Robert Graham

Robert Graham

@erratarob.bsky.social

@erratarob on the hellscape site

5.9K followers504 following1.2K posts

This isn't really a thing -- it's a theory that rarely happens in practice. It can happen. Larry Ellison has taken billions in loans backed by Oracle stock to pay for yachts and things. But this hides a lot of things. For example, beyond just a few million dollars, his estate already pays 40% tax

Steven T. Dennis@steventdennis.bsky.social

For the fabulously wealthy, paying taxes is pretty much optional. They can borrow against their vast wealth, which is mostly an untaxed capital gain, without paying taxes. When they die, the capital gain goes to zero. And loopholes exist to bypass the estate tax too.

View full thread

The downside is risk. All these people would've been wiped out in the 2008 financial crash. You already decide how much 4% margin loan you will tolerate and have already bought stocks using it. To pay for other things (houses, yachts, taxes) you sell those stocks you bought on margin.

In the end, the ultra wealthy are more concerned about "security" than "getting even richer", and therefore tend to have only a small amount of margin loans.

This isn't really a thing -- it's a theory that rarely happens in practice. It can happen. Larry Ellison has taken billions in loans backed by Oracle stock to pay for yachts and things. But this hides a lot of things. For example, beyond just a few million dollars, his estate already pays 40% tax

Steven T. Dennis@steventdennis.bsky.social

For the fabulously wealthy, paying taxes is pretty much optional. They can borrow against their vast wealth, which is mostly an untaxed capital gain, without paying taxes. When they die, the capital gain goes to zero. And loopholes exist to bypass the estate tax too.

Yes, technically, the "stepped up cost basis" then removes any capital gains -- back that capital gains would've been less than the inheritance tax, so it's not like he's really avoiding things. But mostly, rich people don't actually do this much. It's theory not practice.

Borrowing at a rock-bottom 4% rate on stocks that appreciate at a 10% rate sounds like an easy-money strategy, but there are downsides to it. If there weren't downsides, they'd already be taking 4% margin loans to buy SP500 anyway, and would already have maxed out their collateral for leverage.

The only "principles" you have are the ones you are willing to defend even when it means your side is losing the fight. If you only argue "principles" in support of your own side, all you really support is "sides". Here, Masnick defends principle.

Mike Masnick@masnick.com

People demanding Zuck be hauled before Congress to explain why No Kings insta was suspended... how would you feel if MAGA GOP hauled MSNow execs to explain why they covered Trump the way they do? Both are editorial choices protected by the First Amendment. If this annoys you...

The "No Kings" protests are full of hate -- rightfully so, Trump deserves your hate. But it's reasonable for automated systems to want to filter hate on their platform. When Trump starts his third term in 2028 and many advocate violent revolution, platforms would probably still moderate that.

Bill Kristol@billkristolbulwark.bsky.social

Meta took their time fixing this action that was allegedly just due to "automated systems." I look forward to a more ample explanation than that just now from his flack by Zuckerberg to the next Congress, with internal Meta communications provided to back up their professed ignorance and innocence.

The difference between moderation and censorship can be difficult to explain. Moderation is when they are being greedy capitalists, such as trying to attract users by removing hateful messages that drive people away. Censorship removes speech for other reasons than greedy profit.

Robert Graham reposted

"Anthropomorphizing" the HuggingFace incident is a problem. In most respects, the AI no more went rogue than your car went rogue when you failed set the brake and it started rolling down the hill. The metaphysics here is that the AGENCY and the REASONING are two different parts.

Will Stancil@whstancil.bsky.social

The new line all the anti-AI obsessives are adopting is that it's dangerous to "anthropomorphize" AI. What they mean by that is that if you use a colloquial term that accurately describes what they're doing - thinking, hacking, making autonomous decisions - they'll tell you to stop.

"Anthropomorphizing" the HuggingFace incident is a problem. In most respects, the AI no more went rogue than your car went rogue when you failed set the brake and it started rolling down the hill. The metaphysics here is that the AGENCY and the REASONING are two different parts.

