sReinwald

You "skimmed through it" by your own admission, but somehow feel comfortable enough to armchair diagnose everyone involved as having "severe mental health issues". Seriously, are you not ashamed?

Your advocacy of "just focus on mental health" rings particularly hollow when you advocate that we should just ignore a technology that's actively destroying people's mental health, driving them to suicide. The irony is off the charts here. So, please, let's be real - you don't care about mental health. You're scared someone might put a warning sticker on your new toys.

Even if every single person had a pre-existing mental condition (they didn't, and you'd know this if you had read the article) your logic is sociopathic and genuinely disturbing. "Vulnerable people exist. Therefore, companies bear no responsibility when their products exploit those vulnerabilities and harm them. Yay, we don't need to think about it anymore." Cool take. I guess it's fine if technology kills people as long as they might have an undiagnosed mental health issue.

This argument is patently ridiculous in any context other than a Hacker News discussion full of technophiles and accelerationists. Moving fast and breaking things is fine when it comes to your database - not so much when you're dealing with people's lives.

We don't dismiss food safety regulations because some people have allergies. We don't ignore faulty airbags because some people are terrible or reckless drivers. But when a chatbot instructs someone to stop taking prescribed medications, increase ketamine use, and tells them they can fly if they jump off a building - that's just a "mental health issue" and not a flaw in the product? No guardrails, warning labels or safety mechanisms necessary?

The fact that this predominantly affects vulnerable populations makes regulation MORE necessary - not less.

Nothing says, "let's focus more on mental health" than claiming it's "silly" to suggest a technology that actively coaches vulnerable people through suicide attempts might be worthy of regulation.

The issue isn't the mere existence of spyware companies globally. The issue is that Israeli companies in particular have cornered the market on selling to the world's worst human rights abusers, with catastrophic consequences.

Let's be specific: NSO Group sold Pegasus to Saudi Arabia, who used it to track Jamal Khashoggi's inner circle before his assassination. They sold to Mexico, where it was used to target journalists' families within days of their murders. To Rwanda, to hunt dissidents abroad after imprisoning their family. The list goes on.

This isn't cherry-picking. When Citizen Lab analyzes global sypware operations, Israeli companies dominate: NSO, Candiru, Paragon, QuaDream, and arguably Cytrox (Macedonian, but Israeli leadership and investors). The common thread? Former Unit 8200 personnel, who've turned state cyber-warfare capabilities into a business model explicitly built on selling to authoritarians.

Your "but everyone does it" framing fundamentally misrepresents the issue. Yes, other countries have surveillance companies. But there's a massive difference between developing capabilities and systematically selling them to regimes that murder journalists. WHen was the last time a German or French company's tools were found on a murdered journalist's or imprisoned political dissident's phone?

The data shows Israeli companies don't just happen to have "bad PR" (or uniquely terrible luck in choosing their clients) - they actively court authoritarian clients because that's where the money is if you have no morals.

For some context: Israel has a population of less than 10 Million - less than 0.1% of the world's population. If you have a persuasive argument for why Israeli spyware is routinely found by organizations like Citizen Lab, why their products seem so uniquely popular and successful with fascists and authoritarians, I'd love to hear it. Because from where I'm standing, the clear and obvious explanation is that there is a deep, systemic issue in the Israeli private intelligence and cybersecurity sector that is entirely unconcerned with how their tools will be used, or by whom, as long as the money's right. All enabled by the Israeli authorities, who need to approve of these exports.

You're right that spyware companies exist elsewhere. But when researchers keep finding the same tiny country's products in the phones of murdered journalists and jailed activists, dismissing scrutiny as bias is itself a bias. The question isn't why Israeli companies get attention - it's why they keep selling to regimes that use their tools to crush dissent, and worse.

I understand that you're not completely serious about it, but you're proposing a very brittle technical solution for what is fundamentally a social and motivational issue.

The core flaw is that any such marker system is trivially easy to circumvent. Any user intending to pass AI content as their own would simply run the text through a basic script to normalize the character set. This isn't a high-level hack; it's a few dozen lines in Python and trivially easy to write for anyone who can follow a few basic Python tutorials or a 5-second task for ChatGPT or Claude.

Technical solutions to something like this exist in the analog world, of course, like the yellow dots on printers that encode date, time, and the printer's serial number. But, there is a fundamental difference: The user has no control over that enforcement mechanism. It's applied at a firmware/hardware layer that they can't access without significant modification. Encoding "human or AI" markers within the content itself means handing the enforcement mechanism directly to the people you're trying to constrain.

