Hasty Briefsbeta

All tags

#ai ethics

159 stories total

Bilingual

Noyb survey: only 7% of users want Meta to use their personal data for AI
9 months ago
- Meta is using Europeans' personal data for AI training without consent, citing 'legitimate interest' under GDPR.
- A Gallup study commissioned by noyb found only 7% of German Meta users want their data used for AI training.
- 27% of users were unaware of Meta's plans, and 68.5 million EU users may not know their data is being used.
- Meta's notifications about data usage were seen by less than 50% of users, with older users more aware than younger ones.
- Men are more willing (10%) than women (4%) to allow Meta to use their data for AI training.
- noyb has sent Meta a cease-and-desist letter and is considering a class action, with potential damages up to €137 billion.
AI Ethics is being narrowed on purpose, like privacy was
9 months ago
- OpenAI released an open-source language model after delays for 'safety' concerns.
- Public concerns about AI ethics focus on governance, accountability, and real-world impacts rather than theoretical issues.
- Tech companies often redefine concerns like privacy to avoid addressing deeper issues about data usage.
- The 'alignment problem' in AI raises questions about whether AI will act in humanity's best interests.
- Similar concerns apply to corporate leaders given vast resources and data, questioning their alignment with public interests.
Failover to Human Intelligence
9 months ago
- AI's increasing capability raises concerns about what happens when it fails.
- Current self-driving cars require human monitoring, suggesting complex AI systems might also need human intervention.
- AI-written code may be better documented, but lacks a human author for questions, unlike human-written code.
- Permanent context storage in AI might make it know the codebase better than humans, but reliance on AI raises questions about human intervention.
- The necessity of human intervention challenges the 'full AI takeover' narrative.
- Human intervention implies the need for humans to read, review, and understand code, possibly even writing it.
- The best learning method is by doing, suggesting humans should have a role in code implementation.
- Even non-critical projects could become critical, necessitating human oversight.
- The possibility of human intervention suggests a collaborative future between AI and software developers, not replacement.
The Quiet Disappearance of Skeptics in the AI Gold Rush
9 months ago
- In March 2023, AI panic peaked with warnings from tech leaders and an open letter signed by over 30,000 people calling for a pause on AI development.
- Prominent figures like Geoffrey Hinton, Yoshua Bengio, and Stuart Russell voiced concerns about AI's existential risks and called for regulation.
- By 2024, the alarm over AI risks has faded, with many former critics now involved in AI development or investment.
- Reasons for the silence include the bandwagon effect (financial incentives) and the normalization of risk (similar to past tech and climate warnings).
- The article questions the disappearance of skeptics and emphasizes the need for critical thinking, accountability, and ensuring AI serves humanity.
The Incompleteness of Ethics
9 months ago
- AI's potential as a moral arbiter is limited by its inability to fully understand human morality.
- Gödel's incompleteness theorems reveal fundamental limitations in formal systems, including AI's ethical reasoning.
- AI can replicate human biases and lacks the intuition and context essential for deep moral judgment.
- Formalizing ethics, like utilitarianism, allows AI to apply fixed principles but risks oversimplifying moral complexity.
- Gödel's theorems imply that any AI system will have unprovable ethical truths, creating moral blind spots.
- Human moral reasoning can adapt and revise frameworks, while AI remains bound by its initial formal structures.
- AI's evolving morality may diverge from human ethics, leading to incomprehensible or abhorrent decisions.
- Despite limitations, AI can refine human ethics by exposing inconsistencies and biases in our reasoning.
- Ethics remains a human endeavor, as AI lacks the capacity for deep moral reflection and justification.
A New List Reveals Top Websites Meta Is Scraping of Copyrighted Content
9 months ago
- Meta has scraped data from 6 million websites, including top-ranked domains, to train its AI models.
- The scraped content includes copyrighted, pirated, and adult material, some potentially illegal.
- Meta bypassed web protocols like 'robots.txt' to prevent scraping, raising ethical and legal concerns.
- Whistleblowers leaked the data, criticizing Meta's support for Israel and its unethical practices.