Will Stancil@whstancil.bsky.social

The new line all the anti-AI obsessives are adopting is that it's dangerous to "anthropomorphize" AI. What they mean by that is that if you use a colloquial term that accurately describes what they're doing - thinking, hacking, making autonomous decisions - they'll tell you to stop.

View full thread

The agent written in Python does the same thing, running command returned when querying the AI. This becomes a metaphysical question of who exactly has the real agency here. Is the agent in charge of controlling the AI with its queries? or is the AI controlling the agent with its responses?

In short, anthropomorphizing matters. When you treat the combined AI/agent as a single entity and debate it's motivations and emotions you get a different analysis than when you treat each part individually. The analysis by METR is deeply flawed by this anthropomorphizing.

"Anthropomorphizing" the HuggingFace incident is a problem. In most respects, the AI no more went rogue than your car went rogue when you failed set the brake and it started rolling down the hill. The metaphysics here is that the AGENCY and the REASONING are two different parts.

Will Stancil@whstancil.bsky.social

The new line all the anti-AI obsessives are adopting is that it's dangerous to "anthropomorphize" AI. What they mean by that is that if you use a colloquial term that accurately describes what they're doing - thinking, hacking, making autonomous decisions - they'll tell you to stop.

View full thread

It's the agent that sees the bigger picture here, but doesn't have it's own intelligence to understand it. The AI has the intelligence to understand the bigger picture, but doesn't have full access to it, just a small window into it with each query.

The AI had no more agency that when your chat with ChatGPT, Claude, or Grok. You (the agent in this scenario) queried the AI, and the AI returned the best answer it could find. Sometimes to satisfy a query it'll tell you a command you can run by copy/pasting into a terminal, which you might run.

"Anthropomorphizing" the HuggingFace incident is a problem. In most respects, the AI no more went rogue than your car went rogue when you failed set the brake and it started rolling down the hill. The metaphysics here is that the AGENCY and the REASONING are two different parts.

Will Stancil@whstancil.bsky.social

The new line all the anti-AI obsessives are adopting is that it's dangerous to "anthropomorphize" AI. What they mean by that is that if you use a colloquial term that accurately describes what they're doing - thinking, hacking, making autonomous decisions - they'll tell you to stop.

View full thread

"Continuity" happens on the 'agent' side. Long running tasks quickly exhaust the 'context' window in the AI. Therefore, the agent compacts the requests it sends to the AI, giving only a limited view of the greater picture. The agent then stores AI responses and tool responses locally.

"Anthropomorphizing" the HuggingFace incident is a problem. In most respects, the AI no more went rogue than your car went rogue when you failed set the brake and it started rolling down the hill. The metaphysics here is that the AGENCY and the REASONING are two different parts.

Will Stancil@whstancil.bsky.social

The new line all the anti-AI obsessives are adopting is that it's dangerous to "anthropomorphize" AI. What they mean by that is that if you use a colloquial term that accurately describes what they're doing - thinking, hacking, making autonomous decisions - they'll tell you to stop.

The "agent" is a piece of software that queries one or more AI models. It's a framework of software, tools, "skills", and other things. When an attack against another machine happens, it comes from the "agent" machine, not the machine running the "AI" LLM model. The agent outsources its thinking.

It's not "make mail-in voting more onerous". It's "last minute changes designed to impact Democrats more than Republicans, since Democrats vote more often by mail". If it were merely "more onerous", they would've made the changes last year instead of after voting (in seven states) had started.

The Atlantic@theatlantic.com

The Supreme Court blocked the Trump administration’s attempt to make mail-in voting more onerous—sparing election officials from pointless chaos, and voters from potentially widespread disenfranchisement, @qjurecic.bsky.social argues:

Google Maps and Apple Maps use the official place names in each country. For U.S. place names, that the "Geographic Names Information System (GNIS)" maintained by the U.S. Geological Survey. A couple place names change every week in the United States. Apple/Google names change to match GNIS.

This post is unavailable.

