
Research software engineer @ UCL. previously a chemist at Oxford, Bristol. #openscience
London
Joined September 2017
- Tweets247
- Following332
- Followers455
- Likes943
I was moderately worried about the future of AGI. youtu.be/VcVfceTsD0A did nothing to help
having spent all of 5 minutes thinking about it: the AI alignment problem seems too difficult. how do you build a system with empathy?
967
Open sourced an hour ago and there’s already some spicy PRs 🌶
Replying to @Engineering
Ready to look under the hood? Check out our repos on @GitHub: github.com/twitter/the-algor…
1
3
1,778
Writing for web with (almost) no javascript is awesome. Molecular visualisation in rust 🦀: t-young31.github.io/mol-wasm… (from a .xyz file), UX needs some work to say the least..
code: github.com/t-young31/mol-was…
3
18
3,033
Just as we hit 100k downloads (😲) autodE v1.3.0 is out 🚀
anaconda.org/conda-forge/aut…
Lots of improvements and a few bugfixes – thanks to all involved! Highlights include: a new TS search method, better bond definitions, frequency scaling and NWChem defaulting to autodE optimisers
I *will* have introduced bugs so please do create an issue (at github.com/duartegroup/autod…) when you find them – thanks
1
2
Without wanting to start a 🔥 war: what defines a bond?
-
Distance? Electron density? Dissociation energy?
prompted by github.com/duartegroup/autod…. Tagging @ASterlingChem @HarryWTMorgan1 who I'm sure have opinions..
Ever wanted a fast UFF optimisation code that's MIT-licensed and built in 🦀+🐍?
opt-rs (github.com/t-young31/opt-rs) has just been open sourced 🚀
(Very fortunate to be able to work on side projects like this as part of my role in @ucl_arc)
TIL (that I really should have known already): pow from cmath in C++ is abysmally slow for integer powers
3
3
It's my final day in this building... 4.5 years. masters→phd→post doc. It's been great and I'll miss everyone. Thanks y'all 🧡
6
75
A bit of a rush in the end but autodE v.1.2.0 is out (github.com/duartegroup/autod…). Huge thanks to everyone that contributed with code/knowledge/bugs. Some efficient-ish geometry optimisers are there, with some that need a little more attention...
2
5
I’ll be moving away from academia so a couple of other projects get to be open sourced inc. a small reaction kinetics package for a project a while ago might be useful for other people: github.com/t-young31/rksim . With hopefully one final thing on the way (@Tristan81946826)
Finally got around to training a NequIP potential and DAMN is it impressive. Congrats for the amazing work 🎉
This quoted post is unavailable.
1
9
going to be working on autodE development with the help of @OxfordRse, alongside some related projects
5
Looking forward to chatting about our (@Tristan81946826) work at #AIChem21
P33 Efficient construction of accurate interatomic potentials using machine learning
Tom Young, University of Oxford, UK #AIChem21posters
teapawt.net/RSC_CICAG/status/14395…
1
17
code: github.com/duartegroup/gap-t… paper: pubs.rsc.org/en/content/arti… scripts to generate the example in the poster: github.com/t-young31/gap-tra…
1
Further to our initial work we found that you can use MP2 forces in combination with CCSD(T) energies to avoid costly numerical gradients and using a maximum diversity selection scheme on a cheap QM active learnt configurations (github.com/t-young31/gap-tra…) to maximise efficiency.
1





