Crossposted on my personal blog. This is post number 16 in my second attempt at doing Inkkaven in a day, i.e. to write 30 blogposts in a single day.
MATS is an organization that pairs up-and-coming AI Safety researchers (who I call participants) with the world’s best (this is not an exaggeration) existing AI Safety researchers (called mentors), for a minimum of 3 months research experience, followed by 6 or 12 months of further time to pursue their research further if they meet a minimum standard.
The most common role at MATS, called research manager but I prefer the term research coach, is all about providing 1-1 support to the participants. The participant-mentor relationship is purely based on the research: by default they meet weekly for 30 minutes and only discuss what research has happened, and what research tasks to tackle over next week. The research coach works with the participant on literally everything else, which is very broad. Some examples are accountability (e.g. for the research goals, other non-research goals that the participant sets like applying to jobs), interfacing with MATS (so that MATS can track patterns or engagement of participants), people management (e.g. helping with any interpersonal conflicts, or, helping them make the most of the limited 30 minute time slot with their mentor), career planning, general life improvements (a common one is sleep), …
What do I like about research coaching?
I like to be a jack of all trades and research coaching exposes you to many different skillsets. It has been great to flex and improve many different skills.
I like to learn about many different research areas, rather than going deep into one niche sub-sub-question. Working with various participants allowed me to do this.
I fundamentally like helping and teaching and coaching people, so the role naturally fits my personality here.
I do not enjoy the process of doing research myself. I do not inherently find software engineering satisfying and I dislike all the infra stuff. (Looks like claude code is almost good enough that I can just ignore all that, so maybe one day I will do research via coding agents.)
What do I like about MATS. This list is long, and yet there is a high chance I have missed some important considerations.
Socializing with the (vast majority of) staff and participants. Chatting and socializing with the people is great pleasure and likely the biggest reason I like MATS. When I first joined I imagined going into the office 2 or 3 days per week, but then quickly just went every day.
Learning from the (vast majority of) staff and participants. Both the staff and participants are mega impressive and skilful, and there is tons to learn from them.
MATS is a central organization of the AI Safety ecosystem, and its importance will grow with time as it is growing fast. It has connections with most, if not all, the major AI safety teams and organizations in the world at the moment, and a high percentage of these teams and orgs are staffed or founded by MATS alumni.
MATS explicitly has these four values: scout mindset, impact focus, transparent reasoning, and servant leadership. I am huge fan of the first three, and somewhat dislike the fourth because it is too wishy-washy and corporate sounding.
A downside of MATS is that both organizationally and on an individual level, there are not high incentives to actually follow the values, and (in my opinion), most/all staff fall short of meeting the standards implied by these values. Nevertheless, just having these values as a north star is still inspiring and guided a lot of my thinking and actions.
MATS explicitly has a culture of voicing one’s thoughts honestly and openly, including things you are unhappy about in MATS.
MATS is a largely a ‘do-ocracy’. If you have a good idea or find a way to improve things, you are encouraged to go ahead and do it. Various initiatives and improvements start off this way.
MATS is a growing fast, so there is lots of opportunity to contribute and shape how MATS grows. At the time of writing, I actually think this is the highest impact thing one can do in MATS - not the direct research coaching - and something I found highly satisfying.
For the London office only and as of writing: it is based in the Fora Central Street office, which is a fantastic space to be in. Furthermore, you get free access to all the other Fora offices around London (there are around 50).
MATS is a fun place to work. I only speak of the London office, but there is weekly brunch to a nearby cafe on Thursdays, have team shoutouts during the Friday morning standup, lunchtime lightning talks, activities organized on semi-regular basis e.g. there was recently a trip to play table tennis in a local sports center, a piano in the office to allow for music nights, various board games in the office, etc.
MATS is (mostly) a high trust environment. After I had hypomania, I felt comfortable telling the team what happened, rather than keeping it to myself or to the one or two people I trust the most.
MATS takes mental health seriously. Though I did not do anything I regret, in the week after the hypomanic episode, I was taking more and more actions which were riskier than I would normally take, so there was a small risk I would do something I and MATS would regret. Hence, the London team lead intervened (in a highly professional and empathetic manner), and offered two weeks paid medical leave, followed by gradually coming back to work on a part-time basis (again paid full time). This provided time to properly stabilize, ensure I get professional help I need, and also gave me time to improve my life in many ways (e.g. this is why I had time to organize so many events for my birthday).
The pay is great, at least compared to the vast majority of jobs out there. Small compared to what I could get if I optimized purely for total cash (e.g. working in big tech, frontier AI lab or finance), but otherwise excellent. For example, the income made it straightforwardly easy for me to spend £1800 on a piano as a gift for myself, and to still have most of my income go into savings.
Of course, MATS is far from perfect, but that is true of any organization or group of people. I am just about wise enough not to air my dirty laundry in public, but, given the MATS cultural norms I describe above, I did feel comfortable enough to write a detailed memo with my highest level concerns and speculative solutions. It remains to be seen whether the memo sparks the dramatic improvements that I think are possible and necessary, but even if not, MATS is an organization that is hard to beat.
Crossposted on my personal blog. This is post number 16 in my second attempt at doing Inkkaven in a day, i.e. to write 30 blogposts in a single day.
MATS is an organization that pairs up-and-coming AI Safety researchers (who I call participants) with the world’s best (this is not an exaggeration) existing AI Safety researchers (called mentors), for a minimum of 3 months research experience, followed by 6 or 12 months of further time to pursue their research further if they meet a minimum standard.
The most common role at MATS, called research manager but I prefer the term research coach, is all about providing 1-1 support to the participants. The participant-mentor relationship is purely based on the research: by default they meet weekly for 30 minutes and only discuss what research has happened, and what research tasks to tackle over next week. The research coach works with the participant on literally everything else, which is very broad. Some examples are accountability (e.g. for the research goals, other non-research goals that the participant sets like applying to jobs), interfacing with MATS (so that MATS can track patterns or engagement of participants), people management (e.g. helping with any interpersonal conflicts, or, helping them make the most of the limited 30 minute time slot with their mentor), career planning, general life improvements (a common one is sleep), …
What do I like about research coaching?
What do I like about MATS. This list is long, and yet there is a high chance I have missed some important considerations.
Of course, MATS is far from perfect, but that is true of any organization or group of people. I am just about wise enough not to air my dirty laundry in public, but, given the MATS cultural norms I describe above, I did feel comfortable enough to write a detailed memo with my highest level concerns and speculative solutions. It remains to be seen whether the memo sparks the dramatic improvements that I think are possible and necessary, but even if not, MATS is an organization that is hard to beat.