Choose Your Weapon: Survival Strategies for Depressed AI Academics This is a very timely paper that discusses how researchers in academia can keep up with the current pace of AI research. The authors present many ideas, some of them being: – Giving up on doing things that are
RESEARCH
-
Computing Efficiency at Scale: Useful Compute Per Joule
By
–
At scale, you hit limit of available power, so it’s just useful compute per Joule. The “useful” qualifier is important.
-
Direct Experience and Independent Research Find Deeper Truth
By
–
Your direct experience, people you talk to in the subject area & independent research will get you much closer to the truth
-
Understanding Instrumental Convergence and its long-term effects
By
–
Anyone who is curious about the long-term effects of AGI should familiarize themselves with the consequential concept of Instrumental Convergence. One of the best Wikipedia pages on AI
-
Linear Model Training with Sample Weights for Class Balance
By
–
didn't follow but sounds interesting. "train a linear model with sample weights to class balance"…?
-
Training SVMs on High-Dimensional Embeddings: Performance and Practical Considerations
By
–
yeah 😀 it's not as bad as it sounds, e.g. using the example in the notebook, training an SVM on ~10K 1536D embeddings is ~1 second. Sometimes it's possible to precompute. And sometimes it's just not worth it, all depends on setting / application.
-
How to Confuse Your Machine Learning Model
By
–
How to confuse your machine learning model. (credit: @teenybiscuit
) -
ML Research Community Validation Culture and Workshop Dynamics
By
–
And that scratches most of the "research validation" itch for me (don't lie, we all have it). My feeling is that other areas of ML don't have this type of community. More of a rat race, trying to get noticed. Focused workshops help alleviate this, but they can be infrequent 4/n
-
Research Community Loneliness: Building Networks Post-PhD
By
–
There's great communities for jr. researchers to get involved & alleviate this loneliness (eg @ml_collective @forai_ml @AiEleuther
) but may be harder for someone post-PhD, since they're supposed to be "independent." Is research community loneliness a problem? Is it fixable? 5/5