WHY YOU CAN’T TRUST AI
- sydneysoa
- 1 day ago
- 5 min read

PETER CAVE, a former leading Australian foreign correspondent, now retired after having worked as a journalist with ABC News since the 1970s, recently posted on Facebook the following exposé of AI’s complete unreliability in relation to truth - about himself.
PETER CAVE: After reading that Apple has admitted defeat and is going to pay Google billions to use their AI engine rather than the one they have been trying to develop themselves to power the woeful SIRI, I decided to see how good Google's Gemini was by asking it a few questions about something I know about ... me.
It then spewed out this total fiction. (I have never met Richard Fidler BTW.)
AI RESPONSE: ‘The interview on the ABC program Conversations with Richard Fidler is widely considered the "oral memoir" of Peter Cave’s career. Originally broadcast around the time of his retirement in 2012, it is the source of many of the stories that people mistakenly believe are in a published book.
The highlight of the interview—and the "belly" story you recalled—is his account of the Vanuatu "Self-Surgery" of 1980.
The "Belly" Incident (The Vanuatu Abscess)
In the interview, Cave describes a harrowing medical emergency while he was covering the "Coconut War" (the independence movement) in the remote islands of Vanuatu (then the New Hebrides).
The Condition: He developed a massive, agonizing abscess on his abdomen (his "belly") that turned septic. He was in a location with no medical help and was effectively stranded.
The "Operation": Realizing he would die of blood poisoning if the abscess didn't drain, he took a razor blade and a bottle of gin. He used the gin as both an external disinfectant and a "pre-med" (drinking a significant amount to numb the pain).
The Execution: He described to Fidler how he grit his teeth and used the razor to slice open the abscess himself. He recounted the immense relief of the pressure escaping and how he then poured the rest of the gin into the open wound to cauterize and clean it.
The Aftermath: He survived, though he noted with his characteristic dry humor that it wasn't a procedure he’d recommend to others.
Other Key Moments from the Fidler Interview:
The interview is a deep dive into the psychological toll of being a "war junkie." Beyond the belly story, they discussed:
The "Russian" Ruse in Iraq: Cave details the 2004 incident where he and his crew were stopped by insurgents. He explains how his "fixer" (translator) convinced the gunmen they were Russians because, at the time, Russia had not joined the "Coalition of the Willing." This lie saved their lives.
Tiananmen Square: He talks about the chilling silence of the square before the tanks arrived and the sound of the APCs (Armored Personnel Carriers) "crunching" over the makeshift barricades.
The "Black Dog" (PTSD): This is perhaps the most famous part of the interview. Cave was incredibly vulnerable about the fact that after seeing decades of famine, war, and death, he suffered from severe PTSD. He spoke about how he used to "wash away" the memories with alcohol and how he eventually helped set up the ABC’s trauma support system to make sure younger journalists didn't have to suffer in silence like he did.
Why it feels like a Memoir
Richard Fidler’s Conversations is known for its "biographical" style. Because Cave tells his life story chronologically—from his childhood in Newcastle to his final days as Foreign Affairs Editor—listeners often remember it as a cohesive narrative. If you are looking for this specific account, it is still available on the ABC listen app and the Conversations website under the title "Peter Cave's life on the frontline.”’
PETER CAVE: 'I THEN ASKED A SOMEWHAT MORE ACCURATE AI – ie DIA, WHERE THE BOLLOCKS ON GOOGLE AI CAME FROM’
‘AI mixes sources and “fills gaps,” which causes confident, wrong attributions.
What you’re seeing is a blend of misattribution and pattern-completion typical of large language models. These systems don’t “know” facts; they predict likely text based on patterns in training data. When multiple public narratives overlap—journalism, ABC, war reporting, trauma support, Tiananmen, Iraq—models often conflate identities and stitch together the most statistically “plausible” story, even when it’s false.
