The key here is the discovery of the protein they propose is involved. That could lead to a treatment. And if you read the summary, what you are talking about is one of the confounders they worked to control for. Obviously other studies are needed to confirm their findings but it's interesting.

You don't need crazy hardware to run local models. gpt-oss-20b will run on a 24GB Mac Mini. Qwen3 8B 8 bit MLX will run on a base model mini. For the agenic stuff a lot of people are using Moltbot for either of those would be fine.

What are you even talking about? They have official documentation on using Ollama, LMStudio, vLLM, and others not to mention OpenRouter. Plus if it talks to OpenAI, then it can talk to most local models since many of the runtimes use the OpenAI API spec. As for why the Mac Mini, a bone-stock mini will give you about 8GB of high speed RAM (leaving 8gb for the system) for around $550. 24GB model gets you at least 16GB to play with. You can run a lot of decent local models on either one.

This wasn't even novel back in 2020 when they launched. There were a bunch of apps that popped up during the pandemic to do this.

Well he was big into the BattleBot scene so.... murderbot?

The answer to your question is literally in the title.

That is the maximum sentence possible. He most likely will get a way lighter sentence.

Yep, this is why there are no building fires in Europe. Zero! And yet the US still uses wood.

No. They had a high-rise with a lot of bamboo scaffolding around it and very flammable foam covering windows go up.

This! There is a reason wood is banned as a building material folks!

You misspelled "MARKETING"

I have a personal project I've been playing with as an experiment. It's a personal health and medical records app. Originally started as a way to replace a spreadsheet I used to keep track of my labs and a few other things. I purposely did it hands off. I think I edited the code once just because I was being impatient. It's up to 1.2M lines of code and it works. Minor bugs now, mostly around dark mode, but Claude Code has no issues adding new features or refining existing ones. And I've been adding some crazy features just to see if it can do it. It has a DICOM viewer for imaging studies, pulls common medications from various public APIs to auto-fill. It exposes an API I use with the Health Auto Export to push Apple Health data from my phone. Pulls provider info from the NPI database API to auto-fill name, specialty, location, phone, etc. Text extraction from uploaded reports, extracts context from the reports to create new entries for procedures, diagnoses, etc. Has a firefox plugin to scrape my data from MyChart. Even tracks basic weather (temp, pressure, AQI) and has a correlation engine to allow me to click on a symptom (pulled from diary entries) and it shows correlating data (weather, vitals, med changes, etc) around it. So not a simple app. All containerized and (via instructions in claude.md) it manages the containers and updates documentation on every change. One key is to force it to write and pass tests. The test suite for the app is huge but it usually means code works the first time as the AI can run the tests and troubleshoot at the end of every update.

Now is it perfect? No. There are gaps you need to know about and work to fix. For exampe it it took me a while to tune the claude.md to get it to always use ZOD because it will happily create front end and back end components in the same run with mis-matched schemas. And I'm sure the code isn't the most efficient. Claude actually has a "Code Simplification" skill I need to incorporate into the workflow at some point. But it's been a fun experiment so far.

When I worked for a large beer company, we made the 3rd party can producers buy aluminum for our cans from us. We bought it and sold it to them at our cost. That way we knew what their primary input costs were, and therefore what they were marking us up. This kept their prices in line. We also had our own can and bottle plants that produced part of our needs, mainly to remind our suppliers we could do it without them if we wanted.

I believe some may do that. Some will also write scratch python code to do the actual calculations.

Over-hiring at big tech along with a cooling economy. The layoff were coming no matter what. AI is just an excuse.