sxp

The article is forgetting about Anthropic which currently has the best agentic programmer and was the backbone for the recent OpenClaw assistants.

> "White House launches direct-to-consumer drug site..."

> "The site is not selling drugs directly to American patients..."

Just another layer of middlemen. They should go with the proper free market option and allow Americans to buy medication from other countries.

Reading some of the other articles on that site, it's unclear how scientifically sound the original article is. A quick Google search gives different radiocarbon dating for the landslide https://www.sciencedirect.com/science/article/abs/pii/S01695...

I don't know enough about the event to figure out the likelihood of either hypothesis, but this other data point is something to keep in mind.

Claude didn't follow your "Every line must earn its keep. Prefer readability over cleverness. We believe that if carefully designed, 10 lines can have the impact of 1000." from https://github.com/quantbagel/gtinygrad/blob/master/AGENTS.m... given how bloated this demo is.

https://blog.evjang.com/2019/11/jaxpt.html is a better demo of how to render the Cornell Box on a TPU using differentiable path tracing.

> Naina Raisinghani, 00 needed a name for the new tool to complete the upload. It was 2:30 a.m., though, and nobody was around. So she just made one up, a mashup of two nicknames friends had given her: Nano Banana.

Ah, that explains the silly name for such an impressive tool. I guess it's more a more Googley name than what would have otherwise been chosen: Google Gemini Image Pro Red for Workspace.

> And for San Franciscans, the price on the menu is rarely the price you pay. Add in sales tax, a default 18-20% tip on the tablet screen, an SF Mandate or Cost of Living Fee, and 8.625% sales tax. Soon that seemingly cheap $15 lunch might be $20.

This is a key metric that the article doesn't properly account for when it comes to food prices. The asshole restaurateurs got an exception to the anti "drip-pricing" law that required all fees to be rolled into the listed cost: https://oag.ca.gov/hiddenfees

The AI bits look interesting, but the laws appear vague and toothless. E.g, "identify every specific artificial intelligence program used" means all the reports will say “This report was written either fully or in part using artificial intelligence.” without any useful details.

> Artificial Intelligence The popularity of artificial intelligence (AI) has grown rapidly in the last several years and the widespread use of generative AI, or AI that can create original content, presents new legal considerations.

> With the passage of AB 316, a defendant may not say artificial intelligence that they developed, modified, or used that is alleged is to have caused harm to the plaintiff did so autonomously.

> Additionally, law enforcement agencies will need to identify when artificial intelligence was used in official reports and the type of program they used (SB 524).

https://leginfo.legislature.ca.gov/faces/billNavClient.xhtml... https://legiscan.com/CA/text/AB316/id/3223647

There have been hedge funds and other investment groups that claim to use RV. (At least according to Google and pro-RV books.) They also claim to be able to beat the market. Based on those anecdotes, I think any hedge fund that thinks they can use RV will publicly brag about it to get more investors.

None of these groups can replicate their results beyond the initial claims. This is strong evidence that positive results in RV are just due to selection bias, specifically https://en.wikipedia.org/wiki/Publication_bias. If those investment groups could actually replicate their results, they would still be major names and others would be actively trying to copy them since it should only take a couple of millions of dollars to find capable RV candidates.

The non-skeptical view is that if people try to predict the stock market via RV, they will interfere with the future and their prediction ability will decrease. But when weighing this hypothesis against the hypothesis that RV is just selection bias, the latter wins due to Occam's Razor.

For those who want to consider the mundane explanation about remote viewing: remember that humans can still be surprised by the birthday paradox and other synchronicities that occur when sampling data at large scale. So across thousands of sessions, the weak coincidences documented by the Stargate project are expected.

For those who want a skeptical & cynical view: if remote viewing works, it would be part of the standard strategy of every hedge fund. Remember that theses are groups who pay millions for millisecond advantages in information. And you only need an ~51-55% success rate to make a killing in HFT (vs a 50% success rate from a coinflip). The fact that hedge funds don't have remote viewers on staff is evidence against RV providing utility greater than an RNG.

And for curious people who want to try a scientific approach, I suggest joining https://www.social-rv.com/ which is collecting data about RV and trying to make the experiment ironclad via blockchain authentication of predictions.

One option is to find a good doctor/psychiatrist/therapist and try to get diagnosed. (You can also try to self-diagnose via the Interne, your favorite LLM, or just skimming the https://en.wikipedia.org/wiki/DSM-5, but that's a risker option due to false positives and noise.)

If you do have a well-studied issue such as https://en.wikipedia.org/wiki/Autism or https://en.wikipedia.org/wiki/Schizoid_personality_disorder that interferes with socialization, you can find support groups and people who have been able to succeed even with those issues.

Some of these social problems (e.g, ones caused by missing childhood socialization experiences) are curable through practice, but others will require you to learn enough about human social behavior to mask until someone figures out a cure. It can be exhausting, (just like physical exercise can be exhausting) but it's a useful skill to have in order to do well in life.

Good luck.

Between ffmpeg and qemu, I always think of https://xkcd.com/2347/ when I see Fabrice's work. Especially since ffmpeg provides the backbone of almost all video streaming systems today.