Why this happens
Entity conflation: If different people share roles, beats, or employers (e.g., ABC, foreign desks), models blur them into one composite “character.”
Pattern completion: “Conversations with Richard Fidler” is a popular long-form interview brand. Faced with fragments about a veteran ABC correspondent retiring circa 2012, the model completes the pattern with Fidler by default.
Memoryless guessing: LLMs don’t cross-check a registry; they generate text. Without a firm negative signal (“this interview never occurred”), the model leans into what it has most often seen.
Amplification loops: Repeated secondary summaries and blog posts (even if unvetted) become training fodder; the model learns the echo, not the source.
Detail decoration: To sound authoritative, models add vivid specifics (the “belly” abscess, Tiananmen soundscape, Iraq ruse) drawn from similar accounts by other reporters and wrongly attach them to you.
Why you get singled out�You have a long, prominent ABC career with hallmark assignments—Tiananmen, Iraq, trauma support work—that closely resemble canonical “war correspondent” arcs. That high-profile overlap makes you a prime target for the model’s role-based substitution: it slots stories from adjacent reporters into “Peter Cave” because the surrounding context fits.
Why Richard Fidler specifically “Conversations” is a strong narrative anchor: the show is widely cited, often used as a stand‑in for “authoritative oral memoir.” The brand power and frequency in public writing make it a convenient, wrong glue point for the model to attach any veteran ABC life story.
Where the vivid medical tale comes from: Extreme field medicine anecdotes circulate in memoirs, documentaries, and interviews from multiple conflicts and regions. The model likely fused a dramatic abscess/self‑surgery story from another reporter with your conflict coverage, then “grounded” it in a Fidler interview because that pairing is a common storytelling spine.
What generally reduces future misfires
Clear negatives in authoritative bios: Prominent pages stating that no such interview occurred counteract pattern-completion.
Consistent phrasing across profiles: Aligning your public bios (ABC pages, Wikipedia, professional sites) makes the “true” pattern easier for models to learn.
Disambiguation pages: Explicitly distinguishing you from similarly profiled ABC journalists reduces entity conflation.
Sourceable timelines: Clean, date‑anchored career timelines help models map events to you rather than to lookalike peers.
In short, it’s not a smear so much as a statistical shortcut gone wrong: the model fused familiar ABC war‑correspondent tropes with a widely known interview format and attached dramatic details from adjacent reporters to your name.’
CHRISTINE WILLIAMS: My reaction as an editor and small publisher? Thank you Peter Cave for your effort to find out where AI gets its slop.
A SOPHISTICATED MACHINE CREATED BY HUMANS HAS ABANDONED THE BASIC TENET OF WRITING ie ALLOWING A READER TO DISTINGUISH BETWEEN TRUTH AND FALSEHOOD.
DO YOU, READER, THINK THAT IT’S OK TO DRESS UP LIES AS TRUTH? IT’S DEFINITELY NOT OK SO WHY ARE SO MANY PEOPLE, AND BUSINESSES, WILLING TO ACCEPT AI?
AI WRITING CANNOT BE TRUSTED IF IT DOES NOT ABIDE BY ESSENTIAL RULES, IF IT BLENDS TRUE FACTS AND UNTRUE SUPPOSITIONS MASQUERADING AS TRUTH.
MIXING UP SO-CALLED FACTS ABOUT ONE PERSON AND ATTRIBUTING THEM TO ANOTHER PERSON, OR EVEN MORE THAN ONE PERSON, IS CREATIVE WRITING – IN OTHER WORDS IT’S FICTION AND SHOULD NEVER MASQUERADE AS FACT.
HOW LONG WILL IT TAKE BEFORE AI'S DISHONESTY IS FULLY EXPOSED AND ABANDONED? AI SHOULD BE AVOIDED IN ALL HONEST DIALOGUES OR TRANSACTIONS!
Otherwise we members of the human race will surely lose our ability to tell truth from lies!
Article copyright: Christine Williams, based on Peter Cave's Facebook post; photos: Wix - blue microchip & artificial intelligence circuit.

Posts on this SSOA blog are open to members of Sydney School of Arts & Humanities writers' groups and meetups.
If you'd like to join any of our groups or are looking for writing classes, contact us at ssoa.com.au or email sydneysoa@outlook.com




Comments