looking to meet: Researchers and engineers working on alignment, AI control, RL environments for misalignment, evaluations and monitoring. Nonprofits Iām interested in include Redwood Research, FAR.AI, METR, Alignment Research Center, the Center for AI Safety and Palisade Research. Iād also like to connect with MATS researchers and mentors, teams at Anthropic, OpenAI and Apollo Research, and RL-environment providers such as Surge AI or Scale AI.
Iād like to hear which problems these teams are struggling with and where I could contribute. Iād welcome research feedback or opportunities to collaborate, and conversations with founders working on AI safety.
Roles of interest: Research or engineering roles and collaborations in these areas, at either nonprofits or frontier labs. Iād like to work with people I can learn from and help build things that make AI systems safer. Longer term, Iām interested in building an AI-safety startup.
Contact: ishaan@hexaagents.com
Max Perin
Bio: Iām a recent grad (political science + philosophy at UCLA) who, until very recently, was committed to a very traditional prelaw track. I was heavily involved in competitive debate throughout high school and college, which led me to Praxis and then BlueDotās context week, and Iāve been trying to pivot into AIS since.
Looking to meet: Iām most interested in policy research, think tank work, political advocacy, and potentially non-technical forecasting. Would love to chat further with folks from RAND and the AI Futures Project, and would be very interested in intros to people at AE Studio, FAI, and advocacy orgs that may not be on my radar.
Roles of interest: I have some background in nuclear policy/governance and have given a bit of thought to how that does/does not map onto potential governance regimes, but by and large Iām a generalist whoās probably best suited for roles that are highly analytical and where persuasive communication skills are a value add.
Contact: maxvperin12@gmail.com
Cat McGee
Bio: Software engineer focusing on cryptography (especially zero knowledge).
Looking to meet: People in the US or London (preferably London) working on or interested in verification of AI claims ā software or hardware. Also interested in policy people who are trying to build towards Plan A.
Roles of interest: AI verification, but not sure of the specifics yet. āI might build something myself. I am currently interested in just expanding my network to learn and potentially partner with people in future; right now I will spend 6 months doing my own things (& maybe MATS).ā
Contact: hello@mcgee.cat
Ray Che
Bio: I am a second-year CS PhD at UIUC working on AI and cybersecurity. I am interested in LLM security and adopting AI to secure systems. I received my BS and MS from Tsinghua University. My academic website is ray-che.com.
Looking to meet: Frontier AI security and control researchers and engineers at METR, Redwood, OpenAI, Anthropic, Google, Palo Alto.
Roles of interest: Security research scientists and engineers at frontier labs, Redwood/METR, and big techs.
Contact: xijiac2@illinois.edu
James Sykes
Bio: Iām a statistics PhD student at the University of Warwick working on cooperative AI with Dr Paolo Turrini and Prof Paul Jenkins. Currently, Iām doing a project on āMulti-Agent RL for Structured Population Dynamicsā. After this, Iād like to work on personas and explore the use of methods from evolutionary biology to the selection and evaluation of LLMs. Outside of my PhD research, Iāve recently collaborated with UC Berkeleyās CLTC on cyber risk thresholds, as well as with SaferAI on risk modelling through SPAR.
Looking to meet:
People working on cooperative AI (including, but not restricted to, those associated with the Cooperative AI Foundation)
People at UK AISI working on multi-agent systems or risk modelling
People modelling loss-of-control dynamics using agent-based approaches
Roles of interest: An internship/fellowship in Spring/Summer 2027: MATS/ERA/Pivotal/similar, or directly at an org if I could find the right fit/funding.
Contact: jameslpsykes@gmail.com
Alex Chambers-Jones
Bio: I'm a British Physics undergraduate with a programming background and an interest in forecasting and strategy; I'd describe myself as a generalist. I previously researched AI translation, and I've spent the last year competing to build better AI forecasting systems; I'm in the top 5 in the Metaculus FutureEval tournament.
I've been following developments in AI safety since 2022, and since the Huggingface incident I've decided to transition as quickly as possible. I recently started a research project with James Sykes measuring safety-task sandbagging via activation steering.
Looking to meet:
Forecasting AI (e.g. Forethought, AI Futures Project)
Orgs working on alignment research, evaluations, etc. (e.g. Artificial Analysis) ā I want to know where I can help given my previous experience (lots of work on LLMs, but no low-level technical ML/interpretability stuff)
Roles of interest:
Empirical safety research, particularly evaluations
Forecasting and strategy, particularly positive AI futures
Contact: alexcj2802@gmail.com
Jeremy Huff
Bio: I am an economics student with a public policy background who is interested in AI policy, regulation and governance. Would love to meet anyone who does more macro-level safety research and strategy.
