HNHacker News
TopNewBestAskShowJobs

Majromax

3,557 karma · joined February 1, 2019

submissionscomments
Majromax··on MIT asks arXiv to withdraw preprint of paper on AI and scientific discovery
> A weakness that goes hand-in-hand with the lack of peer review

Peer review is not well equipped to catch fraud and deliberate deception. A half-competent fraud will result in data that looks reasonable at first glance, and peer reviewers aren't in the business of trying to replicate studies and results.

Instead, peer review is better at catching papers that either have internal quality problems (e.g. proofs or arguments that don't prove what they claim to prove) or are missing links to a crucial part of the literature (e.g. claiming an already-known result as novel). Here, the value of peer review is more ambiguous. It certainly improves the quality of the paper, but it also delays its publication by a few months.

The machine learning literature gets around this by having almost everything available in preprint with peer-reviewed conferences acting as post-facto gatekeepers, but that just reintroduces the problem of non-peer-reviewed research being seen and cited.

Majromax··on Claude's system prompt is over 24k tokens with tools
> Disney couldn't then claim that Claude does not in fact have permission because Disney's use of the tool in such a way mean Disney now has unclean hands when bringing the claim (or atleast Anthropic would be able to use it as a defence).

Disney wouldn't be able to claim copyright infringement for that specific act, but it would have compelling evidence that Claude is cavalier about generating copyright-infringing responses. That would support further investigation and discovery into how often Claude is being 'fooled' by other users' pinky-swears.

Majromax··on Emergent Misalignment: Narrow Finetuning Can Produce Broadly Misaligned LLMs
> Startlingly, they do not exhibit this behavior when trained on buggy code; only exploit code.

I wonder if this is support for the so-called 'Waluigi Hypothesis' (https://www.alignmentforum.org/posts/D7PumeYTDPfBTp3i7/the-w...). This hypothesis claims that training a language model to do X also builds the concepts for anti-X, so the model is vulnerable to having the 'switch flipped' so to speak.

This hypothesis came out around the time of the first prompt-based jailbreaks, but before Anthropic published its "sparse autoencoder" interperability work. Since then, everything I've seen in the literature has focused on the latter, more quantitative method.

Majromax··on Judge said Meta illegally used books to build its AI
> The RIAA lawyers never had to demonstrate that copying a DVD cratered the sales of their clients. They just got high penalties for infringers almost by default.

The argument for 'fair use' in DVD copying/sharing is much weaker since the thing being shared in that case is a verbatim, digital copy of the work. 'Format shifting' is a tenuous argument, and it's pretty easily limited to making (and not distributing) personal copies of media.

For AI training, a central argument is that training is transformative. An LLM isn't intended to produce verbatim copies of trained-upon works, and the problem of hallucination means an LLM would be unreliable at doing so even if instructed to. That transformation could support the idea of fair use, even though copies of the data are made (internally) during the training process and the model's weights are in some sense a work 'derived' from the training data.

If you analogize to human leaning, then there's clearly no copyright infringement in a human learning from someone's work and creating their own output, even if it "copies" an artist's style or draws inspiration from someone's plot-line. However, it feels unseemly for a computer program to do this kind of thing at scale, and the commercial impact can be significantly greater.

Majromax··on Trump announces 100% tariffs on movies ‘produced in foreign lands’
> There are things to sort out, but it's certainly doable. They can probably use direct COGS to determine the tariff basis.

You could construct a variety of accounting methods to calculate a tax on the 'foreign content' of movies, but how do you actually impose this tax?

A tariff on physical goods is easy. A country requires that goods enter the country through customs facilities, and then the nice customs official doesn't release the thing until the tax is paid. The legal and physical ability to impose these taxes is long, long established.

How do you physically impose a tariff on a movie? If the master is transported physically, what is its value? The value of the fixed copy/master doesn't necessarily include the value of the IP involved, in the same way that a DVD of a $500 million movie might have a retail value of $20. What about movies transferred digitally, since there are no customs checkpoints on fiber-optic lines?

What legal apparatus would be used to impose this tax? Trump is currently getting away with the physical-goods tariffs because the legal infrastructure to collect the taxes is already in place, and remaining legal disputes are just about whether the President can unilaterally set or change tariff rates. If you'd need new law to "tariff movies," then the chance of this whim turning into a real tax drops sharply.

