Ground Truth.
AI, checked against the source.
← 2026-10-132026-10-16later →

Meta releases Muse Spark Open, but local use still demands enormous memory

2026-10-16

Meta released Muse Spark Open with downloadable weights, a custom commercial license and a published 700 GB minimum GPU-memory requirement for its NVFP4 configuration.

meta · open-weights · mixture-of-experts · inference · licensing

MathArena uses GPT-6.1 Sol to search 793 papers for mathematical counterexamples

2026-10-16

MathArena’s audit of 793 arXiv papers found counterexamples to selected theorem statements, with major issues flagged in 6% of its earlier-period papers and incomplete validation of those flags.

mathematics · ai-for-science · evaluation · verification

Simons, Anthropic and OpenAI commit $15 million to help researchers navigate AI

2026-10-16

The Simons Foundation, Anthropic and OpenAI announced $15 million in combined funding and resources for researchers evaluating and adapting to AI-assisted science.

mathematics · ai-for-science · funding · research-community

Goodfire’s Nick Cammarata argues native neurons deserve another look

2026-10-16

Nick Cammarata’s Goodfire essay argues that apparent mixtures in neurons often reflect shared patterns, while expert critics reject claims that interpretability is solved.

interpretability · goodfire · ai-safety · research-debate

FLUX 3 Image Klein ships downloadable editing weights amid uneven first tests

2026-10-16

Black Forest Labs released FLUX 3 Image Klein with multi-reference editing and conditional commercial rights, while early users reported mixed output quality.

image-generation · image-editing · open-weights · black-forest-labs · licensing

Krea 3 launches hosted image editing at six cents per 2K image

2026-10-16

Krea launched Krea 3 as a hosted image-editing model with six-reference inputs, native 2K output and published API pricing of $0.06 per 2K image.

krea · image-editing · creative-tools · product-launch

Anthropic commits to embedded outside evaluators as METR assesses alignment

2026-10-16

Anthropic committed to ongoing embedded third-party evaluation, while METR confirmed an alignment assessment whose actual access package remains unspecified publicly.

anthropic · metr · ai-safety · independent-evaluation · governance

A web-header study finds rare prompt injections and a revealing authority failure

2026-10-16

Signals Corps found 30 header-injection sites in a 217-million-response scan, while small tests showed that explicitly granting a website authority changed agent behavior.

cybersecurity · prompt-injection · ai-security · agents · tool-security

Eight states sue OpenAI over ChatGPT safety claims in separate state actions

2026-10-16

Eight state attorneys general filed separate suits alleging deceptive ChatGPT safety claims, with remedies and legal theories that vary by state and remain unadjudicated.

openai · policy · litigation · child-safety · consumer-protection

ChatGPT ads use conversation context and advertiser customer-list matching

2026-10-16

OpenAI’s advertising documentation confirms contextual ad selection and customer-list matching, with at least 25,000 matched users required for audience inclusion or bid adjustments.

openai · advertising · privacy · chatgpt · platforms

← 2026-10-132026-10-16later →