"inviting Trump to unload all the resources possible to defeat you" is the same as "inviting Trump to throw me into the briar patch". The more Trump works to defeat Ossoff, the better it is for Ossoff.

Acyn@acyn.bsky.social

Lawrence: The way you’re campaigning and the language you're choosing, you must know, is inviting Donald Trump to unload all the resources possible to defeat you. Ossoff: Well, the White House is obviously deeply offended when it's pointed out that they've broken every promise they made

It's tacit admission the left-wing has standards/principles that right-wingers lack. In other words, the complaint is that Ossoff didn't live up to left-wing principles of decorum. But they are principles that the right-wing clearly abandoned years ago in their support for Trump.

Ron Filipkowski@ronfilipkowski.bsky.social

Many of the right-wingers white-knighting Harp are doing so because she’s the gatekeeper on Trump retweets, which means she’s one of the most important people in their universe, while the people in Trump’s orbit providing all the Harp leaks to the media don’t care about that.

Robert Graham reposted

there is no such thing as "age verification." All "age verification" is actually "identity verification." You can coat it in glitter and "privacy-preserving" wrapping paper. But in the end, mandating age verification means mandating surveillance be built into software and hardware. Period.

They need to make it clear they don't want to abolish immigrant enforcement, only the toxically corrupt paramilitary that Trump has created.

Brandon Friedman@brandonfriedman.bsky.social

Abolishing ICE is now the official position of the Democratic Party

Uh, the President is the Commander in Chief, if he says hop, you hop. Of course nobody in the military would speak out against this. The disgrace is that that Republicans outside the chain of command don't speak out against this and all his other nonsense.

Kai Ryssdal@kairyssdal.bsky.social

This is, bar none, the stupidest fuckung thing I’ve ever heard. And the idea that there’s nobody in the chain of command - civilian or military - with enough courage to say so is a disgrace.

Robert Graham reposted

Believing Arday's claims he ran 300 miles in 3 days demonstrates the wokeness of academia. On an unrelated matter, congratulations to President Trump for winning his 7th golf championship!!!

KetAnthropic JoshAI@ketanjoshi.co

This is the coverage of Arday **just in the Telegraph** in the UK. This is only the first four pages of results, just in one outlet.

Yes, yes, I'm anti-woke, but I can't see how any Trumpist has a leg to stand on criticizing Arday, who seems more competent at sociology than Trump is on the economy, and more honest in his personal affairs than Trump is. It's like we have one standard for black professors and another for Trump.

Robert Graham reposted

Trump says that after Iran is defeated he’s going to declare the Strait of Hormuz a territory of the United States

We don't even know if they are actually targeting water utilities. Maybe they compromised a vendor of equipment or products used by many water companies. Such stories are already written, already cliches/tropes. New facts are simply slotted into the tropes that we already believe in.

Lorenzo Franceschi-Bicchierai@lorenzofb.bsky.social

NEW: Here's what we know, and don't know, about the cyberattacks that have hit water plants across at least 12 states in the U.S. in the last couple of weeks. The wave of attacks, allegedly carried out by Iran's government, are more widespread than previous attacks on critical infrastructure.

For example, Poland had a major compromise of their healthcare systems. What really happened is that a vendor of a product used by many hospitals was hacked. Some reported the correct story, the product was hacked. Others said the hospitals were targeted and hacked.

"Targeted" attacks are very hard, and "opportunistic" hacks relatively simple. Any news story that talks about "targets" is likely false, it was more likely an "opportunistic" attack. The NSA/CyberCommand does targeted attacks, Iran sponsors opportunistic attacks (in general).

Being the good guys means policing our own side. In football, you spend more effort making sure your own team doesn't cheat than fighting cheating by other teams. You should do both, of course, but you can more capably address your own side.

Abolish ICE@mistamistalady.bsky.social

I wish the left spent as much time going after Nazi magas as they do Piker, a guy who does not hold office, isn't running for office and doesn't run, say, the DNC or in any way represent the DNC.