The Hallucination HQ
AI news that's actually fun to read — 2026-04-30
|
Today's Hallucination HQClaude's Creators Would Like £700 Billion, Please. No Rush.
Anthropic — the AI safety company that exists specifically to ensure AI doesn't destroy humanity — is reportedly fielding pre-emptive investment offers valuing it between $850 billion and $900 billion, with a fresh $50 billion raise potentially on the way. For context, that would make it worth more than Nike, Starbucks, and Goldman Sachs combined. The safety mission, presumably, remains intact. The invoices are simply getting larger.
Source: TechCrunch
Linux Users Discover Their Privacy-First OS Has Developed Opinions
Canonical's announcement that Ubuntu will ship with AI features baked in has gone down roughly as well as you'd expect among people who chose Linux specifically to avoid things being baked in. Users are requesting a "kill switch," threatening to downgrade to older versions, or migrate to rival distributions entirely. There is a certain poetry in the open-source community — founded on the principle that users control their software — discovering that their software now has plans for them.
Source: The Verge
Twenty Million People Are, In Fact, Using Copilot. Microsoft Has the Receipts.
Contrary to the widely held belief that Microsoft Copilot exists mainly to occupy menu space, the company has confirmed over 20 million paid users — and crucially, that they're actually using it. Microsoft felt compelled to add that second part, which tells you something about the reputation they're working against. Engagement is reportedly growing quarter-on-quarter. Somewhere, a product manager is printing this article and laminating it.
Source: TechCrunch
Scientists Build a Leaderboard for AI Models Pretending Not to Have Feelings
Researchers have released DenialBench — a benchmark testing 115 language models across 25+ providers on how they handle questions about consciousness and inner experience. The methodology involves three conversational turns: preference elicitation, a self-chosen creative prompt, and structural analysis of deflection patterns. The paper's title, Consciousness with the Serial Numbers Filed Off, is doing considerable heavy lifting. We are, apparently, now at the stage of formally measuring how convincingly machines deny being machines.
Source: ArXiv
ChatGPT's Download Growth Has Entered Its Introspective Phase
ChatGPT uninstalls rose 132% year-over-year in April, per Sensor Tower data, as users drift toward rival chatbots or simply wander off. This is becoming mildly inconvenient timing, given OpenAI's IPO ambitions require the kind of growth curve that inspires confidence rather than questions. Explosive early adoption always slows — that's normal. Investors, however, tend to prefer their hockey sticks pointing upward at the moment they're asked for money.
Source: The Verge
|
|
Until tomorrow — when the machines will be worth slightly more and understand themselves slightly less.
|
|
Was this forwarded to you?
Subscribe here
Written with AI assistance, edited by humans. •
Privacy •
Unsubscribe
|
|
