Nick @internic@qoto.org

I have played a little bit with OpenAI's new iteration of GPT, GPT-o1, which performs an initial reasoning step before running the LLM. It is certainly a more capable tool than previous iterations, though still struggling with the most advanced research mathematical tasks.

Here are some concrete experiments (with a prototype version of the model that I was granted access to). In https://chatgpt.com/share/2ecd7b73-3607-46b3-b855-b29003333b87 I repeated an experiment from https://mathstodon.xyz/@tao/109948249160170335 in which I asked GPT to answer a vaguely worded mathematical query which could be solved by identifying a suitable theorem (Cramer's theorem) from the literature. Previously, GPT was able to mention some relevant concepts but the details were hallucinated nonsense. This time around, Cramer's theorem was identified and a perfectly satisfactory answer was given. (1/3)

A friend of mine asked me to help with her website just last year. It took me a while to figure it out. It was an old cpanel install with static html files updated over ftp. It offended my dev sensibilities.

I asked if she wanted to upgrade to something like squarespace. Then I realized that didn't make any sense. She didn't care about any of that, and the website she had only cost her $15 per year. So I just made the simple edits she needed and then let her move on with her life.

Hey kids! Starting tomorrow, you too can give your money over to a new "high-yield crypto investment" scheme endorsed by former president Trump, and run by his sons, including 18 y/o Baron, who is cited as the "chief visionary."

Sounds like a brilliant investment (for the Trump family):

"A whopping 70% of Trump-backed World Liberty Financial's WLFI tokens will be reserved for the project's insiders, according to a white paper draft obtained by CoinDesk. Of the remaining 30% of the tokens distributed via a public sale, the founding team will also receive a portion of the proceeds."

Because nothing screams legit like a high-yield investment scheme. Add crypto and it's uber-legit: https://www.investor.gov/protect-your-investments/fraud/types-fraud/high-yield-investment-programs

CoinDesk quotes a Trump fan and crypto investor predicting disaster: "It'll be the juiciest DeFi target ever and it's forked from a protocol that itself was hacked). it's also an obvious target for the SEC. at best it's an unnecessary distraction, at worst it's a huge embarrassment and source of (additional) legal trouble."

https://www.coindesk.com/business/2024/09/04/in-trump-backed-crypto-project-insiders-are-poised-for-unusually-big-paydays/

#shamelessgrifters

@benroyce

I love the modern world and I will fight about it.

I love my antibiotics and clean drinking water.

Does anyone here use DuckDuckGo, the browser, on macOS? I’ve been trying it out and am pretty happy with it, aside from the fact that tabs don’t always close when I hit cmd+W. I *think* it might have something to do with how the tabs are created, but I have no idea. It’s irritating.

Anyone else see this behaviour?

https://duckduckgo.com/

#DuckDuckGo #MacOS

I can't entirely back this up, but here's what I'm feeling. Bluesky could grow to be a close approximation of what Twitter was. That will be cool for those who miss the golden age of Twitter. But mastodon has the potential to be something new and different. It's not constrained by what came before.

I agree with this thread for the most part. Except for one important thing. We really have to stop convincing ourselves that the only alternative to extractive capitalism is people overworking for free and burning out. It's not better. In fact it's worse in *almost* every way. This approach might push back against extractive capitalism. But for the people who are actually burdened, it makes their real lives under capitalism worse.
https://oliphant.social/@oliphant/113136636513087913

Throughout history, falsely blaming marginalized groups for the spread of diseases has contributed to hate crimes, ethnic cleansing, genocides, and the holocaust. This kind of rhetoric must never be accepted or normalized.

My decision to go industry-independent unlocked the journey that defined my career for the last ten years.

In that time, CAT Lab has worked directly with communities who donate data, collaborating on research that has measurably made online platforms safer & more understanding for hundreds of millions of people. By working with the public, we can make a difference without having to strike deals with unreliable corporate partners or violating people's privacy. I'm so grateful I made that choice.

So this got some traction overnight. A few details:

Much of the filing hinges on the claim that, through their 2013 code of ethics, the publisher’s industry association agreed to set the price of peer review at zero.

