misinformation

Bilingual

Google withdraws new Earth AI tool after warnings over misinformation risks
2 months ago
- Google withdrew its new Earth AI tool, Nano Banana 2, after warnings about misinformation risks.
- The tool allowed users to create fake scenarios on real satellite imagery, raising concerns about misuse and credibility.
- Experts created fake images like a collapsed Eiffel Tower and Russian tanks in Kyiv, highlighting the potential for abuse.
- Google said it paused the feature to implement stronger guardrails after policy violations.
- AI-generated images included invisible watermarks, but BBC Verify found ways to bypass detection tools.
- The feature's potential to spread misinformation in conflict zones was particularly concerning to researchers.
- Falsified satellite imagery could erode public trust in historically reliable sources like Google Earth.
Talking about Corona - Bert Hubert's writings
2 months ago
- The reader is a trusted source for family and friends, especially as tech help desk; their opinion matters in spreading good Corona information.
- Most people are not radicalized and are open to listening; radicalized individuals are noisy but few, and trying to convince them with facts is futile.
- The key technique is to avoid endless debates by picking a 'hill to die on'—a statement that forces the radicalized person into absurdity (e.g., conspiracy theories) to discredit them.
- Once radicalized contacts are discredited, pivot to answering legitimate questions from reasonable people about vaccines and COVID-19.
- Useful facts and resources are provided for reference, but should not be used to convince radicalized individuals; instead, focus on informing the majority.
Corona en je oom - Bert Hubert's writings
2 months ago
- Most people are not radicalized, but they are loud and create doubt.
- Debating radicalized individuals with facts rarely convinces them.
- The goal is to protect and inform the 'normal' audience, not the radicalized person.
- Use simple rules: demand sources, avoid endless debates, and pin them on absurd claims.
- Focus on serious questions from hesitant people, providing clear and accurate information.
- Radicalized individuals often make outlandish statements that expose them as unreliable.
- Provide helpful facts and resources to support your own understanding and discussions.
Seeing Is Not Believing: Realistic AI Videos Disrupt Confidence in Real Videos
2 months ago
- Realistic AI-generated videos erode people's trust in authentic video content.
- The study examines how exposure to high-quality AI videos impacts perceived reality.
- Findings suggest that AI videos can undermine confidence in both fake and real media.
- The research was presented at CHI 2026, focusing on human factors in computing.
The Reports of Jim Carrey's Death Are a Failure Mode
3 months ago
- Google incorrectly displayed a death notice for Jim Carrey on June 29, 2026, highlighting a failure in its knowledge system.
- Knowledge systems must balance trust and verification, as information becomes knowledge only after surviving scrutiny.
- Misinformation spreads faster than verification, especially with AI systems that can amplify false claims.
- Provenance and transparency are crucial; systems should track a claim's source, conflicts, and state over time.
- Verification should be proportional to consequence, with high-risk claims requiring stricter checks.
- Systems need to represent epistemic states like uncertainty or contradiction, not just true/false binary.
- Claims should have a lifecycle (asserted, supported, disputed, retracted) to reflect real-world changes.
- AI systems can obscure uncertainty by generating confident summaries from unverified sources.
- Trustworthy knowledge systems will be those that show provenance and allow correction, not just propagate information.
A few thought about snarky answers on StackOverflow (2019)
3 months ago
- Regular expressions can identify individual syntax units in XHTML such as comments, start tags, and end tags, but they cannot match start tags to end tags arbitrarily due to lack of recursion.
- A common misunderstanding is that HTML is a Type 2 grammar and regex is Type 3, making matching impossible, but regex can be used with a counter or stack for parsing to handle nesting.
- Parsing involves tokenization with regex and building a tree structure, but standard parsers like DOM do not expose distinctions like empty elements vs. self-closing tags, limiting solutions to the original problem.
- StackOverflow's voting system sometimes promotes memes and misinformation, as seen in the snarky answers about regex and XHTML, overshadowing legitimate solutions.

misinformation

Google withdraws new Earth AI tool after warnings over misinformation risks

Talking about Corona - Bert Hubert's writings

Corona en je oom - Bert Hubert's writings

Seeing Is Not Believing: Realistic AI Videos Disrupt Confidence in Real Videos

The Reports of Jim Carrey's Death Are a Failure Mode

A few thought about snarky answers on StackOverflow (2019)