panarky

Then I tell Opus to read the audit report and implement what it agrees with.

Flash is really good at this, and it is blazing fast in Antigravity CLI. Easily 10x faster than Opus.

Can't wait to try 3.8 Flash. If it's good enough, maybe I'll switch Flash to primary and make Opus the auditor.

You say landlords should only have accountability for the crimes of their tenants if they chose to do nothing to prevent their property from being used for crime.

Google doesn't just ignore bad advertisers, they prevent the vast majority of malicious ads and report the worst offenders to law enforcement. Google is already doing far more than what you say immunizes landlords.

You say telcos are in the clear because they give info to law enforcement on demand.

Google also does this, so according to your goalposts, Google is also in the clear.

You say Meta is in the clear because they have been sued for crimes their platform enabled.

Google has also been sued and paid billions in damages.

The standard I'm asking you to generalize is the one I replied to: "hold Google directly accountable for every malicious ad they allow through their network".

If this standard can't generalize to landlords, telcos, Meta and everyone else who has a platform that bad actors abuse, then it's a bogus standard.

Every landlord should be directly accountable for all crimes committed by their tenants.

Every telco should be directly accountable for all crimes committed by their customers.

Meta should pay damages when their users stalk and harass people.

You are undoubtedly correct that at some point there will be new leadership in the US.

Politics makes leadership change a possibility. Biology makes it an inevitability.

But I don't think the evidence is very strong that switching from one man to a different man, even if the new man wears a blue hat instead of a red hat, will make that much difference against capital and its surveillance state.

This isn't a false equivalence "both sides" argument. I'd greatly prefer the blue hat over the red hat.

But the blue hat only makes the underlying forces of late capitalism a little slower and a little less vicious, while simultaneously legitimizing that system.

But in the early days of prestressed concrete (1920, 1930), a single civil engineer performed all of these functions.

The point is that the role of the modern civil engineer has changed and specialized over the decades. You can be a very competent civil engineer without having the knowledge or experience needed to perform every specialty.

Just because our ancestors in the 1940s handwrote binary sequences, and in the 1950s they handwrote assembly, and in the 1960s they handwrote FORTRAN, doesn't mean that today's software engineers need to be able to do any of that.

I'm sure you also realize that when a large holder sells a significant stake of any asset, traders anticipate they will sell more in the future.

Example 1: Berkshire Hathaway selling $47 million of BYD and the market price dropping 8% the next day, wiping out $10 billion of market value.

Example 2: Musk selling 0.6% of his stake in Tesla, and the market price dropping 15% the next day, wiping out $187 billion in market value.

Example 3: UK government announcing a planned sale of 0.3% of above-ground gold, and the global gold price plunging 10% before the first auction.

Ah, here it is https://bpi.com/missing-factors-in-the-cfpbs-analysis-of-ris...

When bond prices drop, US interest rates go up, which hurts the real US economy.

And when bond prices drop, that devalues the remaining 90% of China's holdings, hurting China too.

The Fed could stabilize the bond market by printing dollars to buy the bonds itself, but that would devalue the dollar and drive up inflation in the US. When China tries to repatriate that wealth, they must sell dollars and buy yuan, which would drive up the value of the yuan, driving up the cost of Chinese exports, hurting their manufacturing sector.

It's mutually assured destruction.

Convenience spending by "transactors" (roughly 35% of cardholders) who pay in full every month is something like $200 billion of this.

That's why China will never do this.

Better for both nations to extend and pretend.

Then add unfunded liabilities like pension benefits and healthcare promises for every federal, state and local government, school district and corporation.

A trillion of credit card debt is something like one half of one percent of total indebtedness.

There's definitely value in the "meta-cognition" that an LLM can do even if it is blind to the underlying weights and layers of the model itself.

Not only is search revenue growing, but it is growing at an accelerating rate.

At the same time, operating margins are expanding.

I don't know the name of the logical fallacy where someone personally uses an LLM instead of Google Search and then infers that the search business is dying, without ever reading a financial statement.

A sensational statement that is utterly false.

Per 100g, Norwegian farmed salmon has 0.1x the sodium, 0.5x the saturated fat, 2x the protein, 20x the Omega-3 fatty acids and infinity-x the vitamin D of your hamburger.

