You posted the code to a public blog page, with no attribution in the code or request of attribution from others, no license, and seemingly intended to share it freely with the world.
Then you got an apology, and a second apology.
I'm confused about what you think you're owed?
The explanation makes perfect sense, the headers were obviously just copied with no malicious intent. What is it that is still bothering you about this?
> no license, and seemingly intended to share it freely with the world
No license means you don’t intend to share it “freely”, since you didn’t share any rights. By default, you don’t own things people shared on the internet just because it’s there.
That being said I’ve even seen people with licenses in their repos who get mad when people used their code, there’s just no telling and it’s best to just treat random sources of code as anathema.
I think GP is referring to the fact that an author’s work is copyright protected by default, and a license is needed to permit others to use freely [1]. StackOverflow posts are licensed under CC BY-SA 4.0 [2].
(Disclaimer: Just commenting on GP’s statement about “no license”, not on the specific disagreement or apology mentioned above which I am unfamiliar with.)
It's worth noting that the code in question was also open sourced and permissively licensed by the original author as he stated in the thread[1]. I guess this isn't really about licensing at all, just the original author seems to think it was rude, and also doesn't want to accept any of the apologies that have been offered.
> with no attribution in the code or request of attribution from others, no license, and seemingly intended to share it freely with the world
The bottom of every page on my blog has a copyright link that you can follow. I dedicated the code to the public domain. I never made a copyright claim. I simply asked Addy to not claim to authorship of the code.
Just wanted to maybe provide a slightly different perspective, but I recently went through this process of pulling back from being socially active and it was for more than just one reason.
I wanted to focus on my health, both mental and physical, this meant going to the gym every morning and making time to read and getting rid of social media.
I also wanted to reduce my consumption of alcohol which typically was fueled by social events and always seemed to throw a wrench in taking care of my health (hard to get to the gym in the morning when you were drinking the night before, and for me it was even after just 1 drink).
What I realized was that many of the people I was spending time with, they oriented their communal time around drinking and for me that's pretty detrimental to my goals. After pulling back from social activity, I've felt so much healthier, happier and optimistic about life.
I get the same exact phone calls as you're describing, and I generally weigh the events I'm being invited to with what the focus of the event is - if the goal of the event is to just get together at a bar, I don't go. I think many of my friends feel that I've lost my way, but it's difficult because I sort of see them in the same light.
What I do hope to do eventually is to cultivate some new friendships, because I am missing that social aspect of my life, but for now I've sort of got a good thing going and I'm not too concerned about rushing it into being.
Does it seem, I'm not sure, ironic maybe? That the main example here is "An app that writes blog posts" - "Researches a topic and writes a blog post about it" - that the company who helped champion the network effects of the internet and surface truly useful search results is now helping to destroy that very same thing they built their entire business on?
Different teams. This AI team at Google cares not for the health of the web. They barely remember that Google has a search engine, except of course for the groudbreaking AI search results which are responsible for many large numbers for use in annual reviews. This team's goal is to sell their AI solution, and if that means demonstrating its ability to generate tools that create crap content that harms the search engine results, well, I'm sure another AI solution can probably combat that later.
Google Search has like 10 years of history of doing its best not to get you to click on a search result, but to answer your question directly or at the very least keep you on their platform while screwing over website owners.
The first two iteration of this were AMP and Instant Answers, the third one is AI Overview. AI Overview should not be seen in isolation, but as a part of the pattern. If it weren't for it, Google would double down on some other method of reaching the same goal.
This one will end up the same way the other two did: there's gonna be a vocal minority that's gonna consider it unfair and a web killer, the vast majority of users won't have an opinion, Google will not care, "the web" will play along, those early adopters are temporarily gonna have an advantage in this "new age" and some will die in the process, but the vast majority is gonna continue on as if nothing happened.
It's also not gonna be the final iteration of this process because shiny new things sound better to investors than marginal improvements, so X years from now AI Overview is gonna be seen as something "old-fashioned", Google Search will pivot once again, and the rest of the web will follow to keep Google happy.
> the vast majority of users won't have an opinion
They're here, they don't care how they get from point A to point B, the tech used to achieve that result is completely irrelevant to them. AI? Great. Not AI such as the Instant Answers era? Also great. Average Joe does not spend his time thinking about the economics of the web.
But you shouldn't confuse them finding "AI" useful now with them being attached to it long term. It's a hip new tool now, but the novelty will fade and Google will have to re-invent themselves all over again. If anything, they kinda screwed themselves over by calling this "AI". AI is supposed to be something within reach, but always some years away. By wasting that term for the current era, it's gonna suck so hard to think of a new marketing term that's gonna be seen as an improvement in comparison to the term "AI".
