rdhyee

I'm curious what you as an early-career math student would want the Simon Foundation to do differently from what it says it will do. I hope there is a way for students like to you to get a slice of the pie.

Took this idea and ran with it using Fly's Sprites, inspired by Simon's https://simonwillison.net/2026/Feb/3/introducing-deno-sandbo.... Use case: Claude Code running in a sandboxed Sprite, making authenticated API calls via a Tokenizer proxy without credentials ever entering the sandbox.

Hit a snag: Sprites appear network-isolated from Fly's 6PN private mesh (fdf:: prefix inside the Sprite, not fdaa::; no .internal DNS). So a Tokenizer on a Fly Machine isn't directly reachable without public internet.

Asked on the Fly forum: https://community.fly.io/t/can-sprites-reach-internal-fly-se...

@tptacek's point upthread about controlling not just hosts but request structure is well taken - for AI agent sandboxing you'd want tight scoping on what the proxy will forward.

The interview with Perplexity CEO Aravind Srinivas on DeepSeek is a good listen on a Friday afternoon as we ponder the implications of DeepSeek.

I can get to that setting (when logged in) at https://www.linkedin.com/mypreferences/d/settings/data-for-a...

Full title of news blurb from UC Berkeley's I School:

> Award-winning Research Explores Risks of Using Synthetic Data to Train Facial Recognition Technology

Lecture description:

Physical systems that can learn by themselves Andrea Liu

Brains learn and perform an enormous variety of tasks on their own, using relatively little energy. Brains are able to accomplish this without an external computer because their analog constituent parts (neurons) update their connections without knowing what all the other neurons are doing using local rules. We have developed an approach to learning that shares the property that analog constituent parts update their properties via a local rule, but does not otherwise emulate the brain. Instead, we exploit physics to learn in a far simpler way. Our collaborators have implemented this approach in the lab, developing physical systems that learn and perform machine learning tasks on their own with little energy cost. These systems should open up the opportunity to study how many more is different within a new paradigm for scalable learning.

From the article:

    The fallout from the hack of a little-known but pivotal health-care company is inflicting pain on hospitals, doctor offices, pharmacies and millions of patients across the nation, with government and industry officials calling it one of the most serious attacks on the health-care system in U.S. history.

[....]

    Change Healthcare is a juggernaut in the health-care world, processing 15 billion claims totaling more than $1.5 trillion a year, the company says. It operates the largest electronic “clearinghouse” in the business, acting as a pipeline that connects health-care providers with insurance companies who pay for their services and determine what patients owe. It supported tens of thousands of physicians, dentists, pharmacies and hospitals, handling 50 percent of all medical claims in the United States, the Justice Department wrote in a 2022 lawsuit that unsuccessfully tried to block UnitedHealth from acquiring the company.

From <https://www.nasonline.org/programs/distinctive-voices/upcomi...>:

Wednesday, January 31, 2024 The Future of Artificial Intelligence AI is all around us—recognizing our faces in photos, transcribing our speech, constructing our news feeds, navigating our driving routes, answering our search queries, and much more. But rapidly improving AI is poised to play a much bigger role in all of our lives. In this lecture, AI expert Melanie Mitchell will demystify how current-day AI works, how “intelligent” it really is, and what our expectations—and concerns—about its near-term and long-term prospects should be.

» Melanie Mitchell, Santa Fe Institute

The scammers are very effective. One of my family members almost fell for a similar scam recently.

I think the referenced episode is https://learnbayesstats.com/episode/61-why-we-still-use-non-... Thanks for pointing it out!

Accompanying podcast: https://www.nature.com/articles/d41586-023-02425-8

I'm surprised by the results (from what I've heard so far on the podcast) and a bit skeptical given the involvement of Meta. Looking forward to reactions from various scientific and technical communities.

Description of the service (made by OpenAI):

> The AI Text Classifier is a fine-tuned GPT model that predicts how likely it is that a piece of text was generated by AI from a variety of sources, such as ChatGPT.

> This classifier is available as a free tool to spark discussions on AI literacy.

"Brain on Fraud Apologetics" is a follow-up post by Cory Doctorow on his twitter thread (https://twitter.com/doctorow/status/1628948906657878016) about how he and his friend had been scammed. The Twitter thread was discussed on HN: https://news.ycombinator.com/item?id=34920760

One of the things I find myself doing a lot is trying to make sense of a given HN discussion -- so I fed an archived version of this discussion to the summarizer: (https://labs.kagi.com/ai/sum?url=https://archive.is/HdWx7):

> Universal Summarizer is a tool developed by Kagi that is able to quickly and accurately summarize articles, videos, and other texts. It is able to pick out unique and interesting details from the text, such as Ratz having a prosthetic arm in the bar in the novel Catch 22. It is also able to summarize technical documentation and Moby Dick, as well as provide a summary of the themes in Walden. It is able to summarize videos as well, and is a useful tool for developers to familiarize themselves with academic papers. It is also able to provide summaries in multiple languages, and is able to detect jokes and humorous elements in the text. It is an impressive tool that is sure to be a hit with developers and researchers alike.