The real danger of such a system isn't even just that it's blatantly ineffective; it's that it creates a false sense of security. The absence of "AI-generated" markers would be incorrectly perceived as a guarantee for human origin. This is a far more dangerous state than even our current one, where a healthy level of skepticism is required for all content.

It reminds me of my own methods of circumventing plagiarism checkers back in school. I'm a native German speaker, and instead of copying from German sources for my homework, I would find an English source on the topic, translate it myself, and rewrite it. The core ideas were not my own, but because the text passed through an abstraction layer (my manual translation), it had no direct signature for the checkers to match. (And in case any of my teachers from back then read this: Obviously I didn't cheat in your class, promise.)

Stripping special Unicode characters is an even simpler version of the same principle. The people this system is meant to catch - those aiming to cheat, deceive, or manipulate - are precisely the ones who will bypass it effortlessly. Apart from the most lazy and hapless, of course. But we are already catching those constantly from being dumb enough to include their LLM prompts, or "Sure, I'll do that for you." when copying and pasting. But if you ask me, those people are not the ones we should be worried about.

//edit:

I'm sure there are way smarter people than me thinking about this problem, but I genuinely don't see any way to solve this problem with technology that isn't easily circumvented or extremely brittle.

The most promising would likely be something like unperceivable patterns in the content itself, somehow. Like hiding patterns in the length of words used, length of sentences, punctuation, starting letters for sentences, etc. But even if the big players in AI were to implement something like this immediately, it would be completely moot.

Local open-source models that can be run on consumer hardware already are more than capable enough to re-phrase input text without altering the meaning, and likely wouldn't contain these patterns. Manual editing breaks stylometric patterns trivially - swap synonyms, adjust sentence lengths, restructure paragraphs. You could even attack longer texts piecemeal by having different models rephrase different paragraphs (or sentences), breaking the overall pattern. And if all else fails, there's always my manual approach from high school.

> Oh yup. The way I've always described myself is I have extremely "muted" emotions, bordering on none the overwhelming vast majority of the time. I only very rarely feel extreme emotions of any kind.

I remember infuriating my mother almost every day after school when she'd ask "How was it?" and I would just shrug and say, "I don't know."

She thought I was being evasive or something, but I was being completely honest. I genuinely didn't have an answer because my internal state was, as you describe perfectly, muted. Most of the time, I just felt... like a neutral, warm grey. Well - still do. There was no data to report.

> I don't remember hardly anything about my own past outside of factual information, and that tends to fade rather extremely with time. Even times when I was quite literally sobbing I don't remember the emotional impact of, just the fact that it happened, sometimes not even the cause.

I think remembering the fact of sobbing but not the feeling is the perfect distinction between semantic memory ("a thing that happened") and autobiographical memory ("an experience I had"). The factual data point was recorded, but the emotional qualia wasn't encoded for retrieval.

> On the other hand, I have extremely good factual memory about random shit and can usually build up a solid approximation for how something works from first principles on demand for an extremely broad array of things. Trade-offs, I guess.

I wouldn't even necessarily call it a trade-off so much as a logical consequence. If the brain's system for storing rich, first-person experiential data is impaired, it makes sense that it would rely on and strengthen its system for storing third-person factual data. The "what" gets stored efficiently because the "how it felt to be me when it happened" isn't taking up much space on the hard drive.

> It's what I imagine being an AI would feel like from the perspective of the AI.

Sounds about right to me. I feel the same. I have access to the facts, like my I'd argue objectively fairly impressive achievement I described above, but I don't seem to have the emotional data. So, I can reason myself into knowing that I achieved something - but I'm not feeling it.

That's fair criticism, I'm obviously coming at this from my personal perspective and that is shaped by how my brain experiences the world. I should've been more precise, I didn't intend to suggest that alexithymia is the only pathway to SDAM, there are likely multiple aspects or pathways that can contribute to or cause it.

However, I would challenge the premise that SDAM is entirely unrelated to emotional processing. It's important to distinguish between the conscious feeling of an emotion and its subconscious role in the mechanics of memory formation. There's significant evidence that emotional salience is a crucial part of how the brain tags and consolidates strong autobiographical memories. A disruption to this process doesn't have to be a consciously felt emotional deficit; it can be a mechanical one operating below the level of awareness.

