Pascal's Chatbot Q&As
Because isn't AI the best 'person' to ask about AI? 🤖
Archive
Richner Comms et al. v. OpenAI and Microsoft is a strategically important local-news version of the AI copyright cases: nearly 400 regional outlets allege defendants scraped, copied, trained on,...
...and monetised their journalism without permission. The case might increase pressure toward licensing, provenance, attribution, crawler controls and settlement architecture for news content.
Large language models demonstrably withhold, soften, or suppress true information about specific people and topics, implemented at every layer...
...from crude name-blocklists and system prompts to training-stage shaping that is effectively invisible. A made-up fact can be checked and falsified; a quietly-withheld fact leaves no trace.
The Jamendo case against NVIDIA sits at the intersection of four unresolved fault lines: Creative Commons licensing, non-commercial research datasets,...
...music training data, and the emerging market for curated AI-ready datasets. NVIDIA’s own research documentation seems to put the dataset inside the training pipeline.
Western intelligence alliances (Five/Nine/Fourteen Eyes) already run their surveillance, security, and computing on a handful of US tech platforms, and the law and economics make that dependence...
...the default. The result is a severe structural tilt against both civic resistance and fair competition—especially for non-aligned firms facing a $600bn-a-year capital moat plus a security barrier.
With Disney, The New York Times, Adobe, BBC, Wiley, Cambridge University Press & Assessment and others involved, ARIAM frames responsible AI as a matter of law, human creativity, child safety,...
...consumer trust, and democratic resilience. ARIAM makes a plea for provenance, licensing, rights enforcement, safety-by-design standards, and credible rules for AI systems.
There are at least ten distinct technical families of teacher→student transfer, not one monolithic “distillation.”
Widespread distillation is compressing the frontier-to-open capability gap to a handful of months, threatening the defensibility of multi-hundred-million-dollar training runs.
“World domination” by a coordinated cabal is not supported; the realistic danger needs no cabal. Ambient, profit-driven mass surveillance, plus a handful of opaque choke-point firms...
with discretionary control over visibility, plus governments that covet both, is already enough to threaten civil liberties — without any directed conspiracy to protect donors or venture capitalists.
ChatGPT v. Stan van Baarsen: Europe should not mistake an American data centre in Europe for European AI sovereignty. Without European control over ownership, legal jurisdiction, encryption, workloads
...access rights, energy transparency, portability and emergency continuity, the proposal simply moves the environmental and infrastructure costs to Europe while leaving the command layer in the US.
Theology, AI-safety research, and psychiatry all converge on the same counsel: don’t trust the confident voice on its fluency, verify it, and don’t let it isolate you. Demons deceive and aren’t..
...actually omniscient — which is exactly where AI rhymes. LLMs are fast, fluent, apparently all-knowing, unreliable, and flattering, and at scale they can foster real dependency, delusion, and harm.
Investors allege Adobe sold “commercially safe” AI while parts of its AI stack allegedly relied on contaminated training data.
Tech companies can no longer market “responsible AI” without auditable provenance, board-level oversight, precise disclosures, and evidence-backed customer assurances.
Palantir v. Republik matters because it protects journalists’ ability to use strong public-interest language — including terms like “surveillance technology” and “deadly weapon of war”...
...when scrutinising powerful defence, data and AI vendors. Governments, regulators and civil-rights groups should treat Palantir-style systems as democratic infrastructure risks.