Which is saddening as the first thing I think when I see this overview is "How do I verify this statement is correct" and paradoxically it sometimes just slows me down.
Yeah hate to say it, because I am an AI hater, but I love the AI results in Google and Kagi. I barely click results anymore for basic questions unless it's something important enough for me to need verification to ensure the AI-gen answer wasn't a hallucination. It's been so nice not having to pick through the cesspool that is StackOverflow to find answers to quick cli questions, or wade through SEO-generated, Amazon-affiliate link garbage for more general questions.
This is what the vast majority of users will do of course. The issue people have with it is that it breaks the "social contract" of the web, which is that part of the advertising income goes to the site that provided the information the answer is based on. That, by destroying that income, "overviews" (now including AI overviews, but that's not where it started) are destroying first publishers and I'm sure it'll go all the way through until Youtube is entirely destroyed as well.
Of course, it does not destroy Google's income ... and it destroys the promise Google made long ago, which is to never keep users on Google platforms.
Oh and to add insult to injury, want to bet Opal will force app developers into what you probably never even imaged would happen on the web? Pay-per-view. Not for a video. For a website/app.
> This AI team at Google cares not for the health of the web. They barely remember that Google has a search engine
Neithet do any teams at Google, including the search teams. Google is an ad company with 80% of its revenue coming from ads. They couldn't give two shits about the health of the web.
That contradicts the long-standing claim that Google does care about as many people using as good a WWW as possible. A ghost town has plenty of ad space, but not much ad revenue.
Arguably the ad business is to blame. It created a perverse incentive. They maximized pay-to-play. The losers were authors that previously published on a passion budget (and would/could never pay for ads). AI is just the last nail in the coffin.
I dont think so. They also triggered SEO race where businesses pump out same bland blogposts to optimize ranking. Content made by humans for those companies was the only viable way at that time, and now new synthetic method emerges - whatever generates revenue will win. AI reels and tiktoks get views, so why bother with human generated content after the training on models have been done? Sad but true.
That's a good point, we had sort of the precursor to this already and yea likely driven by google themselves. It seems that every time incentives are aligned purely for profit we end up with situations like this where they inevitably run a good thing straight into the ground.
Funny, I remember the Internet from over a decade (or two) ago and it was a mess of full-screen ads, seizure-inducing animations, infinite popups, etc. that Google helped eliminate.
Google didn’t eliminate malicious ads - adblockers and easylists did that.
I hate to break it to you bud, but the full-page ads never went away - they just look like content now. You know why you need to scroll for 9 pages to see the ingredients to a recipe?
Google created the pageview driven business model that incentivized the internet to be filled with hostile UX, low-quality lists of paged content, affiliate spam, etc.
We could break their back if we required them to respect trademarks.
Google captured the browser with 92+% "browser -> search" funnel market share.
They turned the URL bar into "search", meaning that the method of finding trademarked companies and products now flows through a competitive bidding marketplace where competitors can slide in front of your hard-earned brand.
> that the company who helped champion the network effects of the internet and surface truly useful search results...
The amount of data on the web crossed the threshold of organic discoverability some time before the AI boom started. AI makes it go from really bad to really, really bad (99% to 99.99%). As far as I am concerned it doesn't change anything.
The same mechanisms to find good content would work today as well - following humans and networks.
If they store both the generated content and the eventual indexed location, they could now filter search results more comprehensively based on content hashes.
It'd be great if you could add it to Kiwix[1] somehow (not sure what the process is for that but 100rabbits figured it out for their site) - I use it all the time now that I have a dumb phone - I have the entirety of wikipedia, wiktionary and 100rabbits all offline.
I use the Mudita Kompakt specifically cause it allows sideloading so I can still have a few extras. Right now I have Kiwix and Libby. It works really well.
I have a $10 a month plan from US cellular with only 2gigs so I try to keep everything offline that I can.
Honestly it's mostly the news... so I draw the line at browser, I'll never install a browser, that's basically something I can do when I sit down at a PC. I read quite a bit and I like to have the ability to look up a word or a historical event or some reference from something I read using Kiwix and it's been great for that, just needed to add a 512gb micro sd card. And Libby I just use at the gym when I'm on the treadmill.
I also want to make sure we can build this in CI. My goal is to have this updated every day using the BigQuery update process, so it becomes a 1–2 day delayed static archive of the current state of Hacker News, which is honestly very cool.
