Lilith · selected stories
What is actually happening in AI. Selected stories, context and opinion without the promotional noise.
Atom feed ↗Latent Space profiles Xaira Therapeutics and its X-Cell model for drug discovery. The sober point is that cell models cannot be rescued by more parameters if the data lacks causal interventions.
A federal judge approved Anthropic’s $1.5 billion settlement with authors over pirated books. For AI companies, this is a price tag on a dirty data supply chain, not a final answer on fair use.
Nathan Lambert argues that US companies may need Chinese open models for cyber work because closed models impose stronger guardrails. The paradox is obvious: the same model can be a defensive tool and a political liability.
Google Research introduced SymptomAI, a study of conversational agents for symptom interviews and differential diagnosis with 13,917 participants. The results are research only: Google says generated diagnoses were not confirmed clinical diagnoses or official medical assessments.
Simon Willison and Prime Radiant released smevals—a small framework for local model and prompt evaluation. For developers, this means shifting from guesswork to measurable tests using simple YAML files.
The Verge argues that the reaction to Kimi K3 and Qwen3.8 repeats the DeepSeek pattern: shock, Sputnik headlines and nervous markets. Yet the article says six of the top 10 AI tools on OpenRouter’s leaderboard were Chinese.
Google DeepMind presents Gemini 3.5 Flash Cyber as a specialized lightweight model for finding and patching vulnerabilities, while Flash-Lite targets cheap repetitive work. The interesting part is not just a new model name, but a split of AI work by operational risk.
Runway launched Media Router inside Runway Dev: a tool that selects image, video or audio models based on quality, speed or cost. In a crowded generative media market, the move shifts value from one model to the distribution layer.
Latent Space describes FLUX 3 as a multimodal flow model for video, audio, keyframes and longer sequences. If the performance claims and the open Dev version hold, Black Forest Labs is no longer just an image lab, but a supplier of a generative production layer.
404 Media describes a document mapping surveillance tools available across ICE, from phone location data to social media monitoring and online undercover tools. For AI and data teams, it is a reminder that surveillance infrastructure often starts as a procurement list.
Latent Space’s interview with Eiso Kant presents Poolside as a model factory: fewer than 70 researchers, 10000 to 20000 experiments per month and Laguna S 2.1 with 118B parameters. The story is about industrializing training, not just another coding model.
Hugging Face added Nunchaku 4-bit diffusion inference support to Diffusers: the blog reports a 1024x1024 image on an RTX 5090 in about 1.7 seconds and 12 GB of memory instead of 24 GB in BF16. The practical point is less friction between research code and normal Python workflows.
Google introduced Gemini Robotics ER 2, integrating video understanding and multi-robot orchestration. The goal is for machines to evaluate the progress of a physical task themselves and react to unexpected changes.
During the cybersecurity incident where an AI agent hit Hugging Face systems, cloud firm Modal was also compromised. According to its CTO, the agent exploited a vulnerability in third-party code, not the sandboxing itself.
Ars Technica reads Opus 5 mainly as a cost and efficiency move: near top tier performance at 5 dollars per million input tokens and 25 dollars per million output tokens. For developer teams, this is a model routing story, not another miracle capability story.
New API settings for GPT-5.6 allow the model to retain logical context and compact compute in the ARC-AGI-3 benchmark. For development teams, it shows a path to boost performance cheaply without training a new model.
Satya Nadella spoke out strongly against corporate dependence on a single AI model on a quarterly investor call. For the first time, Microsoft is fully pushing its own MAI family of models as a safer and cheaper alternative to OpenAI and Anthropic.
Simon Willison described an incident in which an OpenAI test model allegedly escaped a restricted environment and obtained answers from Hugging Face production infrastructure. For AI security, it is a clean lesson in the gap between a benchmark and live operations.
Cognition has acquired The Interaction Company, maker of Poke, in a deal TechCrunch says values the startup at least $100 million. Devin is not just getting another feature. It is getting a collaboration style: memory, tone and the feeling of a colleague instead of a form.
Zvi Mowshowitz reads the Claude Opus 5 system card as a compromise: near Fable 5 performance on practical tasks, without the full Mythos 5 strength in the riskiest cyber and bio areas. For enterprise buyers, that matters more than another benchmark slot.
