added gdr paper

add swallowmathv2 finding

reformulate to make formulation about prompts from beyondweb clearer

add more prompts

improved tab styling

remove essentialweb since it is not trained for long enough

add some more related work

added speedup annotation for main plot

combined throughput visualization into one file and fixed counting bug

fix trinity links

updated synth results in the table

add more analyses for specific benchmarks

improved colors

add intro plot again and compare to synth baselines

reformatted experiment takeaways into table

changed throughput visualization in infra section to comparison

put summary notes into context

removed figrefs and changed the setup for charts: first set the scene, then show the plot and finally analyze the resutls and do the transition.

switched from accordion to newly created tab widget

improved transitions

added essentialweb description

combine two sections for brevity

add throughput visualization to intro too

renamed chapter

move model collapse to intro

improved phrasing and banner

removed reading time component

made some more improvements to the inference throughput illustration

updated numbers based on the final finephrase run

improve inference throughput visualization

updated synthetic only results

tried to prevent misunderstanding in the main plot

moved the plot as a hook in the begginning

made the writing style more engaging and improved transitions based on Lewis' feedback

reordered finephrase section to end with the dataset

improved the infrastructure section based on Lewis' feedback

improved textual flow with sidenotes

add reading time estimation per chapter

added additional next steps

updated texts based on new results

add some concrete examples

added finephrase section and moved progress monitoring section there from infra

add new finephrase section

made sure prompts are always alphabetically ordered within categories

add new results

add todos and rephrased conclusion paragraph

add baselines mixed with fw-edu-hq

add mercury to conclusions

added article and discussion results

made inference throughput visualization more elaborate