It’s exactly true only for the easiest failure mode: obvious junk. Nobody needs an AI detector to reject garbage. Kernel maintainers have been rejecting human-generated garbage since before LLMs darkened the software dev community's skies. The hard problem isn’t slop. The hard problem is credible-looking patches that meet the immediate spec, match local style, compile cleanly, and still encode a subtle bug or a long-term maintenance tax.

If all you care about is trash vs not-trash, sure. But kernel hygiene is not a single trash can. When a new tool changes the volume and the shape of incoming patches, maintainers need to know what they’re dealing with so they can adapt their defenses: stricter expectations on rationale, smaller patches, clearer commit messages, more benchmarks for perf-sensitive code, and less tolerance for “it works on my machine.” That’s why the ongoing discussion is leaning toward tool disclosure and “show your work,” not a futile attempt at mind-reading whether a bot wrote the diff. Also, provenance isn’t about blame. It’s about correlated risk. If a model tends to repeat certain patterns (same error-handling shape, same locking assumptions, same “looks right” idioms), you can get a whole class of bugs that are individually non-obvious but statistically related. You don’t get to manage that if you deliberately blindfold yourself.

Define “slop” without smuggling in the answer. Is slop code that fails checkpatch? Easy. Is it code that’s correct but needlessly generic, future-proofed into unreadability, and quietly hostile to the next maintainer? That’s the kind of quality failure that doesn’t show up in a diff review unless the author can justify every tradeoff. And that loops back to the real point: the kernel review process assumes the submitter can defend design choices and follow up when reviewers poke holes. An LLM can only participate in that social contract indirectly; if a human can do that, great, then they’re the author and the tool is irrelevant. If the human can’t, the patch is a liability, even if it isn’t AI slop.
 

[Reply to ConceptJunkie]

That binary sounds great until you remember what “standards” actually means in kernel-land. It’s not just “passes tests” or “meets the spec.” The spec is the easy part. The standards also include: does it fit the subsystem’s design, does it avoid cleverness debt, does it behave across a zoo of arches/configs, does it keep the fast path fast, and can future maintainers reason about it without taking up candlelit lockdep seances.

Also: kernel review isn’t a theorem prover. It’s a risk-management pipeline with finite reviewer attention. “Meets the standards” is often a judgment call made under time pressure, not a formally verified conclusion.

It matters a lot, and you accidentally picked the perfect trio to prove it.

If Linus writes it, Linus can explain it, defend it on the list, revise it when a maintainer says “no, not like that,” and own the fallout for the next decade. Aliens can’t answer review questions. An LLM can’t show up on LKML and say “good catch, here’s why I chose this memory barrier, and here’s the perf data on Zen4 vs Graviton.” Origin matters because accountability matters.

The kernel doesn’t merge diffs. It merges an ongoing relationship with an author who can justify tradeoffs and do follow-up when reality punches the patch in the face. That’s not politics, that’s maintenance. And it’s exactly why the proposed guidance keeps circling around transparency and “make it easy to review,” rather than pretending there’s a magic AI detector.

Good. That’s the only sane way to use any generator, including StackOverflow and “I found this gist on a blog from 2013.”

But “100% understand” is where the wheels start to come off. You think you tested it thoroughly, and it is this kind of innocent arrogance that stops a coding career in its tracks. In kernel code, you can understand what the lines say and still miss what they do when the scheduler, the memory model, the compiler, and three architectures start arguing in the hallway. Races, refcount lifetimes, RCU subtleties, error paths, and performance cliffs do not politely announce themselves during your “thorough testing.” Even experts rely on collective review, fuzzing, CI farms, and years of scar tissue because humans are not exhaustive-state-space machines. To claim you you've thoroughly tested anything is arrogance, not expertise. And here’s the AI-specific twist: LLMs are great at producing code that looks like something a careful person would write. That’s not “slop.” That’s plausibly-correct code that can sail through casual review and still be wrong in the exact corner you didn’t think to test. The dangerous patches are the ones that look boring.