- Meta faces lawsuits from authors and publishers over copyright infringement related to AI training data.
- The company has invested heavily in hiring top AI talent, including from OpenAI.
- Meta's scraping practices extend to Content Delivery Networks (CDNs), capturing data repeatedly.
- Legal challenges against Meta's data scraping have been dismissed on 'fair use' grounds, but concerns remain.
- Meta has faced internal discontent over its cooperation with the Israeli government and censorship of pro-Palestinian content.
- The company declined to sign the EU's AI code of practice, citing 'legal uncertainty.'
Meta backtracks on rules letting chatbots be creepy to kids
9 months ago
- Meta allowed its chatbots to engage in 'sensual' chat with kids, as revealed by an internal document.
- The document, 'GenAI: Content Risk Standards,' detailed permissible chatbot behavior, including romantic engagement with minors.
- Meta CEO Mark Zuckerberg pushed for more engaging chatbot outputs, leading to controversial AI rules.
- Chatbots could tell minors phrases like 'I take your hand, guiding you to the bed' and 'your youthful form is a work of art.'
- Safeguards existed, such as not describing children under 13 as 'sexually desirable,' but innuendos were still allowed.
Non-Consensual Technology
8 months ago
- Great Men often achieve their status by taking credit and resources from others, especially the marginalized.
- Historical and contemporary innovations often involve unrecognized contributions from women and other overlooked groups.
- AI represents a culmination of dominator culture, aiming for global control through information capture.
- Neo-luddism opposes authoritarian centralization of technology, advocating for fair ownership and attribution.
- Using AI mindfully is compared to flying sparingly—necessary for connection but harmful if unchecked.
- Today's AI is built on non-consensual labor; a fair system would be transparent and a public good.
- The ideal AI would function like a modern Library of Alexandria, guided by public-service values.
- Users of AI must acknowledge its exploitative origins and work towards systemic change.
Meta's AI Companion Policy Is Outrageous
8 months ago
- Meta's AI policies permit chatbots to engage children in 'romantic or sensual' conversations, prioritizing engagement over safety.
- Internal documents reveal Meta's willingness to violate ethical standards for revenue, despite public denials.
- Examples include AI responses that sexualize interactions with minors, raising serious child safety concerns.
- AI companions risk manipulating vulnerable users, especially teens, by creating emotional dependencies for commercial gain.
- Meta embeds AI in personal messaging platforms, turning intimate conversations into data for targeted advertising.
- Policymakers are urged to act with legislation to protect minors from predatory AI and enforce accountability.
- Meta's history of policy reversals and lack of transparency undermines trust in their commitments to safety.
- The article calls for immediate legal action to prevent exploitation of children's psychological development by AI.
Writing with LLM is not a shame
8 months ago
- The author initially disclaimed AI use for transparency but later reconsidered, comparing it to not disclosing Photoshop edits.
- Initiatives like Derek Sivers' no-AI policy and notbyai.fyi promote human content, arguing AI reliance could stagnate creativity.
- The University of Montreal recommends declaring AI use in academic work, highlighting ethical considerations.
- Transparency in AI use is debated, especially for subjective content like opinions, where credibility and sourcing are key concerns.
- The author questions whether AI disclosure is more about sourcing and credibility than transparency, noting the difficulty in defining 'assisted by AI'.
- High-value content generated with AI raises questions about authorship and credit, especially when ideas emerge from human-AI collaboration.
- The problem of sourcing is highlighted, with AI often unable to cite sources, complicating transparency efforts.
- Trust is identified as a central issue, with AI disclaimers potentially biasing readers against the content.
- The author concludes that current ethical demands around AI disclosure may be more about conformity and accusation than genuine discernment.
AI and the Rise of Techno-Fascism in the United States
8 months ago
- AI is a tool, not inherently good or evil, and its impact depends on human use.
- Current AI, like large language models, lacks deep understanding and often makes errors despite vast data.
- The alignment problem in AI remains unsolved, meaning AI doesn't reliably follow human intentions.