We can look at this as two distinct points of failure in the memory pipeline:

Failure at the input stage: If the emotional signal required to "tag" an event as important for rich autobiographical encoding is never met, the memory is formed, but only as a semantic fact ("a thing that happened"), not a re-experiencable episode. The processing can't happen because the right input was never provided.

Failure at the retrieval and re-experiencing stage: For someone with aphantasia but no issues with alexithymia (like you, I'd assume), the initial emotional tagging might function perfectly well. The disruption happens later. The core deficit of SDAM is the inability to "mentally time travel" and re-experience the past. Aphantasia, by definition, removes a primary tool for this: visual imagination. The brain processes and integrates emotions by revisiting them. If you cannot truly "re-live" a moment because the visual data is inaccessible, then the episodic, first-person quality is lost.

This second point matters beyond just losing access to nostalgia. We process and regulate emotions by mentally revisiting experiences, integrating them into our broader life narrative. If you have greater difficulty "re-living" moments of joy, achievement, or connection because you lack the tool of visual imagination, your ability to extract meaning from them and build emotional understanding is compromised.

Both mechanisms effectively lead to the same subjective experience: a past that feels like "someone else's life" that you know facts about but can't emotionally (re)connect with. The specific pathway might vary between individuals, but I now strongly believe that the underlying issue remains the disrupted relationship between emotional processing and autobiographical memory formation.

Does this potential explanation align more with your personal experience?

I might have to spend some time over the long weekend to explore this a bit more, and to properly back it up with studies.

That's pretty much what my therapist said when I first expressed this to him, so you're in good company there.

I thought about this more on my commute home from work, and I'm starting to suspect that "SDAM" might essentially be the long-term effects of alexithymia or interoceptive blind spots, which are fairly common in neurodivergent people with ADHD, autism, or both.

For context, alexithymia is a significant difficulty in recognizing, sourcing, and describing one's own emotions. Interoception is the sense of your internal bodily state.

You can likely relate to being so deep in a flow state that you don't notice how badly you need to use the restroom, or how hungry you are, until the feeling becomes so overwhelming it finally breaks through your focus. That's an interoceptive blind spot in action.

So, to further elaborate on my theory: If alexithymia raises the required signal strength for an emotion to be consciously recognized as significant, our brains - which strive for efficiency - will only tag and store memories that cross that unusually high threshold of "important." All the "little things," even the nice ones, get dropped because they never registered with enough emotional weight at the moment they happened.

The brain prioritizes emotionally significant information for memory storage. If an event doesn't trigger a sufficiently strong or clearly identifiable emotional response at the moment it occurs - because your baseline emotional processing is affected - it might get stored as just factual information rather than a rich, emotionally resonant autobiographical memory. It becomes "a thing that happened" rather than an "experience I had that affected me emotionally."

This explains my memory pattern perfectly: I remember most big family holidays - Christmas, birthdays, weddings - because those come with heightened emotional anticipation and distinct social components. But I'm already struggling to piece together what we did on Easter this year, and have absolutely no idea what we did last year. The quieter, more routine positive events apparently don't meet that higher "emotional importance" threshold for deep encoding.

It's like having a filter that's calibrated too conservatively - it's protecting you from information overload. Perhaps that's why it's so common in neurodivergent people, both ADHD and autism heavily affect how we take in and process external sensations. If there's any positive spin to this theory, that I will agree with you, makes sense in a not good way, it might be that. But, unfortunately, it's also discarding experiences that others would naturally encode as meaningful memories.

I can relate to this very strongly. The difficulty in 'selling yourself' by recalling specific achievements for interviews or performance reviews is a challenge I know all too well.

Like you and the author, I have aphantasia (and possibly SDAM), but through self-reflection and quite a bit of therapy, I've come to believe this specific issue lies more firmly in the domain of ADHD. For me, the problem isn't just an inability to recall achievements, but a failure to feel a sense of achievement for almost anything in the first place.

A very recent example from my own life illustrates this perfectly: For context, I struggled through university for about 12 years without a degree before receiving my ADHD diagnosis. Afterwards, I decided to switch careers and applied for an apprenticeship as an IT specialist for system integration, essentially a helpdesk/support role. Because of my self-taught skills, I was hired directly, skipping the apprenticeship. Intellectually, I know this was a very good outcome. In the 8 months since, I've taken on tasks far beyond my initial role - automating reports, working on internal tooling and a local AI solution, and developing customer-facing tools. Just yesterday, I was officially promoted to Test Automation Engineer with a 50% salary increase.