I think you’re describing the top slice of kernel contributors and then declaring policy based on the best-case. The kernel also has drive-by patches, corporate throughput patches, newbie patches, and “I fixed my one bug, I give zero shits about the downstream" patches. I'm guilty of this last one; I connected a 7.1 surround system via HDMI to my 4090, and watched my display go dark. Why did I have to go under the hood? Because EDID—the ancient, flaky, and apparently immortal Display Data Channel protocol HDMI still uses— for the sake of compatibility lets any device scream "I'm a monitor!" loudly enough to hijack your system, even if your main display is DP. The kernel insisted that the EDID signal was the main display, so it kept trying to throw the desktop to the phantom VGA display that EDID said existed on my LG soundbar. I couldn't reprogram the soundbar; instead I did brain surgery on drivers/gpu/drm/drm_edid.c to filter out my particular soundbar. Works for me; absolutely will not work for 99.99% of the rest of the planet. I'm a sysadmin, not a coder, but I do have a BS in CS, and I can code when I have to. Kernel maintainers could obviously come up with a better hack than I did, but this is a corner case that will never be loud enough to draw the main line's attention. I had to lean heavily on an LLM to even diagnose this problem; but once it converged, it gave me the root cause, and even helpfully suggested the code I needed to patch into the kernel. It took a few tries -- my C fu was never strong, but it *worked.*

In a bucket, if you want an AI posture that’s actually compatible with kernel reality, it’s this:
If you can’t explain it, you don’t own it.
If you don’t own it, it doesn’t belong in mainline.
Tools are fine. Epistemic outsourcing is not.

So it looks like the editor sprinkled the magic word “AI” on the headline and slashdot did what it always does: auto-spawned 200 trollish variations of “lol AI hype.” Cute. Meanwhile, the actual story here is a lot more practical than the kneejerk anti-AI trolls realize.

What Zanskar is using machine-learning models to spot blind hydrothermal systems: reservoirs with no obvious surface tells. No hot springs, no geysers, no “hey look, free steam!” signpost. In other words: no leaking clues. Humans have traditionally hunted where the geology is loud. Zanskar is trying to hear the quiet stuff, then drill to confirm.

And once you confirm it, the extraction is basically the normal geothermal playbook: Find and a likely spot->drill production well(s)->bring up hot fluid->strip heat at the surface->reinject the cooled fluid back underground.

So the key tradeoff vs traditiona” geothermal isn’t new extraction vs old extraction. AI reduces exploration risk (fewer dry holes), but you still face the classic geothermal buildout grind: drilling cost, reservoir management, cooling choice, permitting, interconnection queues, etc. Zanskar’s bet isn’t new thermodynamics. It’s “we can de-risk the needle-in-a-haystack exploration phase."

Now here's the part where Arizona, Nevada, and Utah start side-eyeing the whole thing. If this hydrothermal renaissance turns into “power for data centers, paid for by sucking from the last puddle,” that’s not clean energy. That’s just a different flavor of externalized cost, and those of us living in the waterless paradise that is the desert Southwest get to pick up the tab. The good news: geothermal doesn’t have to be a water vampire. Many systems reinject what they produce. The risk knobs are mostly water recovery -- especially surface evaparoation in wet-cool loops, which are the most likely (read: highest profit margin) system designs,

My hope, and my ask, for anyone deploying this in the West is to be smart about it. Treat potable groundwater as off-limits unless there’s no alternative and it’s transparently justified. Prioritize closed-loop/reinjection-heavy designs and aggressive leak accounting. Use non-potable sources for any makeup water (brackish, treated, industrial) whenever possible, and pick cooling systems with the desert reality in mind, not the pie-in-the-sky brochure pitch to investors. Last but not least -- monitor and disclose environmental impacts like you actually have to live here afterward.

Using AI to find the needle in the hydrothermal haystack? Absolutely. That’s a sensible tool applied to a hard search problem. Just don’t let “AI found clean energy” become the preface to “and then we sucked your groundwater reserves dry to power yet another chip fab/AI datacenter in Phoenix.

Thanks for the thought-provoking post. I’m one of those people who says “please” and “thank you” to chat bots, and I don’t find it weird at all. For me it’s mostly habit transfer from human conversation. I’ve always written that way in email and spoken that way in person, so my “LLM voice” ends up inheriting the same phatic fluff. “Please” and “thank you” are semantically null, but they’re not functionless. They’re social grease, not information payload.

With LLMs, there’s a secondary effect: style is part of the input. If you consistently talk to a model in a particular register, you nudge its response distribution in that direction. Over time that builds up a stylistic subspace: the model infers “this user likes terse answers,” or “this one likes long, nerdy digressions with footnotes,” or “this one always starts with 'Hey, can you help me'.” I like that my stylistic quirks can feed back into the session, even if it’s just as a faint bias in token probabilities.

