Today is the 25th anniversary of 9-11. I have not seen this video before, which someone posted today. Back then we came together as a country, and vowed never to forget, but sadly, I think we have. Like this guy.
Reason.com agrees: universal healthcare doesn't work. Don't look to Canada or the UK as examples of success. Some wag once told me the Law of Healthcare: there is access, quality and affordability. But you can only pick two. That's so true. You can't have all three because it's not affordable and there aren't enough quality physicians and healthcare providers.
Speaking of which, what med schools are the top two providers of American doctors? You'll never guess. (archive.ph seems to be broken – here's the original article). Meanwhile, Duke Medical School discriminated against White and Asian applicants.
––– 良くない–––
Mathematicians are very upset about OpenAI. NYT has two articles: one about the Navier Stokes issue and the other about a sphere-packing problem that was scooped by an AI model called Gauss. It's getting to the point where mathematicians feel that it's not worth putting in the effort to solve problems anymore. OpenAI is the target of much of the anger. Terence Tao calls it "A Severe Misalignment of AI in Mathematics" in his blog. And so a group of mathematicians have signed a declaration, similar to the Leiden Declaration, which calls for something to be done. This has caused OpenAI to drop out of Caltech's Mathathon, just as Anthropic has already done. Wise move. The number of mathematicians who want it cancelled is growing. I've advocated for AI to be tools to help mathematicians do their work, but not to steal their data and use unreleased super models to finish the work and steal the glory. Not cool.
––– 良くない–––
AI may cut junior coding tasks and weaken the pipeline to senior developers. This is an unintended side-effect. Someone has also posted the problem of when AI-generated code falls apart and there aren't any coder apps to fix the software. You'll need actual coders to fix the problem.
––– 良くない–––
Age verification requirements are picking up steam. Anthropic now requires it to use Claude. And Microsoft planning to implement it for Windows, too. Claude is using the Yoti.app which I've not heard of before.
––– 良くない–––
I've noticed this. Both AI and higher interest rates are working to eliminate bullshit jobs. In polite company, we call them sinecure positions. But whatever.
––– 良くない–––
The AI threat hype is in full swing, with Bernie Sanders wanting to jail the AI tech leaders for 20 years. And did CNN really put this on TV? And Dems want Mike Johnson to cancel Congressional recess to deal with the AI threat. Wow, this is COVID-level fear again.
––– 良くない–––
Why browser fingerprinting is not a solved problem. Good – I want it to stay unsolved. I don't want to be identified by my browser.
––– 凄い –––
Strength Training May Be the Secret to Healthy Aging. It's the sarcopenia story again. And look – Creatine improves muscle mass and cognitive function in older adults even without exercise. The problem is that you have to take so much of it. This study used 10 grams/day, whereas most recommend 5 grams/day.
––– 凄い –––
Duck.ai will allow access to frontier AI models anonymously. It's not difficult to use open-weight models anonymously, but using frontier models does require that your data leave your computer and go to datacenters. And unless you are on an enterprise plan, your data often stays there for 30 days before deletion. And you can't be anonymous if you sign up for an API key, which links your usage to your API_KEY. But to use Duck.ai, you have to sign up and get an account. So it's just anonymous to the frontier model companies, but not really as anonymous as using open-weights. Some of them are getting close to frontier models. And Moonshot has been accused of sending user queries to Claude and using the responses to train their models. So you were actually Claude when you thought you were using Kimi-K3. So your data was going to Anthropic after all. That's the risk with using Chinese models. But at this time, there is no way of getting access to Astra or Fable or Mythos class models without sending data to these AI companies, and you just have to hope that your data stays confidential. Or you can pay for an enterprise plan to have zero-retention.
––– 良くない–––