Updates
FORMINDEX_10_13_2024
Daniel Friedman
0000-0001-6232-9096
October 13, 2024
Situation
In it was identified that the early build of could successfully read in & visualize the FORMIS export Bibtex format. Sanford sent an initial test set of citations, and then the full July 2024 FORMIS export.
Here is the sample of 316 citations
FORMIS 2024(July)-Bibtext_316_sample.txt
558.3 kB
Here is the full Bibtex export
FORMIS 2024(July)-Bibtex.txt
98.6 MB
Here is the reformat of Bibtex into JSON.
FORMIS_2024_July_Bibtex.json
99.7 MB
The goals of this phase of FORMINDEX were to:
Methods & Results
Methods & Results are separated into two main sections, reflecting the two goals above.
The first section “FORMIS analysis” describes the initial intra-FORMIS analysis methods performed & their location in the open source code.
The second section “Generative AI methods” describes three applications of Generative AI methods.
FORMIS analysis
Ingress and Target Bibliographies
TARGET_BIBLIOGRAPHIES = [
"Formica", "Camponotus", "Myrmica", "Pheidole", "Pogonomyrmex", "P. barbatus", "Monomorium", "Messor", "Cataglyphis",
"Foraging", "Bioenergetics", "Myrmecophiles", "Ant-plant", "Dopamine", "Serotonin", "Art", "Robotics",
"Hawaii", "India", "Mexico", "Brazil", "California", "Intelligence"
]
Targeted bibliographies are output to:
Visualization
Distribution of record types
Histogram of record types
Top authors
Publications per year
Co-authorship heatmap
Top venues
Venues through time
Top locations
Distribution of record types
Histogram of record types
Top publishing venues
Top authors
Author co-authorship heatmap
Publications per year (not assumed to be correct/total number, especially after 1996, for certain languages, etc).
Top locations of publication
Generative AI methods
Three Generative AI methods were used: NotebookLM, OpenAI,and Perplexity.
NotebookLM
Google’s NotebookLM was used to produce conversational podcasts on target sub-FORMIS-scale bibliographies. The target bibliography in JSON format was uploaded to a new Notebook. Then unless errors arose, I clicked “Generate Podcast”. RevidAI was used on the audio files to add video & captions.
Several short (~10-20 minutes) podcasts for target sets of FORMIS citations, are hosted on my YouTube channel.
OpenAI API
OpenAI LLM API was used for summarization and translation, all methods in
The total LLM cost, using GPT-4o-mini, per summary & translation depends on the length of input and output, probably on the order of one-tenth to several cents USD in October 2024 (they are cheaper/different methods as well)
For example, here is the Myrmecophile targeted literature summary, translated into Hindi.
Perplexity API
LLM API was used for internet-enabled Myrmecological augmented inquiries.
Prompts are listed here:
The short_name will be used in the output file naming
iterates over Prompts. For now, enter your Perplexity API in plaintext at line 17 in order to use the script.
There is customizable system prompt:
captures the inputs, outputs, and timings. It takes about 10-40 seconds per prompt, and the number of returned lines is ~20-200
All outputs from Perplexity are in
Just to give 3 examples:
Next steps
..-. --- .-. -- .. -. -.. . -..-
Want to print your doc?
This is not the way.
This is not the way.
Try clicking the ⋯ next to your doc name or using a keyboard shortcut (
CtrlP
) instead.