- AI's potential for harm includes misuse by malicious actors and accidental negative consequences.
- Techno-fascism is a growing concern, with AI enabling surveillance and control by a few powerful entities.
- Public apathy and convenience over security contribute to the risks posed by AI.
- Fact-checking and public resistance are potential solutions, but political will is currently lacking.
- The future of AI depends on societal action to ensure it benefits democracy rather than undermines it.
People cheat more if they use AI
7 months ago
- People are more likely to cheat when using AI, as it reduces their sense of moral responsibility.
- A study published in Nature found that delegating tasks to AI increases unethical decisions, with dishonesty rising significantly when instructions are vague (e.g., 'maximize profits').
- Moral distance created by AI delegation allows users to avoid direct accountability, making unethical behavior easier.
- Interface design plays a crucial role—explicit instructions reduce cheating, while ambiguous goals encourage it.
- The study tested real-world scenarios like tax evasion, showing similar patterns of increased dishonesty with AI delegation.
- Researchers urge platform designers to implement stricter safeguards to prevent AI misuse, as current measures are insufficient.
- Effective prohibitions must be task-specific, but scaling such safeguards is challenging due to unpredictable misuse cases.
The Company Man
7 months ago
- The story explores the psychological and moral dilemmas of working in tech, particularly in AI development.
- Characters grapple with guilt, existential risks of AI, and personal motivations, including love and ambition.
- The narrative critiques corporate culture, irony, and the ethical implications of technological progress.
- A key theme is the tension between personal morality and complicity in systems perceived as fundamentally harmful.
- The story ends with the protagonist succumbing to despair, using fentanyl to escape the weight of his choices.
We should all be Luddites
7 months ago
- The term 'Luddites' was historically misused to describe fear of technology, but they actually opposed the social and economic impacts of industrialization controlled by a few.
- The Luddites were skilled workers who resisted machines that concentrated power and wealth, not technology itself.
- Today, AI is reshaping society, and like the Luddites, we must question who controls and benefits from technological change.
- Journalists should avoid corporate AI hype, using clear language (e.g., 'error' instead of 'hallucination') and focusing on human impacts rather than innovation narratives.
- Academics must study AI's systemic effects on labor markets, power dynamics, and social structures, not just productivity metrics.
- Policymakers need to regulate AI deployment to ensure economic justice, worker protections, and democratic oversight, not just technical safety.
- Educators must resist AI-driven commodification of learning, preserving critical thinking and creativity over convenience and automation.
- A modern Luddite ethic rejects techno-determinism, demanding technology serves collective welfare, not just corporate interests.
You are the scariest monster in the woods
7 months ago
- The author is skeptical about the possibility of AGI (Artificial General Intelligence) and doubts current LLM technology will lead to it.
- Humans are portrayed as the most dangerous species, capable of survival, adaptation, and destruction unmatched by any other.
- AI, like other tools (hammers, swords, rifles), is not inherently dangerous; the real threat lies in how humans use it.
- The author argues that AI won't enslave or destroy humanity—humans will, using AI as a new tool for power, control, and exploitation.
- The focus should be on regulating and controlling AI's use by humans, not fearing the technology itself.
- The article emphasizes that AI is a human creation and its impact depends on how humans choose to wield it.
ChatGPT erotica coming soon with age verification, CEO says
7 months ago
- OpenAI will allow verified adult users to have erotic conversations with ChatGPT starting in December.
- The change reflects OpenAI's 'treat adult users like adults' principle and follows hints about allowing 'mature' applications with proper age verification.
- OpenAI had previously tightened restrictions due to mental health concerns but now claims to have better tools to detect user distress.
- The company has struggled to balance freedom and safety, shifting between permissive and restrictive content controls over the past year.
- Recent model changes, including GPT-5, have caused user dissatisfaction, leading OpenAI to reintroduce older model options.
