CJ Blunt

Teaching & AI

Higher Orders of Evidence

1 minute read

In a recent post, I presented the vision of an evidence hiearchy according to DALLE Mini (now ‘Craiyon’). That model is a miniature equivalent of the much more powerful GPT-3 derived DALLE 2 image generator, which takes a text prompt and generates 1024x1024 images to match.

Recently, I was able to get access to DALLE 2, and put some of the same prompts to it. The results certainly showcase the power difference between a relatively small model like DALLE Mini and the full-fledged GPT-3 iteration. Here are the results for ‘Hierarchies of evidence in evidence-based medicine’ and ‘Evidence pyramid in medicine’ from DALLE 2.

DALL·E 2 image for the prompt ‘A hierarchy of evidence with tendrils coming from the bottom reaching deep into the foundations’

DALL·E 2 image for the prompt ‘A set of marionette strings which branch out at the bottom and grow into cracks in the ground’

DALL·E 2 image for the prompt ‘A pyramid made of medicine’

Image generated by DALL·E 2 for one of the evidence pyramid prompts

DALL·E 2 image for the prompt ‘A photorealistic Great Pyramid of Giza but made of medicine’

A second DALL·E 2 image for the prompt ‘A photorealistic Great Pyramid of Giza but made of medicine’

More in Teaching & AI

  1. Cultural Bias in Language Models: The Top 10 Test

    How can AI-generated "Top 10" lists of cultural influencers expose the cultural biases and defaults that are built into our most popular language models? I propose and report an initial test using lists of the Top 10 most influential musicians.

  2. Golem.AI: The Experimenter's Regress and AI Decision Systems

    How can the concept of the experimenter's regress help us understand the problems in AI decision systems? I explore two forms of circularity that can underpin AI decision tools through the lens of Collins and Pinch's 'The Golem'.

  3. NotebookLM: Cast with Care

    Google's NotebookLM "deep dive" feature is taking off in popularity. I subject three of my academic papers to the deep dive treatment, and reveal its tendency to subvert content for a happier ending.