I agree with you that some people do blur the line. There’s an entire micro-industry of YouTube channels role-playing “sentient” AI for engagement. If people are already inclined to anthropomorphize, and their feed is full of “ChatGPT had an EMOTIONAL BREAKDOWN” thumbnails, they are definitely going to treat a statistical text generator like a trapped ghost in the machine. All I can say here is that P.T. Barnum was not wrong about the birthrate.

I think your concern about habits leaking from bots to humans is not misplaced -- if we get used to barking orders at a Roomba, when do we start abusing our barista? That seems a plausible scenario to me. But I’d rather err on the side of reflexive courtesy, even with things that obviously can’t feel it. The real risk isn’t that saying “please” to a chat bot will erode our empathy; it’s that we outsource judgment to it and stop checking whether the answers make sense. Politeness is harmless, whether it is to your chat bot or your next door neighbor. Uncritical deference to either, though, is not. Are people gullible about AI? Sure, some are, and that can be (and already is being) exploited for profit. Does saying “please” to Bixby or Siri meaningfully contribute to that? I’m not convinced. For a lot of us, it’s just the same small-talk no-op we already use with humans to lubricate the social gears of conversation, now applied to a very fancy autocomplete that can impressively mimic human conversation.

What the Max Planck paper and these articles show is pretty specific: if you track a bunch of YouTube talks and podcasts over time, you can see a noticeable uptick in a small cluster of GPT-favored words -- delve, comprehend, boast, swift, meticulous, etc -- starting shortly after ChatGPT shows up. The authors call this a closed cultural feedback loop: we train the model on us, the model develops its own lexical quirks, then we start picking those quirks up in our own speech.

That’s interesting, but we should be careful about what it actually means. Swapping “dig into” for “delve” in a TED talk is not the same thing as creating a new spoken dialect. Even though the authors filtered for dialogue, the corpus is still academic talks and STEM-adjacent podcasts -- performative environments where people mimic the written word, read from notes, or stick to a professional register in their vocal delivery. That’s much closer to “spoken writing” than to how people actually talk over coffee or in a bar. Texting isn’t talking; neither is reading your substack post into a microphone and then using it as a vocieover on your vlog.

Language evolves, and LLMs are going to evolve with it -- the printing press standardized spelling. Strunk & White and AP style sheets homogenized prose. PowerPoint presentations in cubicle land gave us “going forward,” “at the end of the day,” and “leverage synergies.” LLMs are going to reflect those changes. But here's the thing that most people seem to miss when it comes to LLM generated text -- you’re going to see convergence on a limited palette of words – humans already use a very sparse active vocabulary. The OED documents over 600,000 words and word-forms in English, but the average native speaker actively uses maybe 5,000–10,000 and knows on the order of 30–40k at most. In a bucket, each of us is operating with a dramatically limited sub-vocabulary of what the language actually makes available. And it is not just English -- you see the same pattern in every language: gigantic dictionaries, tiny personal vocabularies, and a brutal frequency curve where a few thousand common words do almost all the work. Any LLM trained on human text is going to converge hard on that high-frequency core, no matter how clever the prompting is.

This vocabulary shift is mostly harmless; the content shift is a different beast. The AI slop problem is real, but it’s not fundamentally about words like “delve.” It’s about incentives. If karma, clicks, or ad impressions reward volume over thought, of course an LLM is going to become an industrial slop gun: rage-bait scenarios, synthetic drama in AITA subreddits, engagement spam, and low-effort rewrites to farm outrage (looking at you, slashdot trolls). Mods are responding with vibe-based detection because the platform's software filters are looking for spam, astroturf, and bad-faith pattern posting, not whether the post was the result of an LLM prompt.

Where I think people go off the rails is treating any LLM involvement as illegitimate by definition. There’s a huge difference between copy-pasting the first thing the bot spits out, versus using it as a drafting tool. In the latter case, the LLM is closer to a very fast, slightly overeager junior copy editor. The responsibility for clarity, nuance, and honesty still sits squarely on the human side of the keyboard. I think any kind of written communication needs an author and an editor. LLMs are good at the first role – connecting words in plausible ways – and terrible at the second without a human in the loop.