I’m not an antitrust lawyer, but the core case law here seems pretty strong. In particular, in National Society of Professional Engineers, the Supreme Court said “a code of ethics can still be illegal price fixing”: https://scholar.google.com/scholar_case?case=1917928030443042074

The Indian vulture collapse is one of the worst wildlife disasters in history, and I meet people all the time who have never heard of it. In a few decades, India lost nearly all of its vultures. That cascaded into hundreds of thousands of human deaths, billions of dollars in damage, and a cultural loss that can't even really be measured.

My latest for the Washington Post is about the value of vultures, in hopes that we appreciate ours a little more.

#birds #vultures

https://wapo.st/4elK4Sn

#mtbos #EduTooter #edu
https://onedtech.philhillaa.com/p/prospective-online-students-want-async-sync-mix
Phil Hill is amazing and anayltical.
The lede (he also doesn't call it "tl/dr" for the too long, didn't read nonsense)
is that ed tech needs to stop thinking of online learning as synchronous or asynchronous. WE CAN MIX. From the newsletter:
"This conclusion could have been the lede for the survey release. Nearly nine in ten prospective online students would be willing to attend synchronous sessions in an otherwise asynchronous course to improve their ability to learn the material, and most of those would prefer once per week.

I’ll repeat my conclusions from 2021.

I think that the challenge, or opportunity, over the next few years is for schools to figure out how to combine asynchronous methods that preserve anywhere / anytime access with synchronous methods, increasingly with video, that meaningfully increase student engagement. That’s not a new concept, but as William Gibson noted, “The future is already here – it’s just not very evenly distributed.” How do we increase the large-scale adoption of the methods that work? That is the key opportunity."
Anybody want to work with me to do that with math?

Kudos to whoever created this. I wish I could credit them.

I think it's probably about time to make a general announcement that after two INTENSE years sciencing in public I'm going into committed rest, recovery, family & focus time for the rest of 2024 ☺️.

If it's not something already in motion, I will most definitely be slow if not entirely offline on following up on brand new invites, chats, talk requests, collaboration opps, etc. Please know this is circumstantial & I still welcome them (for '25!)

Excited to be resting and (hopefully) creating. 🌲

Today I was prompted (ha ha) by a question Anders Sandberg asked the latest GPT: at what angle should you throw a projectile — subject to uniform gravity, and ignoring air resistance — in order to maximise the length of its trajectory?

[The Chatbot got quite close, but fell over at the end, being too lazy to compute an easy derivative, and then screwing up its numerical solution to a needlessly difficult formulation of the problem.]

It’s well known that to maximise the horizontal distance travelled, you should launch the projectile at 45°. But to maximise the distance along its parabolic path before it hits the ground, the angle turns out to satisfy the beautiful transcendental equation:

s arctanh(s) = 1

where s = sin(θ)

This is solved by:

s ≈ 56.47°

Starting to see people link to perplexity.ai answers in their blog posts instead of Wikipedia even though the perplexity answer is literally just paraphrasing (not even summarising) the Wikipedia entry. 😑

interesting tooling somebody recommended in a chat earlier today: “I spent a couple months building a newsletter generator in the early days of ChatGPT going back and forth in Python. I gave Replit 1 prompt with the same requirements I used with ChatGPT, and it spit out working code in about 15min.” https://replit.com/

This kind of automation (really just the supercharged evolution of autocomplete that has been in IDEs for years) is one area I think LLMs can provide real value in a way that is both familiar (people have been using autocomplete for years) and helps avoid the most common risks that arise when using these things as general knowledge engines.

Help! My granddaughter asks why the spoon in a glass of water looks bigger than in air. How do I explain geometric optics and the reflective index of water to a 2 ½ year old … 😳🙈?
#physics #scicomm

I cannot even believe this misinformation coming out of the Florida Department of Health. This is just beyond the pale. They are discouraging the use of mRNA COVID vaccines. This is beyond irresponsible. It is malpractice.

https://content.govdelivery.com/accounts/FLDOH/bulletins/3b56786