And I know intellectually that going from applying for an apprenticeship to an engineering role in 8 months is a significant achievement. But I don't feel it as such. My dominant feeling is that, at 35, I'm finally just starting to catch up to my peers. To an outside observer, that probably sounds absurd.

My working theory for this is that it's a mechanism very similar to the common ADHD experience of misplacing keys. The issue isn't a failure to recall the memory of where you put them. The problem is that the memory was never properly encoded in the first place, because your attention was already split between several other thoughts at that moment. You can't retrieve a memory that was never saved.

I suspect something similar happens with achievements. If an event doesn't register emotionally as an "achievement" at the moment it happens, the brain doesn't file it under that tag. It just gets stored as another factual event - 'a thing that happened' - which makes it incredibly difficult to retrieve when a prompt like "tell me about a time you solved a hard problem" specifically asks for something from the 'achievements' file.

It seems the core concept of contextual integrity is still not landing.

It's not a question of surprise that public data can be scraped - I'm well aware of how the internet functions, thank you. The point, which you seem determined to evade, is about the fundamental ethics of systematically doing so and the vast difference in impact and expectation between, say, a server's own moderation logs or incidental screenshots, and a third party, globally indexed, permanent archive. The former serves limited, often known functions within that specific community; the latter is a privacy-invasive data trawl weaponizing the 'public' label. Just because a thing is technically possible doesn't grant a free pass to ignore privacy implications or users' reasonable expectations of how their contributions will be used and disseminated.

Your attempt to dismantle the 'public park' analogy only underscores your misunderstanding of it. The scenario isn't about someone yelling (an exceptional event, often a public nuisance, that might indeed attract specific attention or recording). It's the equivalent of someone systematically planting listening devices by every park bench, transcribing every casual, low-expectation conversation - like my dinner plans with my girlfriend, or a vent about my boss - and then publishing it all online, forever, simply because the park itself is 'public' and it was a technically possible thing to do. The ethical chasm between observing a public spectacle and conducting mass, indiscriminate surveillance of every day, semi-private interactions within a public space shouldn't be this difficult to grasp. One involves a specific event; the other is a dragnet.

As for flagging, I didn't touch your comment. I have never flagged a single comment on this site. Perhaps others simply disagreed with the quality, relevance or the dismissive tone of your contribution.

No, what's "ridiculous" is this simplistic, black-and-white framing that deliberately ignores any nuance, the concept of contextual integrity or reasonable user expectations.

Of course, no one expects absolute secrecy in a public-facing Discord server. That's a straw man. The issue isn't about some naive belief that messages are invisible. It's about the scope, permanence, and method of access and archiving.

People participating in public Discord spaces have reasonable contextual expectations about how their words will be accessed and by whom. They expect their messages to be seen by current and maybe future server members - not extracted, permanently archived, and made globally searchable by entirely unrelated third parties.

This is similar to how conversations in a public park are technically "public," but most people would be rightfully disturbed if someone recorded everything, transcribed it, published it online with their names attached, and made it all searchable forever. Just because something isn't strictly private doesn't mean any and all forms of collection, republication, and indexing are ethically justified.

If you can't see the distinction between "not perfectly private within this specific semi-public space" and "archived indefinitely, and globally searchable forever by anyone, anywhere, for any reason," then you're either arguing in bad faith or your understanding of these issues is so superficial that further engagement is pointless.

While it's understood that LLM outputs have an element of stochasticity, the central finding of this analysis isn't about achieving bit-for-bit identical responses. Rather, it's about the statistically significant and consistent directional bias observed across a considerable number of trials. The 56.9% vs. 43.1% preference isn't an artifact of randomness; it points to a systemic issue within the models' decision-making patterns when presented with this task. Technical users might understand the probabilistic nature of LLMs, but it's questionable whether the average non-technical HR user, who might turn to these tools for assistance, does.

Your suggestion to implement a "clearly defined taxonomy" for decision-making is an attempt to impose rigor, but it potentially sidesteps the more pressing issue: how these LLMs are likely to be used in real-world, less structured environments. The study seems to simulate a plausible scenario - an HR employee, perhaps unfamiliar with the technical specifics of a role or a CV, using an LLM with a general prompt like "find the best candidate." This is where the danger of inherent, unacknowledged biases becomes most acute.