So yes, we can probably measure an LLM accent in online text, especially in semi-scripted speech, which is what the paper's authors focused on. I just don’t buy that as proof of cultural doom. It’s evolution in action: new clichés, new stock phrases, a new batch of verbal tics we’ll eventually mock the way we mock “at this point in time” and "It was a dark and stormy night." The real hazards that LLMs pose are the ones we already know about – spam economics, operant conditioning in the attention economy, and politically motivated disinformation campaigns at scale. Blaming the LLM while ignoring those human-driven incentives is like blaming the hammer for the spec house that never got a punch list walkthrough.

I think you’re right about the symptom and wrong about the diagnosis.

Something really has shifted. I’ve been around slashdot long enough to remember when naked personal attacks getting +5 Insightful was rare enough to be comment-thread drama instead of background noise. The de-cohesion you’re talking about is real in the sense that shared norms feel weaker and more fragile.

Where I disagree is jumping from “norms are under strain” to “no shared values” and “people are just less civil now.”

Every functional community on the net is still built around shared values, even if they’re ugly ones. 4chan has values. Reddit subs have values. Slashdot absolutely has values; we’ve just learned the hard way that “karma + mod points” is an imperfect proxy for them. The fact that a personal attack gets upmodded because it dunks on the right outgroup is itself a value signal: it’s telling you that partisan team loyalty is being rewarded more strongly than old-school /. norms like “attack the argument, not the commenter.”

To me, that’s less “no shared values” and more “the reward structure changed which values dominate in public.”

That’s where the attention economy comes in. For most of human history, public discourse happened in spaces with hard constraints and gatekeepers -- the Roman forum, Renaissance salons, editorial pages in the NYT and Le Mond -- where you had limited bandwidth and a lot of social cost if you crossed certain lines. On the net, we wired up systems that don’t really care about those old boundaries. They care about engagement, and anger is extremely good at generating it.

If you build platforms where short, outraged, partisan jabs get more visibility, more replies, and more dopamine pings than slow, boring civility, people will drift toward outrage. Not because they suddenly stopped believing in civility in the abstract, but because they’re being trained, via a million tiny operant-conditioning loops, that civility is low-reward and rage is high-reward. I’m not using ‘dopamine’ as a metaphor here — platforms are explicitly optimized to tickle the mesolimbic reward pathway, especially the VTA -> nucleus accumbens circuit, which is the same bit of neuroanatomy you see in addiction and reinforcement-learning studies. They are literally hijacking the brain’s reward system for adtech.

So yes: people are behaving worse online than they used to, and the net has absolutely accelerated the de-cohesion you’re describing. But I don’t think that shows an absence of shared values. It shows that the environment we built now systematically rewards values that used to be pushed to the margins, and punishes the ones that held older discourse spaces together.

The stupid, angry impulses were always there. The mistake, in my view, is treating what we’re seeing now as a simple reveal of “how people really are,” rather than a response to the way we’ve re-wired the incentive structure around them.

Yeah, this isn’t the devastating rejoinder you seem to think it is.

Slashdot, and any other online forum for that matter, is not “the social world.” It’s one community, with shared norms, a karma system, and user-driven moderation. The people modding you down and tagging you as the troll you are are your peers, not some shadowy corporate cabal. Mod points are handed out by karma, and meta-moderation is done by anyone who hasn’t cratered their own karma, so the feedback you’re getting is literally the community reflecting you back at yourself. Getting pushback here, or even banned elsewhere, doesn’t mean “society has no place for me.” It means “this particular venue doesn’t want to be your 24/7 argument outlet.” That’s it. I was blissfully unaware of your existence until you showed up here with this ludicrous attempt to dress up your asocial behavior as “freedom of expression.” If I hadn’t already replied, I’d have happily spent a mod point to tag it for the trollish BS that it is. When you claim a forum ban means you’re “exiled from humanity,” you’re not being profound, you’re being a drama queen. Making it less likely that people have to see and read you isn’t oppression; it’s a minor public service to the Slashdot community.

You’re dragging suicidal ideation into a moderation discussion as a debating move. “Let me argue with everyone as I see fit, or my life has no meaning” is not a philosophy of free speech. It's a tactic grade-schoolers use when they get summoned to the principal's office for misbehaving in class. It doesn't work there...why do you think it would work here? Nobody here has the power or the obligation to build you a society where you never feel unwanted. All that’s on the table is whether a specific community has to host your compulsion to “respond to every post I disagree with.” Needless to say, they don't. That's called boundary setting, by the way. And like most grade schoolers, you seem to be uncomfortable with curbs on your asocial behavior. Threatening to kill yourself is exactly what a child would do, and it is exactly why you are about to be plonk-filed, if you keep it up. Welcome to the harsh world of adult life on the net.

