Quick Take  ·  News  ·  Put AI to Work  ·  Looking Ahead

🔎 Quick Take

A new study in JAMA found that by June 2026, about 700,000 doctors, physician assistants and nurse practitioners used one AI tool every month. In January 2024, that number was 355.

That is fast. One of the researchers compared it to electronic health records. Their spread was slow, he said, and "only really occurred after government regulation."

The same thing is happening outside the clinic. Businesses are adopting AI quickly, and free AI models you can run yourself keep getting better. AI is part of daily work now.

So the next questions are ours to push: does this help patients, and can a small nonprofit use the same tools as a big company? This week brought news on both, from a public benefit fund to open tools for tuberculosis research.

Let's get into it.

📰 AI News Roundup

🩺 Lead: AI is already part of medical care

What happened: Researchers at Harvard Medical School and Brigham and Women's Hospital studied how clinicians use OpenEvidence, an AI tool that answers medical questions. The study looked at about 950,000 U.S. doctors, physician assistants and nurse practitioners over two and a half years. JAMA published it on October 15.

Monthly users grew from 355 in January 2024 to 703,265 in June 2026. Together they asked 270 million questions. About half were about clinical decisions, mostly treatment. Many questions were about topics outside the clinician's own specialty.

"An enormous number of physicians are using AI for clinical care," said Anupam Jena of Harvard Medical School, the study's corresponding author. They are "often using it in areas where they don't have specific expertise."

The National Institutes of Health funded the study. OpenEvidence provided the data but did not pay for the research. Three of the co-authors work or worked for the company.

Why this matters: Clinicians decided this tool was useful, and they adopted it quickly. That is a good sign for patients who need a fast, informed answer. The study didn't measure patient outcomes. The researchers say that is the next step. That research needs public funding, and its results need to be public.

💻 Free AI models you can run yourself had a big week

What happened: Three companies released "open-weight" AI models. You can download these models and run them on your own hardware. The test results below come from the companies themselves.

• DeepSeek released a model under the MIT license, so anyone can use it, even to build a paid service. On one coding test, DeepSeek says it scored 77, compared with 74 for Anthropic's Claude Opus 5. Opus 5 still won on two other tests.
• Nvidia released models built to run on new PCs with Nvidia's own chips. People "can delegate everyday projects … without sending those workloads to the cloud," wrote Nvidia's Kari Briski.
• Meta released Muse Spark Open, which it calls the "most capable" open-weight model in the U.S. It is very large. Running it at home needs at least 1 terabyte of memory, so most people will use it through online services.

Why this matters: When good models are free to download, more organizations can afford them, and more can keep their data at home. That matters for groups that handle sensitive information, like legal aid clinics, immigrant rights groups and health centers. It also means a few companies don't control all the best tools.

💰 Anthropic plans to put 5% of its shares into a fund for the public good

What happened: On October 18, Anthropic said it will commit 5% of its shares to a "public benefit fund," subject to stockholder approval. The fund will be a wholly owned entity inside the company. The board may add 2% more for every 50% increase in the company's value above a set baseline.

Anthropic gave a few early ideas for the money, such as "large donations to traditional nonprofits or direct distributions to specific populations." It has also committed $200 million to study AI's effect on jobs. Anthropic says it will share how the fund will be run "in the coming months."

Why this matters: If AI creates a lot of wealth, some of that wealth should go back to the public. A fund like this is one way to do it. The details decide how much good it does. Who picks the projects? How open is the process? Do the communities affected by AI get a voice?

🗳️ Briefly: Georgia launches a chatbot for state information

Georgia's Secretary of State launched GABBY on October 8. It answers residents' questions using information published on the office's website. Secretary of State Brad Raffensperger said residents "shouldn't have to spend their time searching through a government website or waiting on the phone to get a simple question answered."

Source: StateScoop, October 9, 2026

🛠️ Put AI to Work

Practical ways progressives can use AI this week

Run a free AI model on your own computer

Some work is too private to paste into a cloud chatbot: donor notes, client case details, internal staff issues. A local AI model runs entirely on your computer. A free app called LM Studio makes it easy to try.

Check your computer first. On a Mac, you need an Apple Silicon chip (M1 or newer) and macOS 14 or later. LM Studio recommends 16 GB of memory. On Windows, most computers need a processor that supports AVX2. LM Studio also recommends 16 GB of memory and a graphics card with at least 4 GB of its own memory.

Get started:

1. Download LM Studio from lmstudio.ai.
2. Inside the app, download a small model. Smaller models download faster and run better on ordinary laptops.
3. Pick the model you downloaded, then type into the chat window. For private work, skip the app's optional cloud models and web search, so everything stays on your computer.

Three ways to use it this week:

• Summarize sensitive meeting notes. Paste in notes from a staff or client meeting and ask for decisions and next steps.
• Draft donor thank-you notes. Give it names and gift details that you'd rather keep off a company's servers.
• Work without internet. It runs on a plane, at a rural event, or anywhere the Wi-Fi is bad.

Two cautions. Small models can get things wrong, so check facts, numbers and names. And check your organization's tech policy before you install new software. LM Studio's terms allow internal work use, but some features cost money.

Source: LM Studio system requirements

🔭 Looking Ahead

Patient outcomes. The JAMA researchers say their next question is whether AI tools improve care. Watch for that study, and for who pays for it.

Agents that negotiate for you. Anthropic is asking lawyers how AI agents should negotiate for their users, according to The Neuron, citing the Financial Times. No product has been announced. (The Neuron, October 19)

October 26: Colorado's hearing and comment deadline on its AI decision rules. Comment at coag.gov/ai.

Something to think about: Steve Newman wrote about seven Anthropic engineers who rebuilt the company's core code tools in five months. Claude wrote most of the code. His point is that AI doesn't have to be smarter than experts to change what's possible. "It just needs to be superhumanly fast and cheap." Think about what that could mean for small organizations that could never afford custom software. (Second Thoughts, October 14)

Until next time,
Jordan

Read past issues  ·  RSS  ·  Bluesky  ·  Website