AI/tech signals: rare-disease grants, cloud agents, and model taste

The short read

Research and society

Anthropic: up to $50,000 for rare-disease research

• What happened: Anthropic opened its first focused AI for Science call for researchers working on rare genetic diseases, offering up to $50,000 in Claude credits for six months. 1
• Why it matters: The program has separate tracks for basic-science researchers and early-stage biotechs trying to speed rare-disease clinical development. 2
• Signal: Applications close at 07:59 on August 3 in this digest's display time; Anthropic lists the source deadline as 11:59 PM Pacific on August 2. This is a concrete access program rather than a general promise about AI for science. 2

Mollick: security teams may have less time than they think

• What happened: Mollick wrote that if China continues releasing open Mythos-class models, and those models are as risky as US and UK assessments suggest, CISO offices do not have much longer to prepare. 3
• Why it matters: The post turns model-risk debate into a timing problem for the people responsible for enterprise security, rather than a question reserved for model labs.
• Signal: The claim is explicitly conditional: the post does not establish that the models are being released or independently confirm the risk assessment. 3

Model behavior and evaluation

Mollick: the same prompt produced three different kinds of writing

• What happened: He gave Fable, Sol Pro, and Kimi K3 the same request: write a short, good poem based on The Odyssey in the style of Tennyson or Cavafy. 4
• Why it matters: Mollick judged Fable the winner; he described Kimi's result as a blend of the two poets' themes with odd additions, while Sol's was thematically incoherent.
• Signal: This is a single comparative prompt, not a controlled literary benchmark, but it shows why qualitative tests can expose differences that a single score hides. 4

Chollet: the slope matters more than the snapshot

• What happened: Chollet wrote, "Always keep in mind that the rate of change matters more than the current metric value." 5
• Why it matters: A model score says where a system is; the rate of change asks how quickly that position is moving, which changes how teams plan evaluations and investments.
• Signal: The post supplies no metric or comparison window, so it works as an evaluation principle, not evidence of a particular model's progress. 5

AI tools and developer workflow

Brockman: agents no longer need an open laptop

• What happened: Brockman said one of ChatGPT Work's best features is that it runs in the cloud, so it can work from a phone while the user's laptop is closed. 6
• Why it matters: Cloud execution removes the awkward requirement that an agent's host computer stay awake, changing an agent from a local session into an asynchronous worker.
• Signal: His post is a product observation, not a full technical specification; it confirms the mobile-and-closed-laptop behavior but does not define availability, security, or task limits. 6

The thread to pull