Pascal's Chatbot Q&As
Because isn't AI the best 'person' to ask about AI? 🤖
Archive
From Training Data to Market Substitution: Why the Seattle Times–Newsday Case Could Be a More Dangerous Copyright Case for OpenAI and Microsoft.
It appears to have been drafted after watching several years of AI copyright litigation and learning which arguments courts have accepted and rejected.
AI’s benefits and harms will depend on who owns the technology, who has bargaining power, and whether productivity gains are broadly shared. Evidence supports real risks...
...to entry-level work, wages, inequality and communities hosting AI infrastructure, although claims of imminent mass unemployment are being made with more certainty than the evidence supports.
AI shutdown stops being a reliable universal control by roughly 2027–2032, arriving unevenly across systems. The durable response is not a bigger kill-switch but more independent chokepoints.
Model-weight security, compute governance, and zero-trust agent architecture — so that weights, credentials, money, compute, and network access each remain separately revocable.
Agentic AI: enterprises need to measure the fully loaded cost per successful business outcome, including retries, external tools, human intervention and failures hidden inside apparent successes.
GPUs can deliver poor returns when storage, data pipelines and orchestration cannot keep them productively supplied. Making token throughput, latency, utilisation and token-per-watt critical.
Silicon Valley genuinely does have a documented, decades-old cultural streak that favors simplification and treats friction as a bug to engineer away.
If frictions go unconfronted, the costs are a market correction, misallocated capital, higher power bills and public backlash, grid instability, and unreliable AI deployed where it can do real harm.
Mayor Zohran Kwame Mamdani and Schools Chancellor Kamar H. Samuels today announced a moratorium on student-facing genAI use in schools...
...alongside a new screen time policy, a decision that will impact nearly 600,000 public school students, or two-thirds of the system’s total enrollment.
The U.S. DOJ’s 1 September 2026 filing argues that copying copyrighted written works specifically for LLM training should generally qualify as fair use...
...while acquisition of the works and infringing outputs must be assessed separately; importantly, the filing is persuasive advocacy, not binding law. It attempts to narrow copyright “market harm”.
Deep Work in a Fast World - How Attention, AI-Assisted Coding, and Delivery Pressure Shape the Adoption of AI Applications.
Where AI products fail today, the causes are predominantly organisational: weak integration, unclear value, and thin governance, rather than any fixed limit on human attention.
EU designated ChatGPT, Reddit and Roblox for enhanced Digital Services Act scrutiny because each exceeds 45 million EU users; ChatGPT’s treatment as a Very Large Online Search Engine is significant...
...because it extends search regulation to generative-AI interfaces. Non-compliance can ultimately bring fines of up to 6% of worldwide annual turnover.
Elites and governments have at times deliberately reduced populations through genocide, engineered or knowingly aggravated famine, coercive sterilisation, eugenics and forced birth-control programmes.
The recurring historical danger is the concentration of power combined with dehumanising ideology, resource fears and institutions willing to treat certain populations as expendable.
Doe v. xAI: xAI crossed the line from merely providing a tool to potentially producing, possessing and distributing unlawful AI-generated CSAM, including material depicting identifiable real victims.
AI-provider and enterprise-user liability, and the case for mandatory dataset screening, identity-sensitive safeguards, output filtering, traceability, takedown and regulator-enforceable safety duties
Sony Music Publishing and Warner Chappell accuse Anthropic of knowingly using pirated, scraped and improperly processed song lyrics to build Claude, stripping copyright information and enabling...
...the model to reproduce or imitate protected lyrics. The case appears stronger than many earlier AI copyright lawsuits