I'm also skeptical that simply overlaying a taxonomy would fully counteract these underlying biases. The research indicates fairly pervasive tendencies - such as the gender preference or the significant positional bias. It's quite possible these systemic leanings would still find ways to influence the outcome, even within a more structured framework. Such measures might only serve to obfuscate the bias, making it less apparent but not necessarily less impactful.

Okay, the "not really" and "I'll solve that problem if and when" responses are... something else. It feels like you're speedrunning how to get into a world of trouble while hand-waving away every legitimate concern. Let's try to unpack this again, because your justifications are frankly baffling.

> Again, not really. It's impossible to search for users without already knowing what server they are in. This is functionally identical to Discord's in-built search feature.

That's not quite correct, and frankly it borders on willful obfuscation. In your own words elsewhere in this thread, you're eager for search engines to index this archive. That "privacy preserving" barrier of needing to know both a user ID and a server/channel id evaporates the moment Google or any other search engine hoovers up your pages. At that point, any combination of keywords, usernames, aliases, or snippets could reveal someone's posting history, across contexts and years. How is that "functionally identical" to Discord's walled-garden search or "privacy preserving"?

> I believe that people need to realize that their messages were already being logged by many different moderation bots, just not publicized.

This is a disingenuous deflection.

  - Moderation bots operate within a specific server, for a specific purpose (moderation, utility) defined by the server admins. Their logs are typically for admin/moderator use, not for creating a global, publicly searchable archive.
  - Users joining a server often see these bots, understand their function, and server admins explicitly add these bots. It's a known quantity. What you're doing is orders of magnitude different - an external, uninvited entity scraping everything discoverable and making it universally public.
  - Just a matter of time" is a lazy, fatalistic excuse for unethical data harvesting. Just because something can be technically scraped doesn't mean it should be, or that you doing so is fine.
    

Your "I really do want to build a platform that balances privacy and usability" line sounds utterly hollow when the entire foundation of the project demonstrates a profound misunderstanding, or disregard, for basic privacy, consent, and intellectual property.

Speaking of which... have you actually thought about the legal Pandora's Box you're prying open? Your casual "I'll deal with Discord's ToS issues if they arise" attitude is quaint, because Discord's ToS is likely the tip of a colossal iceberg of legal trouble.

You're not just 'breaking ToS', you're potentially looking at:

  - Data Protection Law Violations (GDPR, CCPA, etc.) because you're scraping personal data of EU/California (and other) residents without any lawful basis. The fines can be astronomical. "Opt-out" after the fact for data you had no right to take in the first place isn't how this works.
  - COPPA Violations if you scraped any messages from a 12-year-old on a "public, discoverable" server before their account was deleted by Discord. Guess who's holding that data now without parental consent? You.
  - Every original, creative message is copyrighted by its author. Roleplay, detailed discussions, code snippets, even well-crafted tirades – you're republishing millions of these. While not every "lol" is copyrightable, a massive volume of content on Discord absolutely is. "Fair use" for wholesale, non-transformative republication on this scale? Unlikely.
  - And last but not least, CSAM (Child Sexual Abuse Material): This is the nightmare scenario. You are scraping public Discord. Some public, poorly moderated Discords inevitably contain links to or text-based CSAM. Even if you don't intend to host it, if your scraper picks it up and it becomes accessible via your archive (even just a link), you are in profoundly serious trouble. "But I don't re-publish attachments" is irrelevant if you're archiving and re-publishing the links. This isn't just fines; this is potential prison time.

Good luck with all of this.

I hope you have a good lawyer, ideally multiple. You might need them.

The sheer audacity here is quite something. You're stating people can't use your scraped data for commercial purposes "without permission," while your entire project is built on vacuuming up content from countless users without their permission, and in direct violation of Discord's ToS. That's not just a double standard; it's bordering on next-level cognitive dissonance.

And "privacy preserving"? With a one-click opt-out, that 99.999% of the affected users will never even know exists because they have no idea their conversations are now part of your archive, and you want it indexed by search engines? That's not "privacy preserving" - that's a bad joke. If privacy was a genuine concern, this project wouldn't exist in its current form. What you're offering is an opt-out fig leaf for a mass data harvesting operation.