Farmed salmon is a minimally processed single-ingredient protein, while your hamburger is ultra-processed and packed with refined carbs, added sugar and high-fructose corn syrup and chemicals synthesized in a lab.

> Children and pregnant women should also limit the intake because it can disrupt development.

Presumably you're talking about growth hormones?

Because that American fast-food hamburger has plenty of growth hormones.

Under Norwegian law and European Union/EEA regulations, the use of growth hormones in fish farming is strictly illegal.

Instead of taking syntactic shortcuts, the richer abstractions learned from multiple languages, and code, and math, and images and audio, ultimately make English comprehension and reasoning far stronger.

If you want citations, ask your favorite LLM how linguistic diversity prevents "surface memorization", overfitting on surface-level English patterns instead of representing the deeper concepts in latent vector space.

The results for English and Python are much better because the model is also trained on Mandarin and Greek and Lisp even if you never make a request or receive a response in Mandarin, Greek or Lisp.

Interestingly, the 10x shares held by Brin and Page constitute only 11% of the economic value of Alphabet, but 51% of the voting control.

And the 10x shares automatically convert to 1x shares upon transfer or inheritance, so if the founders cash out or die, the 10x supervoting power disappears.

One exception is GOOG / GOOGL which both trade actively, and there's not much difference in price.

The mechanism for a price divergence could be accumulation of the 10x shares to seize control, but even if you could buy the entire float, it wouldn't be enough to take control, so that mechanism never happens.

they are not humans

they don't think

they don't 'know'

antirez never says they are or do any of that

at least an llm can write a response that is relevant to its context, your non-sequiturs can't even do that.

I want them to be strongly biased in favor of free expression.

But they take my money and use it restrict how people dress, how they pray, what customs they are allowed to follow, what they teach their kids.

The agenda they support is anti-free-expression, and that's worse than neutral.

I don't know if Google would be net-good or net-evil, but I'm pretty sure they would be far less evil than most of the other large corporations on that list.

Certainly less evil than Exxon Mobil, Microsoft, Saudi Aramco, Meta, JPMorgan Chase, UnitedHealth, Coca-Cola, Oracle, Palantir, Goldman Sachs, LVMH, McDonalds, etc.

But it's a paradox that on HN, Google gets far more hate than any of these.

Maybe HN is a weird combination of fatalistic and idealistic, where we assume any for-profit corporation with a do-gooder mission is lying about its mission to hide its essential evilness, while secretly hoping that maybe it really is possible to do good and do well at the same time.

For a while it seemed like Google's do-gooder mission was genuine. Unlike their less principled competition, Google refused to take money to influence search rankings. Google withdrew from mainland China rather than continue censoring search results there.

We secretly started to believe. But all the other compromises along the way felt like betrayals.

Google really is less evil, even today, but we hate them more because we dared to believe, and Google let us down.

Exactly. I don't have to write binary machine code directly, every zero and one artisanally crafted by hand, to have thought deeply for years about a how to solve a problem.

In fact, choosing the right level of abstraction is essential to my ability to solve the problem.

For most problems, the friction of writing binary code by hand is the wrong level.

And we're discovering that many important problems can be solved faster and with greater quality than can be achieved by dogmatically hand-writing every line of source code just for the friction.

I tried building something similar to make 72-hour predictions about the US war on Iran, but found that the persona subagents were far too naive. They believed official statements and media reports at face value, they failed to read between the lines or apply principles of bounded distrust. They accepted spin and wartime propaganda and didn't give enough weight to underlying incentives. They didn't learn from their mistakes from earlier rounds or downgrade their trust in sources after their statements were repeatedly proven false.

It must be possible to improve accuracy with memory, system prompts, progressively changing subagent weights based on historical performance, etc.

I found it helpful to allow the personas to talk to each other. A pure weight of 20% each for 5 personas that are blind to the arguments of the others didn't work as well as personas that modified their rationales after reading the output of the others.

After each prediction resolves, I would have each persona create a post-mortem analysis of what they got right and wrong. Maybe visibility into prior post mortems of their own persona and those of others on the board could allow them to recognize historical cognitive biases and recalibrate for the next prediction.

Presumably the hosted version will have a leaderboard of some sort. Each board might not be able to cheat, but if users can cheaply create many sockpuppet boards, you'll see the Baltimore Stockbroker Scam emerge on the leaderboard. If the public attestation is to be meaningful, it must be difficult and expensive to create new boards.