Exactly — you can control and monitor how weird it’s getting through the example replacement/editing. Also, sampling from a k-shot pool. All the diversity-promoting methods in TinyStories have equivalents here too.
RESEARCH
-
Parameter Size vs Performance in AI Model Graphs
By
–
also, since Kevin Scott didn’t afaik specify what’s plotted in his hypothetical graph, maybe it’s literally just parameter size and not performance lol
-
Model Size Transparency Gap in Recent AI Releases
By
–
Do we have firm data on size for all the recent models? not all have been fully forthcoming, as I recall.
-
Thread Demonstrating k-shot Prompting Method
By
–
I made another thread to illustrate the k-shot prompting method mentioned above:
-
Repeated sampling and k-shot prompting affect creativity in instruction-tuned models
By
–
Repeated sampling from instruction-tuned models can yield low-diversity, stereotyped output. For creative tasks, k-shot prompting base models will improve creativity (measured e.g. as vocabulary size), though generally with worse adherence to the (now implicit) instructions.
-
Every AI Model Since GPT-4 Looks Identical
By
–
Update: On this graph, literally every model that has come out since GPT-4 still looks pretty much like … GPT-4.
-
Papers and Patents: Measuring Research Activity, Not Scientific Progress
By
–
For instance, the number of papers shipped to arxiv (or patents filed…) is pretty much just a measure of how many researchers / students are active at a given time. It isn't a measure of the pace of scientific progress. It is a measure of *inputs*, not outputs.
-
GenAI Engineering Basics Course on Databricks in 2 Hours
By
–
Learn the basics of GenAI engineering in just 2 hours! This self-paced course will help you get started with GenAI on Databricks. Learn how to build a compound AI system, fine-tune models, drive AI governance, and more: https://
bit.ly/4fPUk6S -
Burkov vs Chollet: Conflicting AI Predictions with Mixed Empirical Results
By
–
so @Burkov was kidding, @fChollet was not; empirical data have been mixed.
-
Marcus Test Comprehension Challenge 2014 Performance Debate
By
–
excuse me —- what Marcus test was beaten? My main one was comprehension challenge 2014, certainly not beaten.