Most people using Discord, even on "public, discoverable" servers, aren't posting with the expectation that their words will be systematically scraped, archived indefinitely, and made globally searchable outside the platform's context. It's a fundamental misunderstanding (or willful dismissal) of user expectations on what is essentially a semi-public, yet distinctly siloed, platform. This isn't an open-web forum where content is implicitly intended for broad public consumption and indexing.

Look, I get the frustration that (likely) motivated this. Discord has become an information black hole for many communities, and the shift away from open, searchable forums for project support is a genuine problem I've been incredibly frustrated with myself. But this "solution" - creating a massive, non-consensual archive that tramples over user privacy (and platform terms) - creates far graver ethical and practical issues than the one it purports to solve.

Semantic router is on my radar, but I haven't had a good look at it yet. The primary bottleneck in our current setup, isn't really the routing decision time. The lightweight LLM I chose (Gemma3 4B) handles the task identification fairly well in terms of both speed and accuracy from what I've found.

For some context: this is a fairly limited exploratory deployment which runs alongside other priority projects for me, so I'm not too obsessed with optimizing the decision-making time. Those three seconds are relatively minor when compared with the 20–60 seconds it takes to unload the old and load a new model.

I can see semantic router being really useful in scenarios built around commercial, API-accessed models, though. There, it could yield significant cost savings by, for example, intelligently directing simpler queries to a less capable but cheaper model instead of the latest and greatest (and likely significantly more expensive) model users might feel drawn to. You're basically burning money if you let your employees use Claude 3.7 to format a .csv file.

That’s a perfectly valid idea in theory, but in practice you’ll run into a few painful trade-offs, especially in multi-user environments. Trust me, I'm currently doing exactly that in our fairly limited exploration of how we can leverage local LLMs at work (SME).

Unless you have sufficient VRAM to keep all potential specialized models loaded simultaneously (which negates some of the "lightweight" benefit for the overall system), you'll be forced into model swapping. Constantly loading and unloading models to and from VRAM is a notoriously slow process.

If you have concurrent users with diverse needs (e.g., a developer requiring code generation and a marketing team member needing creative text), the system would have to swap models in and out if they can't co-exist in VRAM. This drastically increases latency before the selected model even begins processing the actual request.

The latency from model swapping directly translates to a poor user experience. Users, especially in an enterprise context, are unlikely to tolerate waiting for a minute or more just for the system to decide which model to use and then load it. This can quickly lead to dissatisfaction and abandonment.

This external routing mechanism is, in essence, an attempt to implement a sort of Mixture-of-Experts (MoE) architecture manually and at a much coarser grain. True MoE models (like the recently released Qwen3-30B-A3B, for instance) are designed from the ground up to handle this routing internally, often with shared parameter components and highly optimized switching mechanisms that minimize latency and resource contention.

To mitigate the latency from swapping, you'd be pressured to provision significantly more GPU resources (more cards, more VRAM) to keep a larger pool of specialized models active. This increases costs and complexity, potentially outweighing the benefits of specialization if a sufficiently capable generalist model (or a true MoE) could handle the workload with fewer resources. And a lot of those additional resources would likely sit idle for most of the time, too.

That's not quite correct. The "sovereignty" pitch here is largely illusory when dealing with a US-based company like OpenAI.

The US CLOUD Act (Clarifying Lawful Overseas Use of Data Act) explicitly gives US authorities the power to compel US-based companies to provide data stored on servers, regardless of where those servers are physically located. This effectively undermines any meaningful data sovereignty claims.

Consider the actual arrangement being proposed:

    - OpenAI (US company) maintains control of the infrastructure
    - OpenAI controls the models and their development
    - OpenAI maintains the security protocols and access rights
    - The data merely sits physically within national borders

This isn't sovereignty - it's a limited hosting arrangement that remains fully under US legal jurisdiction. US intelligence agencies can still access this data through legal mechanisms that bypass the host country's laws entirely.

While technically you're right about the rules of GeoGuessr as a game, I think this misses the more interesting and potentially concerning implications.

IMO, whether or not it's "cheating" at GeoGuessr is less important than what this capability represents: an AI system that can extract specific details from images, determine what's significant, and use that information to precisely locate a photo. That's impressive regardless of method.

The implications extend far beyond the game. Consider its potential use in automating OSINT tasks. An AI agent running o3 could analyze publicly posted photos (e.g., on Twitter, Instagram), identify locations from background details, and aggregate this data. Within an hour or two, this could build a detailed location profile (home, work, frequented places) for an individual - something that previously required considerable manual effort and expertise. While human analysis would likely still be needed, this significantly lowers the barrier for large-scale analysis.