Majromax··on I Tried to Buy an Actual Barrel of Crude Oil (2015)
> Surely there are people trading in these contracts that... don't want their oil delivered to Cushing?

A big-enough buyer will know how to get oil from Cushing to their facility, often by pipeline. One who doesn't really want oil in Cushing is likely to close out their futures trade before the settlement date, treating it like a purely financial transaction.

> It seems like such a bizarre setup to literally require all the oil to come to this one specific town, I assume I'm missing something obvious?

Futures contracts need to be based on the price of something, but the price of a physical good depends on location. Delivery of a barrel of crude to the South Pole would be much, much more expensive – and more variable – than delivery to a big oil terminal. Contracts for physical goods need some kind of agreed-upon reference point, even if most of the time things get financially settled without delivery.

Majromax··on I Tried to Buy an Actual Barrel of Crude Oil (2015)
> All futures should be settled by the actual commodity!

Why? The legitimate hedging role of futures and options is often financial in nature, even for physically-settled contracts.

Take West Texas Intermediate as an example. That's a physically-settled contract, with delivery in Cushing, Oklahoma.

What if I want to lock in a future price of oil but I'm not in Cushing, Oklahoma? Nobody's going to create a liquid futures market with delivery to my loading dock, but most of the time I can get oil on the spot market from a local supplier that already includes/amortizes the transportation cost.

It's far better for me to use the liquid futures market for hedging and still buy on the spot market, closing out the futures contract before delivery. For me, it's as if the futures market is cash-settled, even with a completely non-speculative transaction.

Majromax··on LLM Benchmark for 'Longform Creative Writing'
> You know a burger tastes really good when you eat it, no benchmarks required.

I'd say this is a good example of the opposite, where the problem is finding the quantification of an ultimately subjective experience. Take three restaurant reviewers to a burger joint and you might end up with four different opinions.

Benchmarks proliferate because many LLM domains defy easy, quantitative measurement, yet LLM development and deployment are so expensive that they need to be guided by independent and quantitative (even if not fully objective) measures.

Majromax··on LLM Benchmark for 'Longform Creative Writing'
> Why should I be bothered to read what nobody could be bothered to write?

In an art-criticism sense I broadly agree with you, but I think you go too far from a reader-experience point of view.

> The point of writing is communication and creative writing is specifically about the human experience

That's a point of writing, but the point of reading is only sometimes about communication. It's also about entertainment, enrichment, or expressing half-formed thoughts and feelings.

Take a step back from the autonomously-written novel and imagine something a bit more collaborative. Many players of open-world-ish games develop an emergent story, what if that story could be semi-automatically novelized to document the unique narrative of the playthrough?

SimCity 2000 had "mad-lib" newspaper articles that commented on the city's status; consider ones written by an LLM with full knowledge of the game's state and context.

> something a LLM can mimic but never speak to authoritatively.

Supposing a reader can't tell the difference between an average human-written novel and an average LLM-written novel, where does this authority lie?

Majromax··on Reciprocal Tariff Calculations
> Reciprocal tariffs are calculated as the tariff rate necessary to balance bilateral trade deficits between the U.S. and each of our trading partners. This calculation assumes that persistent trade deficits are due to a combination of tariff and non-tariff factors that prevent trade from balancing.

You know, I need to put a tariff on my local grocery store. They keep taking advantage of me by selling me food, when I export nothing to them!

Majromax··on Trump's Tariffs Wipe Out over $6T on Wall Street in Epic Two-Day Rout
> Industrial policy will always fail if it’s driven by a chimeric political system

I think you're begging the question here that a chimeric political system can craft an effective, long-term tariff policy if it can't craft a long-term industrial policy.

In both cases, the operative phrase is long-term. IF they're to be effective, either set of policies needs to look a decade ahead, and an industrial policy still wins compared to shotgun tariffs.

Majromax··on Trump's Tariffs Wipe Out over $6T on Wall Street in Epic Two-Day Rout
> I started wondering if there's a way for citizens to push the issue that these tarrifs are unconstitutional.

Anyone who imports and is assessed a tariff would have personal standing to sue, arguing that the tax is illegal. I think there are some procedural requirements first to e.g. formally complain, but the dispute would eventually get its day in court.