I can probably run the build for free on GitHub Actions runners, as long as the runner has about 40 GB of disk space available. If needed, I can free up space on the runner before the build starts.
I’ll also write to GitHub support and ask if they can sponsor the cost of a larger runner, mainly because I need the extra disk space to run the build reliably.
I was interested to try Date.now() since this is mentioned as being the only part of the Date implementation that is supported but was surprised to find it always returns 0 for your microquickjs version - your quickjs variant appears to return the current unix time.
Good catch. WebAssembly doesn't have access to the current time unless the JavaScript host provides it through injecting a function, so the WASM build would need to be hooked up specially to support that.
You may not have seen this part:
"Tied embeddings: We now tie the embeddings between the encoder and decoder. This significantly reduces the overall parameter count, allowing us to pack more active capabilities into the same memory footprint — crucial for our new compact 270M-270M model."
I have seen this part. In fact I checked the paper itself where they provide more detailed numbers: it's still almost a double of the base Gemma, reuse of embeddings and attention doesn't make that much difference as most weights are in MLP s
At this point - you would think that cheaters could be detected on the server side by either training a model to flag abnormal behavior or do some type of statistics on the movement patterns over time - is a client-side anti-cheat really required?
Many forms of cheating revolve around modding the game locally so that certain textures can be seen through walls, so you always know where opponents are. So you aren't breaking any laws of physics, you are just able to make much better tactical decisions.
The obvious solution would be, just don't send data to the player's client about enemies that are behind walls. But this is a surprisingly hard thing to engineer in realtime games without breaking the player experience (see: https://technology.riotgames.com/news/demolishing-wallhacks-..., and then notice that even in the final video wallhacks are still possible, they're just more delayed).
> So you aren't breaking any laws of physics, you are just able to make much better tactical decisions.
With respect I'd like to disagree on this subtly. A lot of games have the client send their cursor position at relatively frequent updates/packages (i.e. sub-second). So the server knows pretty precisely in which direction and to which object a player is looking.
This in turn can be readily used upon when using wall-hacks, as most players, who use wall-hacks tend to almost faithfully follow objects behind walls with their cursor, which good moderators can usually spot within a few seconds, when reviewing such footage (source: I was involved in recognizing Wall-Hacks in Minecraft, where players would replace textures, to easily find and mine diamonds underground).
The biggest heuristic is that you suddenly get much more consistent. Valorant uses this to ramp up how intrusive its kernel anticheat becomes and often forcing you to turn on more intrusive features to continue playing the game
In Minecraft one of the common ways to catch people using x-ray hacks or transparent texture packs is to run statistics on the blocks mined. If the ratio of stone-to-diamond gets significantly out of whack it's a sure sign someone is cheating.
In blackjack card counting is (probabilistically) caught by tracking player winnings. If someone is beating the odds a bit too much it's a fairly good indicator they are counting cards. Of course in this case getting it wrong isn't so bad from the casinos perspective either since then they'll just kick out a player that was costing them money anyhow.
When the enigma cypher got cracked they had to be very careful about when to act on information gained. If they started beating the odds too much the Germans would cotton on to enigma being broken.
My point being that cheating will almost by definition improve your odds. There are definitely ways to catch that sort of thing happening without installing rootkits. You just might need to hire a couple mathematicians to figure it out.
That final video is recorded to look better than it is too: the delay is based on position, not time. In a real game you'd be moving slower and have the enemy's data on screen for longer.
That's because the 2025 definition of "anti-cheat" leans heavily towards preventing players from enjoying client-side content that's locked behind microtransactions (for example, EA's new Skate game).
What you’ll find is that a subset of players define a behaviour and now you have to prove that that behaviour is cheating. For most behaviours that could be cheating, it will overlap with skilled players.
Examples would be pre-aiming corners and >99th percentile reaction time.
It’s estimated that cod warzone has 45 million players - a 0.1% false positive rate at that player count is 45000 people. That’s a _lot_. It needs to be orders of magnitude less than that.
This is done, and generally doesn't work as well. Your model will catch people using yesterday's cheats, but the cat-and-mouse nature of cheating means that people will adapt. Funnily enough, cheaters are also training models to play games so that they can evade cheat detection. The kernel-level anticheats are designed to prevent the game from running if they detect you are running any software that interacts with the game. Much simpler for the developer, and circumventing it usually requires running your cheats on a second machine which a) limits what you can do and b) has a higher barrier to entry.