"Google kills Gemini Cloud Services (killedbygoogle.com) 530 points by dang_fan 15 hours ago | hide | 330 comments"

Ha! Is Gemini suicidal?

> At a checkup a few years ago, a doctor told me I was deficient in vitamin D. But he wouldn’t write me a prescription for supplements, simply because, as he put it, everyone in the UK is deficient. Putting the entire population on vitamin D supplements would be too expensive for the country’s national health service, he told me.

Ugh. It's amazing how incompetent medical systems are. I was also deficient in vitamin D and my doctor wrote a prescription. When I did the math, the cost was something like >$.10 per 1000IU. But if I bought the vitamins from a normal store, I would pay <$.01 per 1000IU. Since a person lacking sunlight only needs 1000IU, the price for giving everyone in the UK Vitamin D would be <$700k/day. And probably much less since most people won't need this high of a dose and bulk quantities would be cheaper.

For healthy people, taking extra vitamins is pointless, but giving them to people who are deficient in vitamins is one of the cheapest health interventions for the benefits.

PSA: if you're feeling off, make sure your doctor checks your various vitamin levels and see if cheap OTC vitamins help.

The title sounds like speculative clickbait.

From https://www.airbus.com/en/newsroom/press-releases/2025-11-ai...:

  Analysis of a recent event involving an A320 Family aircraft has revealed that intense solar radiation may corrupt data critical to the functioning of flight controls.

This is different from the core claim that the incident was caused by radiation. What are the prior probabilities that the system was exposed to "intense radiation"? Vs some other mundane cause such as a faulty wire or mechanical issues? And what is the evidence supporting the former hypothesis?

His previous show, https://en.wikipedia.org/wiki/The_Problem_with_Jon_Stewart was canceled by Apple because he wanted to criticize AI, Israel, and China. Unless he wants to go the route of a fully independent podcast (which he already has), he's going to have to bow down to a political master at some point.

Note that their paper includes p-hacking so this is proto-science rather than science. They didn't find data to match their hypothesis, but they were able to find a hypothesis to match their data.

> Follow-up secondary analyses were then conducted to examine in more granular fashion the timing of the association between nuclear testing and occurrence of transients. Table 2 summarizes the association between occurrence of transients and different time windows relative to nuclear testing, ranging from 2 days before a test until 2 days after a test. The only association that reached statistical significance was for the association in which transients occur 1 day after nuclear testing.

To put this in context, it takes about 1 gallon of water per 1g of almond [1]. And in California's dry climate, this water comes from groundwater that doesn't renew as fast as it's depleted. So the next time someone chastises you for your non-low flow showerhead that uses more than 2 almonds of water per minute, remember to put the numbers in context.

1. Numbers from a pro-almond group: https://www.almonds.org/why-almonds/growing-good/water-wise

Another option is to use MathJax and an online editor to convert LaTeX to SVG. E.g, https://latexeditor.lagrida.com/

And if you want to mix it with formatted text or HTML, Markdeep is my preferred system for docs: https://casual-effects.com/markdeep/features.md.html#embedde...

The large differences between gemini-2.5-pro and the gemini-X-flash and gemma models is surprising. It looks like distillation causes an ideological shift. Some, but not all of the other distilled models also show that shift.

Another place where LLMs have a problem is when you ask them to do something that can't be done via duct taping a bunch of Stack Overflow posts together. E.g, I've been vibe coding in Typescript on Deno recently. For various reasons, I didn't want to use the standard Express + Node stack which is what most LLMs seem to prefer for web apps. So I ran into issues with Replit and Gemini failing to handle the subtle differences between node and deno when it comes to serving HTTP requests.

LLMs also have trouble figuring out that a task is impossible. I wanted boilerplate code that rendered a mesh in Three.js using GL_TRIANGLE_STRIP because I was writing a custom shader and needed to experiment with the math. But Three.js does support GL_TRIANGLE_STRIP rendering for architectural reasons. Grok, ChatGPT, and Gemini all hallucinated a GL_TRIANGLE_STRIP rendering API rather than telling be about this and I had to Google the problem myself.

It feels like current Coding LLMs are good at replacing junior engineers when it comes to shallow but broad tasks like creating UIs, modifying examples available on the web, etc. But they fail at senior-level tasks like realizing that the requirements being asked of them aren't valid and doing something that no one has done in their corpus of training data.

>Finally, one winter morning on the school bus, he turned on his tormentors. Curling his fingers in the shape of a pistol, he said, “I hope you all die.”

> Mike Carinci, the school resource officer — a member of the sheriff’s office who worked in the school — viewed and listened to hours of video from the bus, seeing the level of abuse directed daily at the student. “Just horrible things, like nonstop,” he said. Mr. Carinci summoned the students and told them that the bullying had to end. The superintendent told them that they could be suspended or expelled.

> The school traced the “hit list” rumor to a girl who admitted making it up. This quieted the community.

This article makes it sound like the only one who was punished was the victim of the bullying for his emotional outburst and everyone who picked on him got away unscathed. This seems similar to the recent Netflix miniseries Adolescence. Both the series and the discussion around it focused on the main character rather than the bullying that caused him to kill.