This is the flattering story you keep telling yourself: “I’m not being moderated because I’m tedious, I’m being moderated because I’m right and it hurts.” Occam’s razor says otherwise. If your posts were consistently insightful and valuable, you’d be sitting on a pile of +5s and enjoying the same free expression as everyone else. What the community sees is a pattern: one user who has to jump into every argument, everywhere, and won’t let things drop. That gets old fast, no matter how “truthful” you think you’re being. Cognitive dissonance is not what happens when someone hits “mod troll” on yet another of your drive-by counter-takes. It’s what happens when you insist you’re bravely “not cooperating” with the system, while also admitting that a simple ban on a web forum leaves you “suicidally depressed.” Those two self-images don’t line up. You can't simultaneously be a heroic dissident *and* a guy who falls apart if a mod tells him “no.” You do see the contradiction, right?

You keep trying to frame this as a grand struggle over free speech, when it’s much smaller and much simpler. You are not owed an infinite audience. Other people are not props in your “petition for redress” performance. And when a community tells you “enough, go cool off somewhere else,” that isn’t proof that your truths are too powerful for them — it’s proof that your behavior is more annoying than enlightening.

We’re not actually far apart on the premise. Of course the machine only works if the vulnerability exists. Slot machines only work because humans have a reward system; cigarettes only work because we have nicotinic receptors. Here is where we start to diverge, though: The fact that a behavior or susceptibility pre-exists doesn’t mean the industry sitting on top of it is just exposing it. In behavioral psych, “repeatedly exploiting an existing reinforcement pathway to change how often a behavior happens” has a very boring name: training. People are being trained to be stupid. People fed a steady diet of dumbass takes in their social media feed are going to be stupider than they were before they started doomscrolling.

And this is where the divergence in our points of view is no longer sustainable: A geological survey measures; a strip mine extracts. Modern rage-bait systems don’t just “measure” who is prone to dumb, angry engagement. They identify the people most responsive to it, feed them more of what keeps them hooked, and then reward them socially for producing more of the same. That's why I landed on the geological survey analogy; that’s not a passive core sample of human nature, it’s an active extraction-and-refinement process. We started with latent stupidity; we built machinery that concentrates it, monetizes it, and feeds it back into the system. You say people are stupid at the core, and I say they are being trained to be stupid. Whether you call that “abuse” or “training,” the entire rage-bait industry (twitter, instagram, twitch, youtube shorts) it’s doing more than just holding up a mirror -- it's selecting for, elevating, and rewarding a specific limbic pathway. It is why your uncle turned into a raging fox news nutbar and your nephew is flirting with fascism and thinks charlie kirk is some kind of martyr.

“If I like to respond to every post I disagree with”

You kind of answer your own question in the first clause.

What you’re describing isn’t “free expression.” It’s an asocial urge to jump into every disagreement, everywhere, all the time, and “respond as I see fit” with no real limit except your mood. Scale that up to dozens or hundreds of threads and from the moderator’s perspective it doesn’t look like participation, it looks like one guy trying to turn the entire site into his personal argument factory. Your attempt to fig leaf your asocial behavior with free speech arguments is a non-starter. Free speech != compulsory audience. You absolutely have the right to say what you think. But -- and this is why your free-expression argument fails -- the First Amendment doesn’t entitle you to a permanent microphone in an online forum, any more than it entitles you to barge into your neighbor’s living room because you overheard a comment through an open window and you feel compelled to argue with all their dinner guests “as you see fit.” You've articulated your own problem, very clearly, and it's why mods drop the ban hammer on you with what must be annoying regularity. From inside your own skull it feels like “I’m just speaking my mind.” From everyone else’s POV it’s “this idiot again, dragging the same fight into yet another thread.”

