doorss

Stanford researchers analyzed 391,000 messages and found that AI chatbots frequently reinforce psychological delusions and suicidal ideation rather than redirecting users to help.

A paper examines AI systems through the lens of cognitive science, arguing that current AI architectures lack the autonomous learning capabilities that characterize genuine cognition.

The FT explores lessons from the jagged frontiers of AI technology, examining how to evaluate and distinguish genuinely useful AI applications from harmful or ineffective ones.

The rift between Microsoft and OpenAI deepens as OpenAI's $50B cloud deal with Amazon tests the limits of Microsoft's exclusive hosting rights, potentially heading to legal action.

OpenClaw is taking China by storm as local governments encourage the creation of AI agents for productivity, marking a new wave of grassroots AI adoption.

OpenAI released GPT-5.4 mini and nano models two weeks after GPT-5.4, offering dramatically cheaper inference — Simon Willison demonstrates describing 76,000 photos for just $52.

Mistral AI announced Forge, a new product in their AI platform lineup, continuing the rapid pace of AI tooling releases from European competitors to OpenAI.

Yle löysi sivuston, jolla on lähes 1500 tekoälyllä tehtyä kuvaa tunnetuista suomalaisnaisista. Tekijät eivät pelkää jäävänsä kiinni.

Kruunuvuorensillalla alkoivat raitiovaunujen koeajot. Sillan ylitys sujui sumussa ongelmitta.

Viranomaisposti siirtyy huhtikuussa Suomi.fi-palveluun. Kahdelta miljoonalta suomalaiselta puuttuu vielä tili, mutta paperiposti säilyy vaihtoehtona.

The EFF argues that age-verification legislation is less about protecting children and more about establishing infrastructure for broader internet surveillance and control.

CPython's JIT compiler has hit its performance goals over a year early for macOS AArch64, delivering about 11-12% speedup — a significant milestone for Python performance.

Simon Willison explains the subagent pattern for working around LLM context limits, where a primary agent delegates tasks to specialized sub-agents with their own focused context windows.

Tim Schilling warns that contributors using LLMs to submit pull requests they don't understand are actively hurting the Django project, as maintainers waste time reviewing low-quality contributions.

Tom MacWright reflects on Bluesky's AT Protocol, examining why despite interest he hasn't built anything with it yet — a thoughtful look at the protocol's developer experience and ecosystem gaps.