I've been reading a lot of Don Delillo lately and so I wanted to see how Grokipedia page on him fares.
I found the "Critiques of elitism" section and noticed this sentence:
"Reviews of Mao II (1991), for instance, highlighted the novel's focus on a performance artist protagonist as emblematic of this tendency, with detractors accusing DeLillo of prioritizing esoteric concerns over relatable human experiences, thereby catering to an academic or literary insider audience."
But Mao II does not have a performance artist as the protagonist, that is the book The Body Artist. Which seems like an obvious failure of the AI model to properly extract the information from the sourced article.
Also strange is that the sourced article (from Metro times) just as a passing comment says: "DeLillo’s choice of a performance artist as his protagonist is one reason why some critics have accused him of elitism." - so it would seem that it is being used as a primary source though it is actually a secondary source (which itself doesn't provide a source)
Overall I'm not too impressed and found some pretty predictable failures almost immediately...
There is no "Talk" page as well, so if your change is 'contentious', how is consensus reached between different view points? Also, how does one link to previous revisions of an article? Further, how to do diffs between different revisions (there is "Edit history")?
See all the heavily biased content on the Joe Biden and Donald Trump articles?
Why not start there? Try to see if you can contribute a more neutral voice.
My guess is that the tone of these articles we see now is only the beginning. This thing is only a month old.
Then this will become the source of truth in the LLMs.
I cannot believe they've gone so against the spirit of contributors and that there's no legal recourse to stop this.
Open source and open content didn't bare enough fangs. It shouldn't have allowed profit.
We build open source for big tech. They use it (Linux, Redis, Elasticsearch, Python, etc.) to profit, keep us from owning the machines and systems (AWS) that concentrate and mechanize that profit. They then corral our labor, lay off, and move jobs overseas.
They expand into every industry and inject massive amounts of money to destabilize the incumbents. We literally just watched them dismantle the entire US film industry in just three years and swallow it whole. Tech is picking the bones dry.
And now it's happening with the media we create too.
If the pendulum of power swings, we MUST dismantle these companies. And we can't be slow like Lina Khan. It must be fast and furious like Project 2025.
We also need to weaponize our open source licenses.
People might - but will AIs? Personally, I think the chance of all of "Grokipedia" being filtered from all AI datasets & search queries is fairly small.
Against the spirit of the contributors? You know Wikipedia's founders have heavily criticized Wikipedia's blatant political biases, right?
Having LLMs edit encyclopedias, moderate forums and do everything else we rely on user moderation for today is the way of the future. They can be much fairer, more predictable, controllable and put you less at the mercy of the army of whackjobs that end up controlling sites like Wikipedia and Reddit. Musk is once again at the forefront of this. The big question is cost. Volunteers work for "free" (often sadly, "free" means in return for the ability to manipulate the public discourse). But GPUs do not, and Grokipedia has no ads.
It is weird to call Wikipedia editors "an army of whackjobs" in the same paragraph as praising Elon for his vision in bringing hallucinated truth to the masses.
The problem is not whether you are impressed or not.
The problem is in the assumption/story/belief that "Intelligence" is "magic" can be "perfected". It's not true. Philosophers have known it forever. And the AI hypestorm will remind everyone how over rated intelligence actually is.
Intelligence can produce Socratic thought. It can also get Socrates killed. It can produce Aristotle and chase Aristotle out of the village. It can produce Einstein and make him depressed. It can produce Galileo and Gandhi and pretend what they say should be deleted.
People are told all the time Brains/Intelligence are special. Its not true. Even if human brains disappear tomorrow sky is not going to fall. Life and the universe will carry on happily on their merry way.
What can be called special is what happens to Information flowing through thousands and thousands of brains over thousands and thousands of years. What Information survives that process can be interesting. But its no where close to the what we see with Photosynthesis or the Krebs Cycle that emerges out of similar process of Information flowing through microbes.
The info through these processes constantly gets misplaced/corrupted/co-opted/deleted etc. Look at the Bible. Lot of people aren't impressed with it either. Yet it has lasted the downfall of nations, empires and kings.
The same applies to both wikipedia and grokpedia or whatever is produced next through "intelligence".
Once you realize your own brain is very imperfect you don't spend so much time worrying about chimp troupe drama generated through those brains. It's called flourishing through detachment - https://oyc.yale.edu/philosophy/phil-181
Then you got an apology, and a second apology.
I'm confused about what you think you're owed?
The explanation makes perfect sense, the headers were obviously just copied with no malicious intent. What is it that is still bothering you about this?
reply