U.S. drone completes three-day unmanned flight trial The U.S. Navy has completed a successful three-day nonstop flight of a solar-powered unmanned aerial system.

The U.S. Navy has completed a successful three-day nonstop flight of a solar-powered unmanned aerial system (UAS), according to a Naval Air Warfare Center Aircraft Division (NAWCAD)

ukdefencejournal.org.uk/u-s-drone-co...

Fixed.

I asked Gemini deep research about the reasons why European ETFs have such high expense ratios. Here is the (depressing) answer:

docs.google.com/document/d/1...

It's a good read.

As a big fan of Hutter's work, this is now on my reading list: arxiv.org/pdf/2311.12624

Hard to tell in grainy, too much weight on front foot so nose caught water? And wave closed suddenly.

That said: this may be confounded by improved sleep quality and efficiency. I get more deep sleep in the same time, so I am more well rested.

I was surprised++ by the focus effects. I know that as a teen and early 20s I preferred tea because coffee made me task switch too much, so eventually I should experiment w different forms of caffeine, but I really hadn't had expected better focus.

I'll monitor how it goes. I have some weight loss to do, and some amount of cardio to crank up, too.

... sleep while on my caffeine doses, and in spite of the withdrawal headache I have had multiple nights in a row where the proportion of deep sleep was "right" - for whatever values of "right" in the fitbit app.

Aside from the measurable lower systolic blood pressure and significantly increased HRV, I find it *easier* to maintain focus and to give difficult problems sustained attention. Not sure if that is a temporary thing or a result of more efficient sleep (I rarely hit the benchmark goals for deep ...

My withdrawal was pretty brutal -- about 40 hours into it I got a terrible migraine, sweating bullets in cold ambient temperatures etc. - to the extent that some people around me worried I had been taking more severe drugs. 5 days later the withdrawal may be mostly over.

Well-being improvements:

Currents are real, and dangerous. Be careful y'all.

"The past is a foreign country"

HRV up by 50ms (!). Blood pressure down by almost 10 points on the systole.

Kinda wild that quitting caffeine has such a pronounced measurable effect on blood pressure and heart rate variability for me. Way more than I'd expect from the expected value in the literature.

High variance with a genetic component?

Jason ✨👾SaaStr.Ai✨ Lemkin (@jasonlk) .@Replit goes rogue during a code freeze and shutdown and deletes our entire database

This thread is incredible.

"Epstein, Epstein, alles muss versteckt sein" - slight variation on a German children rhyme.

"Epstein, Epstein, alles muss versteckt sein" - slight variation on a German children rhyme.

The difference in spreads, and comfort, and UX, and fees, between IBKR and *any* European bank or broker are just stunning.

IBKR is so much better in literally every possible dimension. Banks are adding so many shadow fees everywhere it is crazy.

Where is the European IBKR?

I mean trying to understand Diracs writing was hard for a bunch of extreme outlier geniuses...

Netherlands rations electricity to ease power grid stresses Country provides early warning for rest of EU if investments in new cables do not keep pace with shift to greener economy

Netherlands rations electricity to ease power grid stresses https://on.ft.com/3TzAqTV

YouTube video by Thomas Dullien [200]*10_7500_0.001_adam

Another training run, 200 neurons in 10 layers trained on 7500 training point. Reaches near-perfection after about a minute, then around 4:30 continued training makes it blow up & diverge into psychedelics:

www.youtube.com/watch?v=lcyt...

YouTube video by Thomas Dullien [400]*20_2500_0.001_adam

I finally managed to automatically upload some more YouTube videos of training neural networks.

This one here tries to overfit 20 layers of 400 neurons on ... 2500 data points. Instead, we get divergence of training and psychedelic graphics: www.youtube.com/watch?v=KpC3...

Thanks for tagging me, good points!

... understand/position something that speaks our language as if it were one of us.

it's a like hearing someone talk with a rich vocabulary, and assuming they are an expert in any field.

Curious why the author didn't challenge the requirement for these systems to adopt a human like persona(*). 4/*

Many risks associated with LLMs, stem from the fundamental problem of developers, users, commissioning authorities, and the media - ascribing and assuming human attributes and abilities to LLMs when there is no evidence of such things. nut as social creatures we have been conditioned to ... 3/*

Understand Neural Nets better, post 5 of N -- Code Assistant shootout In a series of previous blogposts [ 1 , 2 ,  3 , 4 ] I ran some experiments drawing the boundaries of the polytopes generated by a fully-con...

I had a particular optimization for my NN training viz in mind so I asked Claude Code and Gemini CLI separately to implement it.

It worked, mostly, and my code is faster and new questions about training dynamics emerge.

addxorrol.blogspot.com/2025/07/unde...

My data visualisation Python code does about 16 TFLOPs on this H100, and ... as someone whose formative years in computing were 1996-2006 or so, that is absolutely wild to me.

I'm essentially using the compute of the #1 supercomputer in the world when I graduated high school to draw some pics.

AI security notes, July 8th: Improving our (poor) ability to forecast AI so we can do better AI security work AI is changing so rapidly that to invest in security for AI, and AI for security, requires forecasting where AI itself will be in at least the short and medium term.

Great read: open.substack.com/pub/joshuasa...