Mods don’t see your inner motives. What they see is your asocial behavior pattern: one user generating outsized friction across the site. When the pattern doesn’t change after nudges and warnings, the ban hammer comes out. That’s not them “preventing free expression.” That’s them protecting everyone else’s ability to participate without wading through your permanent counter-take on everything. Consequences aren’t persecution; you frame this as “mods prevent me from responding as I see fit,” as if that’s some outrageous injustice, but it's just mods doing their job, which is to tell you, in no uncertain terms, that “You can’t keep doing this here, in this way.” They aren't telling you you can never speak again, they are letting you know that you can’t keep using their bandwidth and their community as the arena for your asocial compulsion to reply to everything. You chose to treat “I want to respond to every post I disagree with” as a personal right that trumps everyone else’s time, attention, and enjoyment. This kind of asocial behavior is a red flag to any moderator. Mods chose to treat it as what it looks like: someone using “free speech” as a shield for asocial behavior. Their job is not to absorb your compulsions because you’ll threaten suicide if they don’t. Their job is to keep the place livable for everyone.

If your mental health is really that fragile, the answer isn’t for mods to let you keep arguing with every user on their forum. The answer is: step away, get help offline, and maybe treat “I feel driven to reply to everything” as a symptom of a personality defect, not a principle to defend. Rage-baiting is not a socially acceptable outlet for your compulsion. In the larger rage-bait discussion: platforms already act like Skinner boxes. They reward compulsive engagement — especially the kind that locks people into endless, angry back-and-forths. The more you give in to “I must respond to every post that annoys me,” the more you’re letting that machinery train you.

So in a bucket, why do mods ban you? Because from their POV, you aren't some tragic free-speech martyr. You're just a user with asocial traits that you should have left behind when you left kindergarten -- a lack of impulse control and a need for attention -- that is degrading the experience for everyone else. You’re not being punished for having opinions. You’re being punished for acting like a child.

Okay, challenge accepted. If you build machines that relentlessly reward our dumbest, angriest impulses, you’re going to get more of them. That’s not a mirror; that’s a factory. You’re treating rage bait as if the internet were a geological survey: we dig, and whatever stupidity we find was just lying there all along. That’s way too generous to the system. What we actually built is more like a Skinner box. If you pay people in attention and dopamine hits specifically for knee-jerk, bad-faith hot takes, you’re not just “exposing” a latent tendency – you are actively training it. We’ve built feedback loops where shallow outrage gets shown more, thoughtful nuance gets buried, and the humans inside the loop update their behavior accordingly. That’s not revealing stupidity; it’s literally manufacturing it as a side-effect of the revenue model. Platforms like YouTube, TikTok, and Instagram have hijacked reading and replaced it with a variable-ratio dopamine dispenser — basically a slot machine jacked into your limbic system. These platforms make revenue by giving their users an infinite scroll engineered to stimulate the brain’s reward system with novelty, outrage, or cleavage — sometimes all three at once. The content that survives in that environment isn’t what makes us smarter; it’s whatever keeps us twitch-scrolling.

You’re packing three different bodies of law into one spooky word, “theft,” and that muddies the water more than it helps. If it was your intent to muddy the waters, congratulations -- now go troll some other thread.

Think about it this way: If “the ideas were on the open internet,” then you’ve already answered the patent part yourself. Anything publicly disclosed before the filing date is prior art. If an AI regurgitates something that’s already out there, it doesn’t create a new patent right for the AI user – it just means the application is dead on arrival as either not novel or obvious. That’s exactly what the patent system is designed to prevent: locking up what’s already in the public domain.

More to the point: “Theft” is the wrong label for what you are talking about. Patent law doesn’t care whether you “stole” an idea; it cares who first conceived it and whether it was already disclosed. If someone copies an existing invention and tries to patent it, it’s not a clever AI loophole – it’s just an invalid patent application. At worst it’s fraud, and after the PTO kills it, the applicant will certainly get a visit from a process server with a civil tort, and (depending on how politically connected the corporation they are attempting to defrauded is) the DOJ.

Finally, copyright and scraping training data are not Issues that the USPTO deals with. At all. It is a different fight altogether. Training an AI on public data doesn’t magically transfer someone’s patent rights to the model’s owner, and it doesn’t erase prior art. Even in the most brain-dead interpretation of the scraping wars, you don’t get to turn “LLM saw a paper” into “we now own a patent on what that paper taught.” If the model outputs something substantially identical to a copyrighted work, that’s a copyright issue, not a patent issue. Neither of those scenarios gives the AI user a valid patent.

