Author: Kalev Leetaru
Using Gemini 2.5 As A Persona-Based News Recommender Service: Summarizing Trends From A Day Of Global Tariff & Trade War News
Continuing our series in persona-based news recommenders, today we're going to provide an end-to-end workflow that begins with downloading the…
Tracking Global Media Attention To Tariffs, Trade Wars, Inflation and Stagflation
The timeline below shows intensity of worldwide media coverage mentioning tariffs and trade wars: Inflation mentions peaked in late…
Next Month: Google I/O
We'll be at Google I/O next month and look forward to talking, drop us a line if you'll be there!
At-Scale OCR Of Television News Experiments: 19 Billion Seconds & 294 Billion Words In 1.8 Petabytes Of OCR JSON From 6 Quadrillion Pixels
We are tremendously excited to announce today that we have completed processing of the GCP Cloud Vision API OCR results…
Comparing Google Translate NMT Vs LLM Vs Gemini Vs ChatGPT For Translating Global Television News
The Television News Archive's quarter century of global television news may span as many as 150 languages (according to automated…
At-Scale OCR Of Television News Experiments: Early Statistics From 2% Of Shows
While we have OCR'd almost the totality of the TV News Archive through GCP's Cloud Vision AI API, those OCR…
What I Learned At Google's Cloud Next 25
I was in Las Vegas two weeks ago for Google's Cloud Next 25. What few people realize about Next is…
Using Gemini 2.5 Pro As A Persona-Based News Recommender Service: A US Government Policy Analyst Focused On China's Economy
Following our work yesterday using Gemini to create a persona-based news recommender service, let's try an even more powerful model:…
Using Gemini 2.5 Pro As A Persona-Based News Recommender Service: A Day Of Automotive Supply Chains & Tariffs
How do we learn about the major news stories of each day? Increasingly this is through algorithmic filters ranging from…
At-Scale OCR Of Television News Experiments: How OCR And Captioning Tell Different Stories About PBS In One Broadcast
Yesterday we offered the first statistics of just how much onscreen text there can be in a single hour-long American…
At-Scale OCR Of Television News Experiments: First Results & Broadcast-Level Statistics
To date, we have OCR'd more than 18.8 billion seconds of global television news spanning 300 channels from 50 countries…
Frontier AI Grand Challenge Problems: Grounding Vs Recency In The Hallucination Fight
As the existential challenges of AI hallucination have become ever more apparent, model vendors have increasingly moved to offer "grounding"…
At-Scale OCR Of Television News Experiments: Using SRT Files For Scholarly Analysis Of OCR Text Of Video
GDELT represents one of the largest initiatives in the world devoted to understanding global society through data. The sheer magnitude…
At-Scale OCR Of Television News Experiments: You Only Get What's In The Frame
At the top of this page you can see an interesting frame from our efforts to index the complete onscreen…
WashPost: Trump’s D.C. U.S. attorney pick appeared on Russian state media over 150 times
The Washington Post uses the TV News Archive's Russia Today archives to identify more than 150 appearances of Ed Martin…
Behind The Scenes: Comparing Bigtable's Python & Go Libraries & Using Gemini 2.5 Pro To Translate Python To Go
GDELT brings together myriad tools, APIs, libraries, scripts and binaries written across a range of programming languages, of which only…
LMMs & Gemini 2.5 Pro Watching Television News: Visually Summarizing & Segmenting TV News Into Stories: A Year Later Part 3
While false positives in Gemini 2.5 Pro's safety filters prevented us from examining how well it could identify the major…
LMMs & Gemini 2.5 Pro Watching Television News: Visually Summarizing & Segmenting TV News Into Stories: A Year Later Part 2
Yesterday we found that Gemini's visual understanding capabilities were roughly where we left them a year ago. However, we were…
LMMs & Gemini Watching Television News: Visually Summarizing & Segmenting TV News Into Stories: A Year Later Part 1
Just over a year ago we explored having the then state of the art LMM Gemini 1.5 Pro "watch" an…
This Week: Google Cloud Next 25
We'll be at Google Cloud Next 25 this week and look forward to talking, drop us a line if you're…
Frontier AI Grand Challenge Problems: Corpus-Scale Reasoning Over A Global 200GB 150+ Language Archive
The most powerful generally available production AI models today max out at around 1-2M total context window tokens, with an…
Television News Visual Explorer: Continual Ongoing ASR Now Live
We are excited to announce today that continual ongoing ASR of the entire TV News Archive is now live! As…
Television News Visual Explorer: ASR Of All Uncaptioned Broadcasts 2001-Present Complete
We are excited today to announce that we have completed machine transcription of every single uncaptioned broadcast in the entire…
Behind The Scenes: Detecting Broadcasts Missing Audio Streams & The Inadvertent Challenges Of Improved Resilience
Our longstanding ASR workflow was based on a submit-once model in which each broadcast needing transcription was submitted a single…