However, these tariffs are being applied at least under a thin veneer of law regarding declared emergencies. It's clearly beyond the intent of the law for Trump to do this, but it's hard to draw a bright line between this and (e.g.) ambitious environmental regulation.

Majromax··on Trump's Tariffs Wipe Out over $6T on Wall Street in Epic Two-Day Rout
> Trump is crashing the T-note yield to refinance $9T (which is how much US debt is going to mature in 2025) at a lower rate

Even if this were intended (unlikely), destroying $6T of notional stock market value so far to save $270b/yr of interest (if 10-year yields drop by 3%, which hasn't happened yet) is a bad trade.

Majromax··on Trump's Tariffs Wipe Out over $6T on Wall Street in Epic Two-Day Rout
I know that a quotation is not necessarily an endorsement, but this is a fantastically short-sighted idea.

> it becomes increasingly burdensome for the United States to finance the provision of reserve assets

How dare countries want pictures of dead Presidents so badly that... they're willing to send real goods to the United States for its consumption, with no expectation of reciprocation.

> and the defense umbrella

... and agree to a de-facto policy of minimal militarization, such that they could not pose a threat to America's global security interests, no matter what it decides those are.

A few hundred years ago, this would look like a tribute system. It's unfathomable that a nation in such a privileged position would decide it's tired of such a systematic set of advantages.

> as the manufacturing and tradeable sectors bear the brunt of the costs...

What's the saying, they've tried nothing and are all out of ideas? Industrial policy is a known and understood thing. The US has this in its defense sector even it doesn't want to admit it, and the CHIPS act was another attempt for the semiconductor industry.

This sort of policy is far more targeted and far less disruptive than blowing up international supply chains in hope of repatriating sweat-shop sneaker sewers and factories for five cent injection molded widgets.

Majromax··on The Colors of Her Coat
> That’s to say, it may be a legally permitted form of copyright violation, or it may be a legally prohibited form of copyright violation.

Where exactly is the copyright violation supposed to be occurring? Is it distinct from the 'copyright violation' that happens in the brain any time someone learns what a Ghibli frame looks like? What if I study Ghibli in an art class and write notes to myself about the art style, to jog my memory later?

If the problem is that model training is considered a weird kind of copying, what's the legal line between that (allegedly infringing) copy and taking a JPEG and re-encoding it as a PNG?

Majromax··on The Colors of Her Coat
> Since you cannot (it is technically not possible) create an ML model reproducing a style without having it ingest original works in that style

I very much disagree with this. Certainly no gen-AI model currently does this, but you could reinforcement-learn your way to Ghibli style. In language modelling, that's the entire point of the RLHF step.

As long as the style can be described or even recognized by humans, you can nudge your way towards it with multiple-choice sampling.

> such a tool is producing derivative works based on original work (let’s say, by Miyazaki or Studio Ghibli) at scale and for profit, then I have difficulty seeing how copyright law is not involved.

How is this different than a fairground caricaturist that draws subjects in the Simpsons style? They "ingested" lots of Groening-style media, and now they're creating works that are derivative of that style, at scale and for profit.

Learning from a work and being inspired by a work do not create 'derivative works' for the purposes of copyright. The boundary between that and an infringing quotation is fuzzy (note 'sampling' cases in music), but a style itself needs to be on the non-infringing side of things if art is to progress at all, even by humans.

Note that I'm leaving aside the issue of data-gathering; if OpenAI was downloading pirated copies of the Ghibli corpus then that act itself was copyright infringement, independently of the AI training and output.

Majromax··on The Colors of Her Coat
> act so callous towards the copyright of another artist

It's not copyright. Specific works of art are copyrightable, but a style is not. We probably do not want to extend copyright protection to style because its boundaries are fuzzy and because art advances through incremental improvements.

However, we give social value to originality. A human artist who could just reproduce the Ghibli style is 'lesser' than one who creates their own.

This speaks to the tension described by the author. On one hand, we deem that human creativity is worthy of status and that copied styles are 'imitation' or 'knock-off' goods. On the other hand, as consumers we gravitate to the familiar, and success breeds just those cheap imitators.

Majromax··on The disappearance of Gaia, ESA spacecraft will be turned off on 27 March 2025
> Folks don’t appreciate how much energy it actually takes to “fall” into the Earth or even the Sun.