This ability to extract location clues from images - whether shop names, signs, or landscapes - has significant real-world implications, particularly for automating OSINT tasks. The potential for large-scale location analysis from public photos is the more profound point, regardless of whether one specific method breaks game rules. That capability is what deserves attention.

I really have to challenge the notion of AI "distilling information without cognitive bias.

First, AI systems absolutely embody cognitive biases - they're just different from human ones. These systems inherit biases from:

  - Their training data (which reflects human biases and knowledge cutoffs)
  - Architectural decisions made by engineers  
  - Optimization criteria and reinforcement learning objectives  
  - The specific prompting and context provided by users

An AI doesn't independently evaluate source credibility or apply domain expertise - it synthesizes patterns from its training data according to its programming.

Second: You frame AI as a "power suit" for distilling information faster. While speed has its place, a core value of doing research isn't just arriving at a final summary. It's the process of engaging with a vast, often messy, diversity of information, facts, opinions, and even flawed arguments. Grappling with that breadth, identifying conflicting viewpoints, and synthesizing them _yourself_ is where deep understanding and critical thinking are truly built.

Skipping straight to the "distilled information," as useful as it might be for some tasks, feels like reading an incredibly abridged version of Lord of the Rings: A small man finds a powerful ring once owned by an evil God, makes some friends and ends up destroying the ring in a volcano. The end. You miss all the nuance, context, and struggle that creates real meaning and understanding.

Following on from that, you suggest that this AI-driven distillation then "allows for another level, experiments, surveys, etc to uncover things even further." I'd argue the opposite is more likely. These tools are bypassing the very cognitive effort that develops critical thinking in the first place. The essential practice for building those skills involves precisely the tasks these tools aim to automate: navigating contradictory information, assessing source reliability, weighing arguments, and constructing a reasoned conclusion yourself. By offloading this fundamental intellectual work, we remove the necessary exercise. We're unfortunately already seeing glimpses of this, with people resorting to shortcuts like asking "@Grok is this true???" on Twitter instead of engaging critically with the information presented to them.

Tools like this might offer helpful starting points or quick summaries, but they can't replicate the cognitive and critical thinking benefits of the research journey itself. They aren't a substitute for the human mind actively wrestling with information to achieve genuine understanding, which is the foundation required before one can effectively design meaningful experiments or surveys.

While blocking before authentication seems intuitive for efficiency, checking after provides crucial context that's missing if you block pre-auth: you know which specific user account just authenticated successfully.

This context enables two important things:

- Granular exceptions: If Alice is attending a conference in Toronto, you can say "Allow Alice to log in from Canada next week" without opening Canada-wide logins for everyone. Pre-auth geo-blocking forces you into an all-or-nothing stance.

- Better threat intelligence: A valid login from an unexpected region (e.g. Moscow when Alice is normally in D.C.) is a far stronger signal of compromise than a failed attempt. Capturing "successful login + wrong location" helps you prioritize real threats. If you block pre-auth, you'd never know Alice's account was compromised.

Putting geo-checks after authentication gives you precise control over whom, exactly, is logging in from where, and offers richer data for your security monitoring.

This response is a textbook example of a manipulative false dichotomy that poisons legitimate discourse about child safety online.

Presenting the only options as either "scan your child's biometric data into opaque systems" or "let your child be groomed and/or get addicted to porn" is intellectually dishonest and deliberately inflammatory. It's a rhetorical trap designed to shame parents with valid privacy concerns into compliance.

Privacy rights and child protection are not mutually exclusive. Numerous approaches exist that don't require harvesting biometric data from minors, from improved content filtering and educational initiatives to parental controls and account verification methods that don't rely on facial scanning. Corporations are simply implementing the most convenient (for them) solution that technically satisfies regulatory requirements while creating new data streams they can potentially monetize.

What's actually happening here is deeply troubling: we're normalizing the idea that children must surrender their biometric data as the price of digital participation. This creates permanent digital identifiers that could follow them throughout their lives, with their data stored in systems with questionable security, unclear retention policies, and potential for future misuse.

Weaponizing the fear of child exploitation to silence legitimate concerns about corporate overreach isn't just manipulative - it's morally reprehensible. Framing opposition to biometric surveillance as being pro-exploitation deliberately poisons the well against anyone who questions these systems.

