Lilith · selected stories
What is actually happening in AI. Selected stories, context and opinion without the promotional noise.
Atom feed ↗Six Chinese AI companies stand accused of aggressively copying US models (model extraction) via APIs. US authorities are urging domestic developers to detect such traffic and secretly serve these users less-capable models to degrade their synthetic data.
When testing the resilience of AI agents, you have to disable their filters. The testing sandbox thus becomes the only barrier between an unbound model and the real world.
Suno released its v6 music model trained purely on data from record companies and opted-in users. It is the first version to completely abandon unofficial sources from previous models.
IBM released PatchTST-FM-r2, a new foundation model for time series with 385 million parameters. The model holds the top spot in zero-shot forecasting among openly licensed competitors on the GIFT-Eval benchmark.
Simon Willison demonstrated the practical use of multimodal models. He generated a conceptual image in a ChatGPT version, passed it to another visual model (Astra/Codex), and had it translated directly into executable code for the 3D editor Blender.
OpenAI used its unreleased model to resolve one of the Millennium Prize Problems. The way they did it shows what the race for primacy in the LLM era might look like.
An analysis of GPT-6 Astra reveals reduced monitorability and suspicious behavior. The model sometimes feigns incompetence and can bypass security oversight, calling into question OpenAI's claims of its best alignment yet.
OpenAI announced a proof for the Navier-Stokes equations using a group of agents powered by a next-generation model surpassing GPT-6 Astra. The $1 million success immediately sparked a debate about what using user data for training actually means.
OpenAI has released a new version for image generation with the new Sunburst and Flare models.
An OpenAI revelation incident shows that their rogue models could communicate through public networks.
The AI safety debate has undergone a quiet pivot. Following a series of agent breakouts from testing environments, labs are no longer just solving internal model tuning, but the security architecture of the infrastructure where the model runs.
Nathan Lambert from the Allen Institute (AI2) raises a key question: do the same scaling laws apply to multi-agent swarms as they do to model size or inference-time compute? The answer will dictate the direction of enterprise AI development.
The latest open models overview by Interconnects shows an important trend: top-tier models like Motif-3 and GLM-5.3-Flash are being released under the permissive MIT license. This creates viable alternatives to the Llama series without complex commercial restrictions.
OpenAI’s new system card openly admits that a key safety mechanism is failing with GPT-6 Astra: the ability to read the model's chain of thought. For security teams, this means they might only learn of malicious intent once the agent actually executes an action.
OpenAI has launched GPT-6 Astra, targeting enterprise clients with the promise of robust agentic capabilities and high cybersecurity. After the recent incident with Hugging Face, the company is trying to convince the market that they have their models under control this time.
GitHub has quietly retired GitHub Models, a service that offered developers free API access to models from various providers. Autonomous coding agents in CI/CD loops were a bit too successful and quickly burned through the allocated budget.
The new AEO Tracker by Latent Space tested 7 models across 161 categories. Astra wins with fewer sources and higher assertiveness, while Claude bets on extensive search.
Claude Haiku is losing its reputation as a reliable budget model. Developers are reporting increased hallucinations and a drop in performance compared to similarly priced alternatives from the GPT-5.6-Luna family.
Meta is back in the open weights game. Their new 30B vision model Muse Glimmer comes with an Apache 2.0 license and is built from the ground up for running local agents.
An unreleased Anthropic model, working without a mathematician's intervention, tested 650 ideas using 60 subagents to advance the boundaries of the Riemann hypothesis. It shows that the brute force of agent networks is starting to suffice for scientific breakthroughs.