There is a strong chance that there are no actual "drones" involved and these sightings are just due to random people mistaking airplanes as drones/UFOs: https://x.com/MickWest/status/1971656713230270794

When people have analyzed the flight logs, the sightings match with known planes: https://x.com/ThomasH_Synth/status/1972236703864586463

https://x.com/MickWest is a good source to follow since he and others are willing to actually track down the data related to UFOs.

This seems completely unrelated to Google’s nano banana. Is it using the Google API or is it just piggybacking off the name?

Only because of RLHF instructed them to do so. Prior ones without this training responded differently: https://en.wikipedia.org/wiki/LaMDA

> Put simply, intelligence is all about doing things, while consciousness is about being or feeling.

Unless one believes in p-zombies or a magical soul, robots & LLMs can "be" and "feel". We can distinguish LLMs which "are" from random noise which "isn't". And multimodal LLMs & robots have sensory inputs.

One can always make up some untestable notion of "consciousness" and then say that LLMs don't have it without being able to define which humans (i.e, what level of functioning brain between adult, child, fetus, zygote, corpse, etc) are conscious vs which are not. If one arbitrarily draws a line somewhere, then it's just as valid to arbitrarily draw the line somewhere else.

https://thelastdriverlicenseholder.com/2025/02/03/2024-disen... says it's about 40% as good as a Waymo if you use disengagement as a metric.

As far as I can tell, OP's argument is bullshit. I did a bunch of research into max contrast displays and haven't found any reason to not use them. There are reasons to not use max _brightness_ since a bright screen can be problematic in the dark, but it's better to use max contrast on a slightly dim screen than to use max brightness with a slightly lower contrast.

There was some minor effects for people with dyslexia and high contrast, but no science-backed a11y guidelines provide evidence of reducing contrast for the general population. And since most people have crappy cheap screens instead of fancy Macbooks that UI designers use, reducing contrast makes it worse for them. Particularly at off-screen angles.

I'm not sure who started this UI trend of grey-on-gray, but it's shitty and should be avoided unless there is an intentional stylistic reason to sacrifice form for function.

The problem with this is that the verifier (Bank, DMV, Government, etc) can track that you went to gambling.com. E.g, maybe the gambling.com string is "gambling.com-123". Or if it's a random number, maybe gambling.com is storing it somewhere and the bank/government files a legal request to get a copy. You have to trust the government in this case which isn't ideal.

A better one is Chaum's Ecash protocol: https://en.wikipedia.org/wiki/Ecash

To use a metaphor for that protocol:

  1) The gambling site gives you a piece of with a UID on it.
  2) You put the paper in a opaque envelope. (This is a cryptographic way of hiding the contents)
  3) You take the envelope to the bank (or DMV, police station, etc) along with a copy of your birth certificate/RealID and they emboss it with their stamp saying you are over 18. The embossing is transferred through the envelop to the paper, but the bank hasn't seen the contents of that paper. (This is a cryptographic signing method which can affect the UID in step 2 without the bank needing to see the UID: https://en.wikipedia.org/wiki/Blind_signature)
  4) You take the paper out of the envelope and destroy the envelope. The paper with the UID has the stamp but doesn't have any tracking information that the bank might have put on it.
  5) You give the UID to the gambling site. They see that you got it stamped by the bank so they know you have an account with the bank and are over 18, but they don't know the RealID that the bank saw when the bank stamped the envelope.

This is secure because the bank never saw the UID so they don't know if it was gambling-123 or disney-123. The gambling site can save that stamped UID and give it to the bank (or government) if required, but bank can't figure who came in to get that UID embossed. The only person who knows all the tracking information is the user. And as long as they burn the envelope (which is cryptographically secure), there is no usable tracking information.

As long as the bank is Good and uses the same stamp for all users (i.e, they don't use alice-stamp, and bob-stamp, etc for different users), there is no way for anyone to connect that Alice got her gambling-123 UID stamped. But this stamp is normally using PKI so anyone can check the public key of the stamp.

This algorithm was originally conceived to create anonymous "cash" since the bank would charge $1 to stamp your envelope and the gambling site could sell their UID paper to the bank to get that $1 from the bank.

You're right. It's clear I should have had an LLM write my comment rather than do it myself before a cup of coffee. I've already spent an order of magnitude more energy thinking about this article compared to an LLM.

Also, since I live a first-world life style which consumes multiple KW of power, I've probably consumed multiple orders of magnitude energy more than an LLM on this topic.

Since a human uses ~100W of power, the .24Watt-hours of energy for an AI prompt is about 40human-seconds [Edit: 9human-seconds] of energy.

And unlike the human who spent multiple hours writing that article, an LLM would have linked to the original study: https://services.google.com/fh/files/misc/measuring_the_envi...

[ETA] Extending on these numbers a bit, a mean human uses 1.25KW of power (Kardashev Level .7 / 8 Gigahumans) and the mean American uses ~8KW of power according to https://en.wikipedia.org/wiki/List_of_countries_by_energy_co.... So if we align AIs to be eco-friendly, they will definitely murder all humans for the sake of the planet /s