It's not that bad from the Lagrange points thanks to multi-body gravitational dynamics. The Lagrange points form part of the so-called 'interplanetary transit network' (https://en.wikipedia.org/wiki/Interplanetary_Transport_Netwo...) that supports extremely low-energy transfers in the solar system, provided you're willing to wait a very long time.

That being said, there's no reason to send a deep-space probe back to Earth for reentry when it's even easier to send it into a deep-space graveyard orbit of some description with no risk of any potential debris-related problems.

Majromax··on xAI has acquired X, xAI now valued at $80B
> I have no idea what corporate law is like in Texas.

Nobody does, really. Texas is relatively new to its 'corporate registration friendly' game, and at least on paper the statutory law is the same. Texas courts don't have to follow Delaware's case history, but it's also not a given that they will rule in a vastly more owner-friendly manner.

Majromax··on xAI has acquired X, xAI now valued at $80B
I'm not sure the SEC would obviously have jurisdiction, since neither Twitter nor xAI were public companies. The FTC might care about such a merger, however, and I suppose minority investors in xAI would have some ability to file a lawsuit if they think that X was acquired over its fair value.
Majromax··on I asked police to send me their public surveillance footage of my car
> I don't think there should be an expectation of anonymity for the specific case of operating a car on a public road.

Technology allows surveillance at scale, and that attacks privacy in a new and deeply unintuitive direction.

In the pre-information era, people still didn't have a reasonable expectation of privacy in public, in that everything they did could be noted. However, that notice was far from automatic, and people were only remembered if they were memorable. That could happen from doing something or appearing out of the ordinary, or it could happen from specific interest (e.g. being followed).

Nowadays, the surveillance is automatic, and hard drives don't forget. The technology allows aggregators to answer retrospective questions about the target's ordinary behaviour. People now need be concerned about how their actions today will be viewed weeks, months, or years from now if they later attract scrutiny.

This technology is here today for cars and license plates, but it's only a matter of scale before it's applied to pedestrians with face, gait, or other individual recognition modalities. We don't have the social scripts to properly deal with it.

Majromax··on Is it safe to travel to the United States with your phone?
> yes, you will be asked to unlock it

To elaborate, per the CBP (https://www.cbp.gov/travel/cbp-search-authority/border-searc...), if you refuse to unlock your phone or other electronic device it may be confiscated.

If you are a non-citizen, they can also use that refusal as evidence that you are not 'entitled to enter the United States.' In theory that alone is not sufficient to result in denied entry, but in practice it seems that border guards have a great deal of subjective leeway about things.

If you are a US citizen, they cannot deny you entry at the border. However, the device might still be confiscated.

Majromax··on Is it safe to travel to the United States with your phone?
> My gut tells me that the people getting blocked at the border are either running into extremely overzealous border guards OR are making some kind of ruckus anyway and complaining to the media after provoking guards somehow.

My default guess is Goodhart's Law.

The current administration expressly values 'tough' border enforcement, and it publicly uses removals and associated statistics to describe how 'tough' it is being.

Border agents respond to these incentives by looking for any excuse to deny or remove would-be entrants, minimizing the independent judgment and discretion that's an ordinary part of the job.

It's not a 'top down order', but it's an entirely predictable consequence of the administration's messaging.

Majromax··on Apple restricts Pebble from being awesome with iPhones
Battery life. Radios are energy-intensive, and the processing required to run "full apps" is also often pretty steep. In the meantime, while battery chemistries are getting better space is very much at a premium on one's wrist.

From the other recently-posted article, the new Pebble watches are targeting 30-day battery lives between charges. That would be unthinkable for a watch that manages its own wifi/cellular connections.

Majromax··on AI-Generated Voice Evidence Poses Dangers in Court
> I could make plausible sounding recordings that did not reflect reality by, for example, cutting recordings up with freeware like Audacity,

Cutting up what recordings, exactly? If you're mixing-and-matching, you need a pretty broad corpus of source material with fairly consistent recording quality (background noise, etc), and even still you're limited to reproducing words that are already present. You can't cut-and-splice together a recording of me saying 'Yes, dghlsakjg, I embezzled $1 million and blew it on the ponies' because there will be no recording of me saying your username, 'embezzled', or 'ponies.'

