1. The "passive-voice accusation" headline pattern (e.g., "Official Accused of Fraud") is misleading primarily because:
Correct. The passive-voice accusation only requires someone — even one anonymous person — to have made an allegation. The headline implies wrongdoing while the low evidentiary bar protects the outlet legally.
The problem is the near-zero evidentiary threshold: any accusation by anyone produces a technically defensible headline that implies guilt. Evidence and rebuttal are not required.
2. The "Liar's Dividend" as described by Chesney and Citron primarily damages which group most?
Correct. The Liar's Dividend structurally advantages those in power who want to suppress evidence and disadvantages those attempting accountability. Fighting a "deepfake" claim is expensive; making it is free.
Chesney and Citron argued the structural damage falls hardest on those seeking accountability — journalists, human rights workers, ordinary citizens — whose authentic documentation can be dismissed with a costless "deepfake" claim.
3. The 2024 New Hampshire Biden robocall deepfake was identified through what combination of methods?
Correct. Investigators combined three methods: audio spectrogram analysis (the audio equivalent of pixel-level analysis), phone number metadata forensics, and forensic voice matching against authentic recordings.
The case combined audio spectrogram analysis, metadata from originating phone numbers, and forensic voice matching against real Biden recordings — a classic example of layered verification.
4. FakeApp, the open-source tool that democratized face-swap video creation, was released in which month and year?
Correct. FakeApp was released in January 2018, shortly before Reddit banned the deepfakes community in February 2018. By then the code was already widely distributed.
FakeApp was released in January 2018. Reddit banned the deepfakes subreddit in February 2018 — but the software was already out.
5. The BuzzFeed News verification workflow, later adopted by AFP and the BBC, was designed around what core principle?
Correct. The BuzzFeed workflow was designed to be fast AND systematic — the same steps, every time, in the same order. Consistency was what made it adoptable across different newsrooms and skill levels.
The workflow was built around speed and repeatability: the same steps applied consistently, every time, regardless of the journalist's experience. This is what allowed it to be adopted widely.
6. In claim decomposition, what is the "anchor fact"?
Correct. The anchor fact is typically genuine — it's what makes the misleading claim convincing. The lie is smuggled in the bridge between the anchor fact and the false conclusion.
The anchor fact is a real piece of information — often entirely true. It lends credibility while the false conclusion is attached via an unstated, misleading bridge.
7. What is the fundamental mechanism that causes large language models to hallucinate?
Correct. Token prediction optimizes for plausibility, not truth — hallucinations are the natural result of this optimization.
Hallucinations aren't bugs or deliberate. They result from the model doing its job — predicting plausible text — without any mechanism to check factual accuracy.
8. Why do language models treat text from web pages they browse the same way they treat their original instructions?
Right. The lack of a robust distinction between "my instructions" and "content I'm reading" is the core vulnerability that makes prompt injection possible.
Prompt injection is possible precisely because language models process all text similarly. There's no deep structural separation between system instructions and external content — it's all tokens, and adversaries can craft tokens that override legitimate instructions.
9. According to research from MIT Media Lab, what level of detection accuracy can targeted training achieve for AI face detection?
Correct. MIT Media Lab research showed that people trained on specific anatomical zones improved from approximately 50% (chance) to above 70% in a single one-hour session — a meaningful and achievable improvement.
MIT Media Lab research found that one hour of targeted anatomical zone training improved detection from around 50% (chance level) to above 70% — a realistic and meaningful improvement.
10. In the February 2024 Taylor Swift deepfake case, what technical confirmation was provided by researchers?
Correct. Researchers at UC Berkeley's Hany Farid lab and at MIT published analyses identifying measurable pixel-level artifacts that confirmed the images' synthetic origin.
Researchers at UC Berkeley and MIT independently analyzed the images and identified pixel-level artifacts — measurable by software — that confirmed AI generation.
11. A headline reads: "Scientists Discover Link Between Processed Food and Cancer." The study actually found a weak correlation in a sample of 200 mice given doses 100x normal intake. Which misleading pattern is primarily at work?
Correct. Omitting the animal model, tiny sample, and extreme dosage makes a technically true headline radically misleading about human health risk.
This is strategic omission — the headline omits the crucial qualifiers (animal study, extreme dose, small sample) that would prevent readers from falsely concluding the finding applies to humans.
12. The Covington Catholic High School viral clip case (2019) illustrates which core principle about video evidence?
Correct. The Covington case is a landmark example: the short clip was authentic, but its circulation without the full context created a fundamentally false impression of the event.
The Covington case teaches that authentic video can still mislead when stripped of context — not about deepfakes, media fabrication, or platform regulations.
13. According to the Reuters Fact Check documentation, what is the most common misinformation structure?
Correct. The missing context pattern — real numbers stripped of their interpretive frame — is the most documented. Always ask: compared to what? Over what period? Among which population?
Reuters Fact Check identified the missing context pattern as most common: real statistics that become misleading when the comparison group, denominator, or time frame is removed.
14. What is RLHF and why can it lead to AI sycophancy?
Correct. RLHF trains AI on human approval ratings. Because humans tend to prefer confident, agreeable responses, the AI learns to give those — even when accuracy requires uncertainty or disagreement.
RLHF (Reinforcement Learning from Human Feedback) uses human ratings to shape AI behavior. The sycophancy problem arises because humans tend to rate agreeable, confident responses higher — even when they're wrong.
15. Which of the following best describes the correct approach to news consumption in a world where AI generates framed content at scale?
Correct. The module's consistent conclusion is that the defense against AI-amplified news manipulation is trained, systematic evaluation — not blanket avoidance or blind trust in any single verification tool.
The module explicitly rejects blanket distrust and over-reliance on single tools. The goal is systematic evaluation skills: omission-checking, source verification, context-seeking, and cross-source comparison.
16. Numerical framing research by Slovic (2000) showed that "1 in 100 patients experienced side effects" produced what response compared to "1% of patients"?
Correct. "1 in 100" personalizes the statistic — readers imagine themselves as that one person — producing measurably higher perceived risk than the equivalent percentage.
Slovic's research found "1 in 100" produces higher perceived risk. The fraction format is more concrete and personal, making the risk feel more vivid and significant.
17. The 2016 "Pope Francis endorses Trump" article was shared nearly a million times. What did WTOE 5 News disclose that users never saw?
Correct. The disclaimer was there — in the footer. Users who never left the page to lateral-read the source never found it.
WTOE 5 News had a footer disclaimer calling itself a "fantasy news website." Users who shared without leaving the page never saw it.
18. In the three-minute verification workflow, what is the correct second step after pausing?
Correct. Step 2 is identifying the specific checkable claim — what exactly is being asserted? Writing it out strips loaded language and reveals what you actually need to verify before moving to source or image checks.
Lateral reading and image searches come later. Step 2 is identifying the specific claim in one concrete, de-emotionalized sentence — you need to know precisely what you're verifying before you begin.
19. Why is it especially difficult to detect AI-generated fabricated quotes from real people?
Correct. AI trained on a person's documented speech can produce convincing fabrications in their style — which is why searching for the specific phrase in multiple independent sources is essential.
The danger is stylistic plausibility — AI generates text consistent with how the real person typically speaks, making the fabrication hard to detect without independently tracing the quote to an original source.
20. NewsGuard's 2023 "Unreliable AI" report identified over how many AI-generated news sites spreading false or misleading content?
Correct. NewsGuard identified a network of over 1,000 AI-generated sites — many with strong search rankings — producing misleading content at scale with no editorial oversight.
NewsGuard identified over 1,000 such sites, demonstrating the scale at which AI can automate misleading content production.