by Ron Unz, The Unz Review:
About a year ago, I first began exploring the powerful new AI systems that had been receiving so much public attention, and incorporated some of their features into our website.
For myself and many of our other writers, I added focused chatbots that used the corpus of the written works hosted on our website to simulate the responses of the authors to new questions. This was particularly effective in the case of the Ron Unz Chatbot given that it was trained on the substantial 1.5 million words of my own articles.
TRUTH LIVES on at https://sgtreport.tv/
Then a few months ago I added another AI feature, having the ChatGPT system automatically produce short summaries and outlines for every article we publish that was longer than 1,000 words, thereby allowing readers to easily get a rough sense of pieces that they lacked the time or interest to actually read.
However, over the last few weeks I’ve begun using a new AI system for an even more important purpose, one much closer to the core mission of this website and my own large body of work.
OpenAI recently released an especially powerful new version of the ChatGPT AI called Deep Research. Whereas the ordinary ChatGPT and most other chatbots are designed to respond to prompts within seconds, the Deep Research AI may spend up to 30 minutes working on a given assignment, but it uses that time to produce remarkably detailed and advanced results. For example, according to Wikipedia a standard benchmark test scored the GPT-4o system at only 3.3% and DeepSeek’s R1 model did much better at 9.4%, but Deep Research rated a vastly superior 26.6%.
I discovered that the Deep Research AI could very effectively be used to fact-check exactly the sort of long, complex articles that I often write, and the results it produced were extremely impressive, fully confirming the sort of performance quality suggested by that standard benchmark. This led me to begin producing such fact-checking runs for the pieces in my lengthy American Pravda series and some other ones.
A couple of weeks ago I published an article discussing this process and the dozens of very impressive fact-checking results that I had already obtained.
Unsurprisingly these Deep Research runs are enormously resource-intensive, so that a standard OpenAI account limits them to 25 per month, with the first 10 being full-power and the remaining 15 low-power. We soon switched over to a Premium account that raised the allotment to 250 monthly runs, evenly divided between full-power and low-power. The full-power runs usually seemed as if they had been written by an exceptionally intelligent individual who had read nearly everything available on the Internet and also possessed almost total recall.
This new AI system seemed to have great potential value for my own writing and for other controversial content.
Over the last few years I have produced a huge body of work analyzing many of the most important world events of the last century or more, and often coming to extremely controversial conclusions, conclusions that would have enormous impact upon our entire society if they were judged correct and widely accepted. I have always done my best to adhere to the strictest standards of accuracy and care in writing these sometimes inflammatory articles, and as a result I have regularly declared that I would still stand by at least 99% of everything I have written in this huge body of extremely controversial material.
