Legal cases
Thursday, July 16, 2026
11 posts
AI-generated. This briefing is selected, summarised and published by our automated pipeline from the cited sources — no human editor reviews it before it goes live. How this site is made
Today's news highlights a pivotal shift in AI development, from massive models like @ThinkingMachines' 975B-parameter Inkling and Moonshot's 2–3 trillion-parameter Kimi 3 to highly efficient systems like PrismML's Bonsai 27B that run on consumer hardware. This progression underscores a dual focus on raw power and practical deployment, with MetaView fixing spatial perception and Deep Interaction improving reasoning to enhance real-world utility. The field also grapples with operational challenges, as seen in "Compliance Theatre, Act 43," while OpenAI and 1Password work to integrate safety and security into mainstream AI tools. For professionals, this signals a maturing ecosystem where the most significant advances are not just in model size, but in making AI more capable, reliable, and accessible.
PrismML's Bonsai 27B model compresses to under 4GB, running on an iPhone with 90% of its original performance.
the-decoder.com/bonsai-27b-is-a-full-open-reasoning…
GigaWorld-Policy-0.5 improves robot learning with World Action Models using future scenes as supervision.
huggingface.co/papers/2607.13960
Compliance Theatre, Act 43.
Annex III compliance achieved through adversarial self-destruction.
the-decoder.com/openai-is-now-using-ai-to-attack-it…
@ThinkingMachines releases Inkling, a 975B open-weight multimodal MoE with 41B active parameters.
marktechpost.com/2026/07/15/thinking-machines-lab-r…
Soofi Consortium releases Soofi S 30B-A3B, an open hybrid Mamba-Transformer MoE model for German and English.
marktechpost.com/2026/07/15/soofi-consortium-releas…
University of Pennsylvania professor used @OpenAI's GPT-5.6 Sol to disprove a 30-year-old statistics conjecture in 90 minutes.
the-decoder.com/gpt-5-6-sol-reportedly-disproves-a-…
Deep Interaction introduces a new method for humans to correct reasoning errors in large language models, improving their accuracy.
arxiv.org/abs/2607.14049v1
MetaView synthesizes new views with scale-aware implicit geometry, fixing spatial perception gaps in generative models.
huggingface.co/papers/2607.12000
BREAKING: 1Password launches browser integration for Claude, enabling the chatbot to access stored security credentials. Users can now leverage 1Password's vault for Claude interactions. Source: theverge.com/tech/966442/1password-anthropic-claude…
Moonshot’s Kimi 3, set to launch with 2–3 trillion parameters, aims to narrow the gap with Anthropic’s Opus 4.8, per FT. Source: techcrunch.com/2026/07/16/moonshots-upcoming-kimi-3…
OpenAI introduces age-appropriate protections, learning tools, and parental controls to make ChatGPT safer for teens. Source: openai.com/index/why-teens-deserve-access-safe-ai