↓ Skip to main content
Chemical Cognition Lab
Chemical Cognition Lab

6. Computing & Infrastructure

Last updated:

Tools
#

  • Zulip for chat
  • Google Docs and Google Sheets for writing and planning; Microsoft Teams is also available through U of T
  • Zoom for meetings, sometimes Teams
  • Go-links for short links
  • Linux, VS Code, and Jupyter notebooks for coding
  • Mamba for Python environments

Setup and clusters
#

Our hpc_setup repo covers the first things to set up on your laptop, Python environments, and how to connect to Canadian compute clusters such as Balam, Killarney, Trillium, and others. New projects start from the project template.

Agentic coding
#

Agentic coding tools are good, and I encourage you to try them. They will keep developing, and we’ll be using them. Our agent skills repo collects the skills we’ve written for them.

Agentic workflows work well when you already know what you’re doing. The tricky part is that during your training, you’re still learning how to do things, so make sure you still understand how the underlying pieces work.

Be careful not to accrue too much technical debt with agentic workflows. Debt is useful when it lets us move quickly, but you have to pay it back eventually (Martin Fowler explains it well), especially if the codebase grows faster than your understanding of it.

Generative AI
#

The main thing to ask yourself is whether you understand your work. Sometimes the only way to understand something is to build or write it yourself.

It’s easy right now to produce a lot of output without really understanding it. Everybody wants to be productive, but you are responsible for any code and writing you put your name on.

  • For code, use AI freely, as long as you read, understand, and test what it produces.
  • For writing, AI is fine for polishing and brainstorming, while the ideas and arguments should come from you. Follow each venue’s disclosure rules, and never use AI to generate citations.
  • For confidentiality, don’t paste unpublished data, collaborator or industry material, or manuscripts you’re reviewing into tools that may retain them.

When you use AI for writing, think about who’s reading. If it’s a person, people tend to like shorter, more concise writing. If it’s a machine, length matters less.

Be careful of the AI aesthetic, in writing and in visuals. It’s very bland, and readers and reviewers can judge it unfavorably. See The AI design aesthetic that’s taking over the internet in The New Yorker.

U of T’s AI guidelines apply to all of us.

How I think about AI is close to how Terence Tao frames it for mathematics (paper, blog post) and how Ted Chiang writes about it: Why A.I. isn’t going to make art and his conversation with Ezra Klein (summary). Also worth reading: No, artificial intelligence is not conscious in The Atlantic.