Looking to meet: METR/Redwood/Encode, high-leverage AI safety organizations and policy.
Roles of interest: Interested in seeing orgs like METR grow into third-party auditors, and would love to apply policy/law lenses there.
Contact: jdhuff2048@gmail.com
Charlie Freiberg
Bio: Charlie Freiberg shapes the stories that make you known, understood, and chosen first. A narrative + communications strategist for founders and leaders, she translates innovations for the people who matter most, so you grow.
She is the cofounder of STUDY HALL, a creative building series where all disciplines play together with frontier partners (Plastic Labs, Collab Fund, Columbia University). She produces Story Is the Enzyme (raid, Edge Esmeralda), a talk series on how stories teach society to use tech responsibly ā and Storyhack, a workshop for scaling your brand and business through self-discovery. She publishes xx, an advice column (š) on finding whatās human while making the future.
Looking to meet: Comms leads from any and all orgs ā even just for field building!
Roles of interest: Anthropic ā unrivaled comms work at scale! Anyone across lab comms, and startups growing hard-to-explain things. Those running comms in safety (Seismic, METR, Apollo etc.). As a meta, anyone bringing the practice into the future with AI-native workflows.
Contact: charliejfreiberg@gmail.com
Daniel Hines
Bio: Iām a software engineer with a career in hard tech: counter-drone electronic warfare at Anduril, MRR spectroscopy at BrightSpec, blockchain protocol design at Marigold. Iām a full stack engineer (lowest-level firmware all the way to user-facing apps), and thrive in scrappy teams building something new. I use functional programming (Haskell, OCaml, Rust, Nix) and type theory for high-assurance software that canāt afford to fail.
Iām most excited about theoretical alignment research (agent foundations, deep learning theory, etc.), technical governance mechanisms for pause/pacing (e.g. verified inference) and LLM content detection.
Looking to meet: Orgs in theoretical alignment (met folks at Resolution and one guy at Iliad, but thereās probably more people/orgs I could meet). Iām also interested in formal methods for AI security. I think I could slot in as a generalist or research engineer into a lot of orgs.
Roles of interest: Any role at Resolution or Iliad (probably research engineering or generalist software engineering). Iām interested in meeting folks looking for technical cofounders.
Contact: d4hines@gmail.com
Thatās everyone!
If someone above made you think āoh, they should talk to so-and-soā ā it would be lovely if youād send a quick email! It can be two lines.
And if youāre transitioning into AI safety and want to be featured in a future connections post, fill out this form! I might do this again, depending on how well things go.
Finally: thanks to Harry and Josh for a lovely time. You can read their write-up re: Context Week here.
What an impressive group of people. It is reassuring to see that there are people like this with so much talent and such a range of skills wanting to transition into AI Safety or move to more impactful roles.
I wonder ... if these 10 people, or some of them, were to get together, maybe recruit a couple more for specific roles, what would they do that might make them eligible for a Tailwind grant from Coefficient Giving.
For example: right now the biggest question in AI Safety is about how to align the US and China on slowing down - which involves a lot of strategy and influencing and policy, but also some technical questions about verification, and communication.
Hoping against hope that the GOP gets trounced in the mid-terms, by January there should be a chance to have serious discussions with China despite Trump. What might a group of people like this do that would enable that or give it the best chance of success?
You have people from US / Europe / UK / China. You have people interested in policy and people with technical experience. Even if it were just a short-term collaboration, it could be a stepping stone into other AIS fields.
Context here: I see so many people fighting for very few roles in AI Safety. And at the same time, there is a lot of funding for new AI Safety orgs. So what does it take to put a group of very talented, passionate people like this and turn them into a funded AI Safety org working on a critical problem?
IMHO (but I could be wrong), there probably are gaps for people with generalist / org skills - how to manage budgets and how to set up companies and how to coach / train and how to lead long-term research projects - but there are lots of generalists like these looking to work in AI Safety - maybe instead of the usual model where you start with generalists and recruit the technical people, there's a way for the technical people to set up, get funding, and then go recruit some generalists ...
(if people are interested, I'm broadly happy to help with funding applications, a very tedious but critical tasks, maybe the bottleneck).
yooo jack my goat too
edit: this seems like a very useful thing for a platform like wat Denis is saying where you have a database of people trying to work in this field, their skills, seems like something very useful for project team leaders and stuff if you just automate the search, I wonder if this already exists and it kind of does with Linkdin but at the same time, I don't know if there is much guarantee of quality on linkdin it makes me want to make an EA app for employers where I describe people just as you have like match attacks or a fantasy football team for team assembly in a project, wht do u think?