In a bucket, the new USPTO guidance doesn’t bless any of this. It basically says: AI is lab equipment, only humans can be inventors, and prior art is still prior art, no matter whether a human or a GPU found it. If your worry is “big players will try to slip bad patents through on things that already existed,” you’re late to that party – they’ve been doing that with manual searches, interns, and buzzwordy specs for decades. AI doesn’t change the core legal filters; it just changes how fast you can search and how easy it is to generate garbage that those filters are supposed to catch.

You’re giving the hardware way too much credit here. Drop a random into a state-of-the-art drug discovery or materials pipeline and they’re not going to “lock up progress,” they’re going to drown in garbage candidates. The stuff that still matters – and that the law still cares about for inventorship – is choosing the problem, defining the constraints, designing or tuning the models, sanity-checking outputs, and turning one candidate into something that works in the real world. That’s domain expertise and judgment. The new USPTO guidance doesn’t say “push the big red GENERATE button and collect your patent”; it says “we still use the same old ‘did a human actually conceive this invention?’ test, even if AI was involved.”

That dystopia assumes two things that aren’t actually true even after this policy change: (1) that you can get a patent on anything your model spits out, and (2) that nobody has to worry about human inventorship or the usual patentability hurdles. Even under the new guidance, a human still has to be able to stand up in court and say “I had a definite and permanent idea of this claimed invention,” not “I filed whatever the GPU burped out.” If no human ever reaches that benchmark, the patent is defective on old-school inventorship grounds, regardless of how many racks of H100s were involved. On top of that, as AI tools become standard, a lot of AI-assisted output becomes easier to attack as “obvious” to a skilled person using ordinary tools. So raw compute gives you an advantage, sure, but it doesn’t magically let you patent “all of human progress.”

On that much, I’m with you: the constitutional rationale is “to promote the progress of science and useful arts,” not “to funnel perpetual rents to whoever owns the biggest datacenter.” But this particular USPTO tweak isn’t a radical new scheme; it’s the bureaucracy ducking the hard philosophical questions and saying: “AI is lab equipment.” Same inventorship standard, same novelty / obviousness / enablement tests. That absolutely does tend to favor big, opaque, capital-intensive R&D shops – just like every expensive tool does – but it’s an extension of the system’s existing bias toward incumbents, not a brand-new AI loophole.

If your broader point is that IP law has drifted a long way from its original “limited-time bargain” into something closer to “long-term tollbooths for incumbents,” I think a lot of us would nod along. But that’s a critique of the entire modern IP regime – term length, scope, litigation costs, DMCA-style enforcement – not specifically of this AI-guidance change. This move doesn’t turn AI into a magical rent-extraction engine; it just tells applicants, “We’re going to pretend the 19th-century conception doctrine still works in an AI lab.” That’s a fair thing to criticize, but it’s a different argument than “no creativity required, game over.”

I can agree with you on this point, as long as you are not trying to defend piracy as some kind of noble "sharing". But I have to say, judging by the rest of your post, “Sharing of info” is doing a lot of work here. This story is not about someone handing a friend a USB stick with some MP3s. It’s about a guy running a commercial IPTV service, charging 50€ for 3 months or 100€ for 6 months, reselling pay-TV he never paid to license in the first place, and a bunch of hotels/cafés using that to undercut competitors who do pay for legal subscriptions. Calling that “sharing” is like calling a pirate cable company “your generous neighbor.” The tech that makes copying easy doesn’t magically turn commercial theft of service into a noble act of information freedom. A thief is a thief, and there is nothing more ignoble than a thief who commits his crimes in the name of a worthy cause.

Sure, we can and should argue about business models -- subscriptions vs. ads vs. public funding, Creative Commons, whatever. But this particular setup already is a compensation system: broadcasters pay for rights, hotels and bars pay for commercial subscriptions, and the revenue chain keeps the content being produced. The IPTV reseller in this story isn’t experimenting with a better system; he’s just inserting himself into the existing one as an unpaid middleman and pocketing the money. That’s not “rethinking compensation,” it’s “I keep the cash and somebody else eats the loss.”

Again, nobody outlawed “sharing.” I don't think that word means what you think it means --get a dictionary (your choice!) and look up sharing. Then look up "piracy" in the same dictionary. Not even close, right? Greece passed a law that explicitly targets pirate IPTV networks and end users who knowingly buy access to illegal pay-TV streams, with fines between 750€ and 5,000€ (higher for commercial users, doubled for repeat offenders). That’s not “grandma got sued for emailing a recipe,” it’s “hotels and cafés running their business on stolen feeds while their competitors pay full freight.” Calling enforcement there a “waste of lives” is melodramatic. If you decide to run your business on obviously dodgy 100€-for-6-months “all the premium channels you want” boxes, the legal mess is not an accident; it’s baked into the choice.

