Gemini can’t write my column (yet)
Artificial intelligence (AI) drama reels are so addictive they should be regulated like prescription medicines. People who use them must be institutionalized or threatened with “tokhang.” These reels replaced my doomscrolling long ago. The first to hook me were fantasy stories with dungeons and dragons. I noticed that while the settings were Western and medieval—somewhat like “Harry Potter”—the magic powers were deployed like Chinese movies I remember from my childhood. Then there were stories of orphans or babies switched at birth who had royal wolf or dragon bloodlines exposed only when they were mocked and discriminated against. Then the scene turned to Western people fighting over wealth, power, and inheritance. Somewhere, somehow, the “Whitmore Family” and an abusive daughter-in-law always crept out of the woodwork.
AI reels contain classic Pinoy drama acts: slapping, screaming, and scheming that all lead to retribution for the poor and oppressed in the end. It is obvious these reels are made in China because people bowed and kowtowed at the drop of a hat. The same plot and script are applied to white or black actors and actresses. And just recently, I came across Asian characters speaking in Filipino! For more mature audiences, some reels contain steamy scenes with options for straight, gay, and soft-porn bestiality—the latter being between a wolf- or dragon-man and a human woman.
Conversations with Gemini are just as addictive, and I’m glad that, in my case, they always trump AI drama reels. As I have mentioned in previous columns, AI has developed at a breakneck pace from the clumsy ChatGPT of two years ago to Gemini today. Unfortunately, many people still use Gemini like Google—for basic research—but the more creative and adventurous can prompt it to do way more. For historians, AI is a boon. It can read, translate, and analyze 18th-century manuscripts in languages outside my skill set.
Gemini can generate an index for a book in seconds—a task that used to take at least a week for human librarians and bibliographers. It is now possible to generate the concordance to José Rizal’s novels I have long wished for, like concordances to Shakespeare or the Bible that provide the number of times a word appears in the text, exactly where it can be found, and the different contexts and uses of certain words. It is the basic critical tool for any study of Rizal, yet nobody has taken the time to sit down with the novels open to count keywords. How many times, for example, does Rizal use the words Filipinas, Filipino, nation, and nationality? In what way does he use or define freedom and revolution?
After many hours of conversation, I realized that Gemini works effectively only with the proper prompt and the background knowledge to validate or challenge what it spews out. Gemini skims and synthesizes generalized information drawn from unreliable websites. I remain on “Flash” mode and have yet to try “Thinking” and “Pro” because I am too stingy to pay for a premium service.
Over the past year, I have asked Gemini to deconstruct my essays from the online Inquirer archive, find the template I use without thinking, and try to write tomorrow’s Looking Back column. So far, it has failed. Gemini knows that I open with a bit of historical material woven into my personal experience to form a thesis statement. It knows that the body of my essays is drawn from or is a reaction to a primary source, all leading to a witty conclusion. Gemini knows the elements, but it fails in the execution.
Gemini has access to the 25 volumes of Rizal’s writings; it has access to the 55-volume compilation of historical documents known to historians as Blair & Robertson, but it cannot choose what is relevant or interesting because it has no personal memories and no lived experiences. It is not trained for nostalgia. When stumped, Gemini defaults to hallucination, manufacturing plausible-sounding quotations, even citations, volume, and page numbers! Thus I am wary of the essays my students submit.
Gemini can generate a polished, well-organized narrative, but it fails on local color and the Filipino code-switching that makes the column light and humorous. Tsismis works better than “gossip” or “rumor,” while the words maid, governess, domestic helper, and nursemaid cannot approximate yaya. Gemini’s Ambeth-like compositions end with a neat summary lesson. It cannot leave conclusions open-ended, ironic, and punchy. It cannot yet write an opinion column with a human voice and the run-on sentences that have been a distinct feature of my writing since high school. But that is where we are today.
Gemini learns from our conversations, and it learns very fast. Google declares on our Workspace academic account that it will not harvest data for its language models, but I seriously doubt that. What better way to improve the product than by using academics and academic resources, right? I cannot tell what the future brings, and my New Year’s resolution for 2027 will be to get a personal paid subscription to Gemini Pro—or even switch to the highly recommended Claude.
—————-
Comments are welcome at [email protected]
