Edward Kmett

Fellow, Head of Technology and Architecture @ Groq, Inc.
72 karmaJoined Apr 2022Working (15+ years)Farmington Hills, MI, USA


Formerly at MIRI. I'm currently Head of Technology and Architecture at Groq and I've been helping on the boards of Topos Institute and the Haskell Foundation where I can. I'm more of an anti-malarial bed nets style effective altruist than a hardcore longtermist, though I have strong concerns about AI X-risk, and feel I have more to contribute on the latter front than the former.

How others can help me

Help me find out how to help you.

How I can help others

I happen to write a lot of Haskell. I'm also pretty well plugged into what is going on with the current and next generation worth of AI accelerators. I'm a passable category theorist. Please feel free to reach out to me with questions on any of those fronts. If you need a board member or adviser or just a critical ear and feel I might be a fit, please feel free to reach out.


In my experience from actually renting  spaces to run tech conferences and events, a convention hall in a reasonably dense city runs in the tens of thousands for a single <5 day long event.

Consider just 3 breakout rooms, typically ~$90-$180/room/hour, 8 hours a day, 5 days takes you into the $20-$40k range, that gets you no food, no lodging for attendees, once you add that stuff in and the all-in cost for a fully-paid for typical 30-40 person event can easily land just inside $80-100k, with costs growing slightly sublinearly per person for a while before they balloon again once you rise above the size where there are multiple competing venues in your area.

If the convention center you are using isn't a hotel now you have to move your attendees back and forth.  For tech conferences you can charge a few hundred $ per person, and have them book with your hotel out of a bloc, incurring obligations from your org to a hotel about occupancy, and then you can generally come up with a way to make your event break even. Conferences are expensive for a reason. 

Locating an event in the middle of nowhere? Maybe that room becomes maybe a fixed rate $350 space at a cheaper hotel, $1000/day for 3 becomes so $5k for a 5 day event, but now its harder to get folks to the venue.

At the end of the day, in either of these scenarios you are subsidizing the hotel and event space industry rather than focusing in on your events.

On the other hand, using the model of a typical CFAR event, using a venue that is already owned, you are dealing with some cleanup, a couple of members of the team providing operations support, catering costs, but the venue itself remains relatively customized to purpose. People can sleep on site, enabling folks to drift in and out of conversations, the event can run all night with people talking in smaller more intimate groups long past when a hotel would have kicked you out of the space. If your goal is to built a community and have folks come together, this is a much better model.

If you can keep this above 50% utilization, this even strikes me as way more than break-even or even if I'm delusionally out of touch and hotel prices have some how plummeted in the last 3 years while the cost of everything else has gone up, at least far closer than the 10-100x swing you are arguing for here.

Lightcone, by comparison, has been getting about 75% utilization out of the home it has been using as a conference space.

On the flip side, it seems the Abbey is likely to require less in the way of upfront renovations, which seem to me to be the significant at-risk part of the Inn, because the sorts of changes you're likely to want to make it a good venue for running workshops probably make it an objectively worse 'hotel', which is what almost anyone else would be buying the space to get.

It seems the two are within a factor of 2 of each other for long term cost, most likely in your favor, while they may have you beat a bit in terms of non-recoverable conversion expense.

In the end it is surprisingly close, all things considered, given quite how different Oxford and Berkeley are.

The Rose Garden Inn is even something at a comparable price point to pressure test against. As in it is the same ballpark general distance to most of the potential users, roughly the same price, within a factor of 2 in room count, etc. but way more run down, and as recent breakins have shown, though perhaps way more vulnerable to people just walking on premises and stealing construction materials as they work to fix it up.

I do think the Lightcone example is a large part of why I'm not up in arms about this. They've demonstrated in their existing somewhat smaller spaces that with the spaces they have they get ridiculously high utilization: ~75% utilization of a house they use for events, packed guest spaces, overflowing the office, etc.

At the end of the day there's a reason why real estate is considered an investment. None of the uses folks are going to put it to as a venue are going to appreciably damage that investment, and from an annual operating expense perspective, it is likely significantly cheaper than renting office/retreat space, and putting folks up in hotels.

There's something to be said for going somewhere and staying immersed in what you are doing, especially when its most likely cheaper than the more disruptive alternative.

The lack of a 2 hour commute is nothing to sneeze at though. CFAR has (had? I haven't checked in on it lately) a venue a couple of hours away from Berkeley that they've used for organizing workshops and events, and the tribulations of organizing getting everyone to and from the venue pretty much ensured it was only used for running 4-5 day events. It made it significantly more difficult for folks at CFAR or MIRI to pop up to make "guest appearances" at workshops and the like significantly reducing value to participants. 

Speaking from personal experience, that distance rather complicated the value proposition for me, for whether it was worth showing up for a day at the end of an event to get to know some of the participants.

At the end of the day, the optics seem poor, but the actual cost for the space seems to be what I'd expect for a space that can sleep that many people, zoned so you can use them as actual bedrooms and have people stay on site. By the time it is kitted out in proper group-house density with beds in every nook and cranny you can find, you'd be able to fit a rather large number of attendees into the space.

You can go grab some concrete office space in an industrial park somewhere, but at the end of the day you generally can't legally have people sleep in that office space -- no matter what Elon is trying to do with Twitter HQ this month, so you'd wind up needing to sublet nearby apartments and the like for attendees, assuming you can find ones that legally allow you to do so, or pay premiums for hotel stays.

Part of why the CFAR venue wound up as far outside of Berkeley as it did, was they literally couldn't find any place closer that would legally let them treat it like a bed and breakfast for hosting attendees. 

https://www.rationality.org/resources/updates/2017/cfar-2017-venue-update is an older post describing the rationale for purchasing the space I mentioned above.

Habryka and the rest of the Lightcone Infrastructure team seem to be wrangling the same sort of considerations as they try to provide gathering space for EA and rationalist folks in the bay area, today, except there a roughly equivalent amount of funding doesn't buy anything like a fancy-looking but run-down historical abbey.