Month: August 2025
Lipton Soup As A Lesson In The Importance Of Clarity When Asking Advanced Models To Visually Catalog Video & Visual Hallucination
We recently spotted a brief appearance of Lipton Chicken Noodle soup in a 2023 television evening news broadcast discussing Boston's…
Using Gemini To Catalog 16 Years Of 24/7 Coverage Across Three Cable News Channels: 20B Tokens Yielding 3.15M Stories
Continuing our story segmentation experiments, we used Gemini 2.5 Flash Thinking to "watch" 16 years of three English language 24/7…
Using Gemini To Catalog A Year Of Ukraine TV News Coverage Across Four Countries Totaling 345K Stories For Just $1,190
There is great interest in helping journalists and scholars understand and cover the world's biggest stories each day. Yet, no…
Enriching Democracy: A First Glimpse At Cataloging & Annotating A Year Of The Legislative Process
Earlier this month we demonstrated an incredible first glimpse of applying our new story cataloging workflow to a handful of…
Gemini At Scale: Understanding Tokens In The Real World: A Small Batch Analysis Of 400M Tokens
What does it look like to apply LMMs like Gemini at scale? Specifically, given that "tokens" are the central currency…
Translating Television News From Across The World: Early Experiments Comparing Gemini Versus Google Translate
As we continue to scale up our experiments cataloging the underlying stories being covered across the world's television news coverage…
Combining Chirp ASR + Gemini To Story Segment Television News From Around The World
Continuing our series on segmenting television news broadcasts into their underlying stories, thus far we have explored segmenting only captioned…
Story Segmentation Of TV News: The Impact Of Input Chunking On Token Counts & Index Richness
We've demonstrated that input chunk size appears to have only a minimal impact on story segmentation, with larger chunk sizes…
Story Segmentation Of TV News: Assessing Segmentation Stability On Non-English Broadcasts
Continuing our television news story segmentation experiments, using 2-second chunking of our Chirp ASR, how consistent is Gemini's story segmentation…
Story Segmentation Of TV News: Impact Of Input Chunking On Segmentation Results
Segmenting television news broadcasts into their underlying stories has the added complexity of needing to transparently pass timecode information through…
Story Segmentation Of TV News: Adding Language Detection With A Single Sentence
As we expand beyond English, it is important that we add language detection to our television news story segmentation workflow…
Story Segmentation Of TV News: Experiments With Spanish-Language News Broadcasts
Given the tremendous success of our new story cataloging workflow applying Gemini 2.5 Flash to index and catalog the stories…
Enriching Democracy: Cataloging & Annotating The Legislative Process: First Experiments
Two years ago we explored the ability of LLMs to read through congressional transcripts to catalog all of the legislative…
Using Gemini To Explore The Narrative Undercurrents Of Television News At Scale: First Experiments
One of the most remarkable elements of our Gemini-powered story segmentation of television news is Gemini's incredible ability to perform…
Story Segmentation Of TV News: Expanding From Evening News To General News & News Commentary Broadcasts
Yesterday we unveiled the incredible results of a massive new collaboration with the Internet Archive's TV News Archive to use…