I collect links and share a few that I find interesting on Elder Research’s Slack during the week.
Pen and Paper Exercises in Machine Learning
With all the focus on LLMs, I’m more grateful than ever to find resources like this (from 2022), helping those who really want to understand (some of) the guts of machine learning to do so. This seems like a useful instrument for deliberate practice, which is something I’m not totally sure how to do well in the analytics/machine learning field.
Tracing the thoughts of a large language model
More interesting work from Anthropic, diving into the guts of these LLMs to try and understand what their computations look like more or less in vivo.
There’s an interesting duality here: these models are simple at the fundamental level, but the scale at which these simple operations are combined and processed make their operation very difficult to understand at a higher, often more useful level.
Introducing 4o Image Generation
I began paying for ChatGPT Plus recently to get a better sense of when and where these chat interfaces are useful to me. This update is a remarkable improvement over my previous experiences (presumably with DALL-E). I haven’t found many applications for generated images yet that aren’t over the top or in poor taste, but if they turn up then this looks like a good solution.
“Create a stylized, fanciful image of a machine learning model training on data so that it can make future forecasts.”
Related Posts
- Analytics Link Roundup, March 10–14, 2025 (March 2025)
- Analytics Link Roundup, Feb. 17–21 (February 2025)
- Analytics Link Roundup, March 17–21, 2025 (March 2025)
- Analytics Link Roundup, February 24–28 (March 2025)
- Analytics Link Roundup, March 3–7, 2025 (March 2025)
- A Few Points on Generative AI (July 2023)
- Multi-GPU Evaluation Loss with Detectron 2 (June 2020)