Atlas Computing updates: forgotten Q1 update
Evan Miyazono
Hello Atlas community,
I opened the doc to write my start-of-Q2 update and realized I forgot to finish and send my start-of-Q1 update 😨. Our last update was in October). I didn't update tenses, so forgive me on that (or just wait 2 weeks for my next update)!
Project updates
Most exciting: We got a grant to prototype tooling for specification generation and validation (It hasn’t been announced yet, but here’s a sharable copy of part of the proposal). We’ll be focused on prototyping tools for formalizing natural language descriptions of cryptography, but I’m hoping we’ll be able to test out the framework for autoformalizing other types of policies as well. Sign up for our blog or follow our github if you want intermittent updates and demos, though, as we’ll be sharing progress there.
Events: We’ve been scaling up events. That started with helping to organize a small workshop in Berkeley early December on scaling proofs with AI. But now we’re also supporting a workshop to advance field-building for Guaranteed-Safe AI on March 7-9 in the San Francisco bay area, and we’ll be organizing a workshop on AI for Formalization in Ottawa in May next to ICSE. If you work on either topic and you’d like to join, please reach out to he...@atlascomputing.org. I can’t promise we’ll have room, but would love to hear from you.
Growth:
We’ve hired Jason Gross to join us for the 10-month period of the grant! We’re really excited to have him working with us on the aforementioned grant.
We’re spinning out our FlexHEGs work as Earendil, LLC to continue the work given funding from the Survival and Flourishing Fund’s FlexHEGs process. Evan will continue as a senior advisor to the effort, and our main efforts will consist of developing technical design specifications and cleanroom recipes for security layers.
Travel updates
Upcoming events: catch us in person or let us know if we should meet anyone
Engage more
We’re looking to form an industry consortium, gathering industry leaders to discuss their issues with AI unreliability who might be interested in scaling and automating their review processes. Please email me if you know someone who might be interested.
Anyway, thanks for reading!
- Evan
p.s. What Evan’s reading…
I’ve been really into scifi again (also was on a lot of flights) this quarter
On some strong recommendations, I read Permutation City by Greg Egan. This was a delightful and trippy exploration of the hard problem of consciousness and possible implications of whole brian emulation/uploading, but I think I might have liked Diaspora better. I bought and am reading a collection of his short stories and looking forward to reading Distress, which is now on my shelf.
This quarter I also read both Player of Games and Use of Weapons by Iain M. Banks. These seems like a much better introduction to the Culture Series that Consider Phlebas, which I don’t tend to recommend. Some people have pointed to the world of the Culture Series as how they want the world to look post-ASI, and I think it’s a perfectly reasonable outcome (perhaps even desirable in a “humans as autonomous pets” kind of way), but the path we’re on does not seem like it’s headed in this direction. Interesting world-building for both; great story-telling for UoW but I wouldn’t recommend without reservations
Phoenix Project - I found this to be a surprisingly engaging, stressful, and sometimes delightful nonfiction about operations (specifically dev-ops). I’m planning on reading The Goal by Cox and Goldratt because I understand this is a modernized, devops retelling of that book.
I also read Moonbound by Robin Sloan and Let’s Explore Diabetes with Owls by David Sedaris, probably wouldn’t blanket recommend either.