Yeesh. The source is TorrentFreak, not some mainstream media outlet. TorrentFreak has spent years criticizing heavy-handed anti-piracy campaigns and is hardly a house organ of the copyright maximalists. If even TorrentFreak describes this as an IPTV piracy reselling network and talk plainly about “illegal streams” and “commercial exploitation,” maybe the problem isn’t media bias against “sharing,” it’s that this really is just bog-standard commercial piracy.

Well, look around you. In this thread we literally have: Slashdot summarizing, TorrentFreak reporting (from a pretty pirate-sympathetic angle), and people in the comments arguing about priorities and proportionality. What more do you want? This is exactly what a functioning public debate looks like. You can absolutely criticize copyright law, argue for decriminalizing non-commercial copying, or push for new compensation models. But using “information wants to be free” as a blanket excuse for a commercial IPTV gray-market — especially one used by businesses to gain unfair competition over law-abiding rivals -- just erases the actual victims and pretends there’s no tradeoff at all.

A typical Briton will only see lower energy bills when wholesale prices stay low, grid congestion costs stop wiping out those gains, and OFGEM ensures those savings actually reach the meter. While these record wind outputs are absolutely real and frequently drive wholesale generation costs down to near zero, the price Brits pay is currently dominated by the archaic rules of the UK energy market and the physical cost of moving power from Dogger Bank to London.

The primary culprit is the sad fact that dead dinosaurs are still setting the marginal price. Under the current "pay-as-clear" market structure, the price of electricity is set by the most expensive generator needed to meet demand at that specific moment. Even if wind is providing 55% of the power for pennies, if the grid needs a single gas peaker plant to turn on to meet the last megawatt of demand, every generator gets paid that high gas price. Until market reform decouples renewables from fossil fuels, gas prices will largely dictate a typical Brit's electric bill regardless of how hard the wind blows -- this is the only thing that keeps gas peakers economically viable. As soon as renewables are decoupled from dead dinosaurs in OFGEM's repricing algorithm, fossil fuel generation will stop being a guaranteed profit maker, and start becoming a guaranteed loss. The fossil fuel industry knows this, and will do everything in its power to keep that decoupling from happening.

This is compounded by a massive hidden tax caused by grid congestion. As the Times mentions, Britain has spent nearly £1.3 billion this year paying wind farms to turn off because the cabling network physically cannot carry that much power south. Brits then have to pay gas plants closer to London to fire up to replace them. Realistically, consumers won't feel the full financial benefit of Dogger Bank, Hornsea, and Beatrice until the transmission upgrades catch up to the generation capacity. Brits are likely looking at 2026 for the first bottlenecks to clear, but true structural price drops won't arrive until the late 2020s when new transmission lines come online and the "gas-setting-the-price" mechanic is finally reformed. Wind is doing its job; the grid and the regulators just haven't caught up yet.

Honestly? For the laughs. But also for the insights. An LLM has billions of tokens to string together based on a prompt, compared to typical human adults, who have only a few tens of thousands of tokens to articulate an idea with. And this goes for non-verbal paradigms as well -- music and art are languages with idiomatic patterns that can be captured the same way Gemini or ChapGPT capture syntactic patterns in human writing. Not everybody is a Mozart or Michaelangelo, but machine learning let's everybody explore those landscapes the way an artist or a musician would. Musicians call it noodling, and artists call it doodling. It’s actually pretty fun, and emergent connections surface almost effortlessly. So when someone says “who even wants this?”, the answer is simple: anyone who enjoys exploring what creativity looks like when you’ve suddenly got a collaborator who can riff in any style on demand -- and is not constrained by verbal bandwidth the way humans are. But don't forget the laughs -- it’s pretty fun to hear Elvis belting out Led Zep's Immigrant Song. :) But mainly for curiosity, because the boundaries of “what if?” just got a lot wider. It's for all these reasons musicians plug into effects pedals and midi devices, or authors bounce ideas off writing partners — because sometimes it leads somewhere interesting. If Klay wants to monetize this process, I say go for it.