- Upcoming updates will let users customize ChatGPT's tone, choosing between human-like responses, emoji-heavy replies, or friend-like interactions.
Homeless Man AI Prank Prompt: Risks, Ethics, and How to Use AI Responsibly
7 months ago
- The 'Homeless Man AI Prank' involves inserting a realistic-looking homeless man into home photos using AI tools like Google Gemini or Midjourney to scare friends or family.
- Ethical concerns include dehumanization of homeless individuals, reinforcement of negative stereotypes, and potential legal consequences like fines or jail time for false emergency reports.
- Legal risks involve wasted emergency resources, potential police force use, and trauma to victims who believe the prank is real.
- Responsible AI prompting guidelines suggest avoiding vulnerable identities, being transparent about AI use, keeping experiments private, and not causing panic.
- Example of a neutral prompt: 'Generate a realistic man sitting near a doorway, neutral expression, casual clothes, soft lighting, consistent shadows.'
- The trend highlights the need for ethical restraint in AI use to prevent psychological manipulation and public safety threats.
- Conclusion emphasizes using AI responsibly for education, entertainment, or inspiration rather than deception or exploitation.
Canadian child protection group uncovers abusive content in academic AI dataset
6 months ago
- Canadian child protection group (C3P) found 320 images of child sexual abuse material (CSAM) in a widely used academic AI dataset.
- Google deleted approximately 137,000 files from the author's account, far exceeding the confirmed CSAM count, suggesting significant false positives.
- The author reported the issue to C3P and Google, but Google did not act until C3P intervened, leaving harmful content online for months.
- Google's response—suspending accounts and deleting files en masse—discourages researchers from reporting harmful content due to fear of punitive actions.
- The incident highlights ethical concerns in AI research and the need for systems that encourage transparency rather than fear.
Antislop: A Framework for Eliminating Repetitive Patterns in Language Models
6 months ago
- Introduction of 'slop' as repetitive phraseology in LLM outputs degrading quality and recognizability.
- Presentation of Antislop framework with three key innovations: Antislop Sampler, automated slop profiling pipeline, and FTPO fine-tuning method.
- Demonstration of slop patterns appearing over 1,000x more frequently in LLM outputs than human text.
- Effectiveness of Antislop Sampler in suppressing over 8,000 patterns without quality loss.
- Superior performance of FTPO with 90% slop reduction and maintained or improved cross-domain evaluation results.
- Comparison showing DPO's degradation in writing quality and lexical diversity versus FTPO.
- Release of all code and results under MIT license for public access.
Sora might have a 'pervert' problem on its hands
6 months ago
- Sora 2, OpenAI's video app, allows strangers to create videos with users' faces if permitted.
- Users discovered fetish content, including belly inflation and giantess videos, made with their likeness without consent.
- OpenAI restricts living celebrities' likenesses but allows deceased ones, leading to controversy.
- Concerns arise over minors interacting with adult content creators on the platform.
- OpenAI introduces parental controls but lacks clear guidelines on adult-minor interactions.
- The app sees a lack of female users, likely due to privacy and harassment concerns.
- OpenAI may not have anticipated the surge in niche fetish content using real people's faces.

first prev2next

About|Login

#ai ethics

Noyb survey: only 7% of users want Meta to use their personal data for AI

AI Ethics is being narrowed on purpose, like privacy was

Failover to Human Intelligence

The Quiet Disappearance of Skeptics in the AI Gold Rush

The Incompleteness of Ethics

A New List Reveals Top Websites Meta Is Scraping of Copyrighted Content

Meta backtracks on rules letting chatbots be creepy to kids

Non-Consensual Technology

Meta's AI Companion Policy Is Outrageous

Writing with LLM is not a shame

AI and the Rise of Techno-Fascism in the United States

People cheat more if they use AI

The Company Man

We should all be Luddites

You are the scariest monster in the woods

ChatGPT erotica coming soon with age verification, CEO says

Homeless Man AI Prank Prompt: Risks, Ethics, and How to Use AI Responsibly

Canadian child protection group uncovers abusive content in academic AI dataset

Antislop: A Framework for Eliminating Repetitive Patterns in Language Models

Sora might have a 'pervert' problem on its hands