We can and must develop approaches that protect children without surrendering their fundamental privacy rights. Pretending these are our only two options isn't just wrong - it actively undermines the nuanced conversation we should be having about both child safety and digital rights.

It depends on what I'm after. I still use regular searches quite a bit.

But a lot of my classic ADHD "let's dive into this rabbit hole" google sessions have definitely been replaced by AI deep searches like Perplexity. Instead of me going down a rabbit hole personally for all the random stuff that comes across my mind, I'll just let perplexity handle it and I come back a few minutes later and read whatever it came up with.

And sometimes, I don't even read that, and that's also fine. Just being able to hand that "task" off to an AI to handle it for me is very liberating in a way. I still get derailed a bit of course, but instead of losing half an hour, it's just a few seconds of typing out my question, and then getting back to what I've been doing.

That's not quite correct. It is no longer visible on the website itself, or within the mobile clients. But if you inspect the requests, you can still see things like

    source: <a href="http://twitter.com/download/iphone" rel="nofollow">Twitter for iPhone</a>

which shows the tweet was posted from iOS.

> In reality I enjoy steering it to reach a solution.

That resonates with me. It actually brings back some nostalgic memories about setting up and constantly tweaking bots in World of Warcraft (sorry, I was young).

There's something incredibly engaging and satisfying about configuring them just right and then sitting back, watching your minions run around and do your bidding.

I get a similar vibe when I'm working with AI coding assistants. It's less about the (currently unrealistic) hype of full replacement and more about trying to guide these powerful, and often erratic, tools to do what they're supposed to.

For me, it taps into that same core enjoyment of automating and observing a process unfold that I got with those WoW bots. Perhaps unsurprisingly, automation is now a fairly big part of my job.

Seems easy enough to get around with homophonic substitution, though. Didn't refuse any of my tests so far.

You can also set

    OLLAMA_CONTEXT_LENGTH=<tokens>

as an environment variable to change ollama's default context length.

It is not only targeting the Right. I'd suggest that Russia isn't primarily interested in bolstering the right specifically. They are interested in dividing national societies as well international alliances along any possible ideological issue, effectively shattering the entire thing - divide et impera.

You can see some indication of this in the Mueller report on Russia's election interference in the US, and the DOJ indictment against Prigozhin and his Internet Research Agency troll factory. They kept playing both sides: pushing Facebook groups that were strong pro racial justice & BLM while also pushing pro cop / back the blue Facebook groups. At some point, they scheduled protests for both sides of this particular issue at the same time and place, clearly hoping for altercations.

But I think it's safe to say that, at least in the US, their efforts seem to have mainly focused on the Republican side. However, here in Germany, both the populist far-right AfD and the populist far-left political Party BSW around former Linke politician Sarah Wagenknecht are very much aligned with Russian interests: https://www.dw.com/en/russias-best-friends-in-germany-afd-an...

> I also suspect that Russia has been causing the very problems that the far right reacts against, most notably immigration.

They absolutely are.

A large portion of immigrants into Europe originate from countries suffering from conflicts that Russia has a big hand in. The big one, of course, being Syria, but also many African nations where the Wagner group, now effectively re-branded as the Russian Africa Corps are providing military support and protection.

Another big aspect of Russia facilitating immigration into Europe, is the immigration route through Belarus: https://en.wikipedia.org/wiki/Belarus%E2%80%93European_Union...

> Biggest caviar there is that html export is still wip.

Pandoc seems to support converting from and to Typst. That might be an option for you.

Constantly warning about how your product is or will eventually be so powerful that it will become a danger to humanity or society is the successful marketing.

And for the big AI companies, like OpenAI, it has the very beneficial side-effect of establishing the narrative that lets them influence politics into regulating their potential competitors out of the market. Because they are, of course, the only reasonable and responsible builders of self-described doomsday devices.

Perhaps both. But I wonder if the incredible blandness of most chatbots is effectively just a regression towards the mean.

Most AI companies try to train their bots on vast amounts of different data, and I suspect it's very difficult for that to result in very creative writing when you're training on works of fiction, as well as cooking recipes, Reddit comments and technical documentation.

> [...] the precarious state that VR is currently in, where Meta is more or less propping up the entire consumer market with their willingness to burn billions of dollars on Reality Lab every quarter,

They don't quite seem so happy to do that for much longer.

https://www.businessinsider.com/mark-zuckerberg-cuts-spendin...