Goals

Decrease time from feature release to public knowledge of feature

Increasing utility of tools to academics - gives more heft via publishing - increases visibility (how do we get this? academic groups already like open source things, could make examples and docs more visible/discoverable, and ensure that the initial installs)

Improve discoverability, esp. of examples

• Social media blitz, conference talks, virtual workshops

• Automate/calendar-schedule social media blasts

• Could ask on social media which examples they want to see

• Encourage users to spread the word, or submit their notebooks to anything-goes

• Do an AMA of some sort? Reddit/Twitch/Zoom office hours

• University visits/workshops

Add examples that are relevant

• Some heavy manual steps here, lots of consideration

Evolve examples/features to meet user needs (ex PTM concept evolution from workshop to repo)

Some sort of regular project updates - newsletter? publish ad board updates? AI meeting summaries?

How do we do product differentiation? Ex espaloma vs. nagl? Is bespokefit an openff product.

• Bespokefit is a little different from other OpenFF projects in that we’re not open to new features/contributions, whereas we’re open to contributions/new capabilities in most other packages.

Confusion about purpose - Feedback on the landing page has been that folks don’t have a clear idea of what we do - we should have a thing that says “we make force fields”

(Who maintains our website? JW: Let’s say that the infrastructure team does)

How can we make meeting notes more accessible to outsiders?

• AI summary? Mixed feelings about accuracy.

• Who is this summary intended for? Maybe somewhat infequent meeting attendees who need to occasionally catch up.

• Could publish monthly ad board summaries

• Could do 1 paragraph AI summaries as part of a newsletter, and the paragraphs could link to the meeting notes.

•  

Channels

Internal

Slack

Confluence

Recurring meetings

External

Website

Github

Docs

Workshops

Conferences

Bluesky

LinkedIn

Blogs

Youtube

Problems to address:

•  

• What are we trying to accomplish?

  • MT – Recall that JW told a person at the in person meeting that we can now load PDBs. We should look at improving this ratio of staff time to external person time.

• How do external but interested parties find out status?

  • Meeting types – a central list of all science meetings, purposes, channels, notes and agendas, somewhere public

• Product differentiation

  • Espaloma vs NAGL

  • general collaboration research work vs OpenFF’s

• OpenFF website is very old

  • Need to cut out a lot of cruft

  • It should primarily be a map to our online footprint that explains and contextualises how pieces fit together

  • From the IRL meeting, good place to ask what to keep

  • Ideally: update site once a quarter

  • A splash page that explains all parts of the project and links to all relevant resources sounds like a good idea, but what we have is an approximation of this that is very hard to keep up to date.

• Confluence

  • KCJ put a lot of work into structure, but now no one uses it

• Example notebooks

  • Barrier to getting people to find it

  •  

Blog Post Ideas

• NAGL

  • How you can use nagl to make your life easier - LW

  • Why we need nagl - LW

  • Use AI to set up your molecules for simulation! - JE

• How to win at conferences

• What does it mean for a force field to be self-consistent? - CC

• Why do we need a general force field?

• How hard is it to train a protein force field? - CC

• Is a molecule a molecule?

• Why do we need to track stereochemistry?

• Why all of our molecular file formats are terrible and (we shouldn’t make another/we should just use Interchange instead)

• Why you should never (in MM-land) infer a molecular structure from elements and 3D coordinates alone - JuH

• Types of thinking you can’t do on 6 hours of sleep

• The cost of free in open source - JW

• Other fruits are expensive but bananas are cheap

• Why are physical models relevant in the face of machine learning?

• If someone else made the world’s best force field using our tools and methods, we still win

• What we’ve done in the past x_ - JE

• Boston is too expensive

  • Let’s have conferences in Charlotte instead

• Bad AI/ML models

  • We have the tools to discriminate between good and bad models

• Leaked meeting transcripts

• Areas of opportunity

  • What are we NOT good at?

• OpenFF tools are designed to be used (and expanded upon)

• My favorite github issues - PB

• Lily’s improper dihedral rant

  • the value of collecting scattered information in a single location

• An apology for aromaticity

• Savage self-criticism

• Why we should drop (graph-based) bond order and no longer use aromaticity - DM

• What happened with the WBO project (and is it still a good idea?)

  • When is it time to pull the plug on a project?

• Cheminformatics can’t comprehend transition metals

• We make no warranty but we do provide (some) support

• When to deprecate and how to maintain trust - MT

• The importance of open source scientific software - BM

  • See http://aip.scitation.org/doi/abs/10.1109/MCSE.2009.19 which is a paper version of some of the key ideas (DLM) ← Thank you! (BM)

• How is our project structured and funded? - JE

• How does a grad student work with OpenFF? - AF

• Python dependencies suck - TG