The problem comes with the voice-cloning technology that can construct entirely new sentences based on relatively short voice profile samples.

> I've been able to sound like Morgan Freeman using a phone app for at least half a decade

You've been able to sound like Morgan Freeman because of specific, hard tuning work put into the voice changer. Now, you can sound like your boss, or your neighbour, or your ex.

Majromax··on The US stops sharing air quality data from embassies worldwide
> please don't conflate the current administration with the whole American people.

The whole? No, of course no leader should ever be conflated with the whole people of a country. A plurality, however? Definitely.

Trump won an evidently free and fair election, growing his vote-share over essentially the whole country. He has not conducted a heel-face turn since entering office, and his actions are broadly consistent with his campaign statements.

Moreover, Trump isn't the only elected official in the federal government. If Trump were some aberration then Congress would have every ability to thwart his agenda or remove him from office; they have not. The majority-Republican Congress approved even his controversial cabinet nominees on party-line votes, and to my knowledge they have not taken any major investigative or punitive steps since.

A majority of American voters may not have voted specifically for this, but they did vote for the party and candidate who are content to allow this to happen in the supposed service of other goals. Voters who earnestly believe the misinformation shared by demagogues and social media are a tragic case, but that does not excuse voters who 'voted with their wallet,' foreswearing their obligation to the common good in hopes of a tax cut.

Do not mistake me for a rabid partisan. I do not let the Democrats off the hook either for reasons that are another rant and are directionally in alignment with Nate Silver's published opinions.

Majromax··on Electric Propulsion Magnets Ready for Space Tests
> What other space apart from 3 dimensional space exists? And what does that even mean?

Aside from the one and two-dimensional spaces mentioned in a sibling comment (ℝ¹ and ℝ²), there's also spaces with more interesting geometry.

For example, S² is the two-sphere, more conventionally known as "the surface of a ball." On the sphere, the equivalent of straight lines are the great circles. S¹ would be the one-sphere (the circumference of a circle), and S³ would be an interesting, periodic three-dimensional space – kind of like what our universe would be if it were closed rather than (probably) open.

Majromax··on GibberLink [AI-AI Communication]
> A computer should never talk to you first.

Isn't that exactly what happens with every IVR (phone menu) system?

Majromax··on DOGE's only public ledger is riddled with mistakes
> Domestic help was once available because there was an extreme surplus of dirt poor people.

That's precisely the Baulmol effect. When your next-best job is subsistence farming, being employed as domestic help is a step up. When your next-best job is (e.g.) something relatively well-paid in a factory or a phone exchange, being domestic help is no longer so attractive.

> That, or the effect isn't all that it is claimed to be; it's almost tautological that as supply of workers for a low paying job dries up, the wages for the job have to go up to retain workers.

I think we agree here. The mechanism of the Baulmol effect is pretty boring and pedestrian, but the outcome is surprising in aggregate. It's "why are we paying X times more for the same number of teachers?" and "why are there fewer tailors and more fast fashion?" and "why can't I find a handyman who won't refuse to fix something around my house because the job is 'too small'?" all wrapped up in one.

Majromax··on DigiKey's Tariff Resources
> I’d think that low income people would spend a larger proportion of their money on stuff made primarily in the US

The distributional effects are still bad for consumers, even purchasing made-in-America goods. The domestic good will still be higher-cost than the pre-tariff foreign good, so low-income consumers will be worse off than the status quo.

If the tariffs are set at a moderate level (i.e. not so high as to make imports effectively impossible), basic supply and demand suggests that the sticker price of American goods should equal the post-tariff price of imported goods. That's essentially what consumers saw with tariffs on appliances in Trump's first term.

Protective tariffs could still arguably make sense in a demand-constrained economy where there was a lot of unemployment to go around: individual consumers would be worse off but more people would be employed. However, breathless media aside this conflicts with basic statistics that show the prime-age employment[†] rate (https://fred.stlouisfed.org/graph/?g=1DUXA) near cyclical peak levels, currently above the pre-financial-crisis level.

Tariffs are a lousy solution for the more common complaint that stuff is just too expensive.

[†] — I use the prime age employment rate to avoid problems with "discouraged workers" or disability being excluded from unemployment and to correct for basic demographic shifts. If fewer people are employed because more people are retired, that's an entirely different problem.

← PreviousPage 7 of 27Next →