yorwba

Yeah, if you want to generate an aesthetically pleasing SVG, you'd be better off asking a pixel-based image-generation model for "vector art" and then deconstructing it into an equivalent SVG with something like LayerPeeler. https://layerpeeler.github.io/

The participants of the Geneva Conventions were well aware that in war, harming innocent civilians and their property is often hard to avoid. So when they delineated what should henceforth count as a war crime, they were careful to weigh collateral damage against the expected military benefit of an attack. So killing civilians isn't automatically a war crime, but pointlessly killing civilians is.

Saying Russia is "deliberately targeting the civilian population" is implicitly a war crime accusation, but it could very well be that the Russian officer who made the targeting decision was actually trying to hit a ball-bearing factory and accuracy just wasn't good enough. Though for FPV drones that can be retargeted in real time, such an argument is harder to make. In any case, whether or not something is a war crime depends on the state of mind of the decision makers, not just the outcome.

It seems like the article title was inspired by this sentence "By contrast, the provider of a VPN or similar services that are used in order to circumvent an ineffective geo-blocking measure and are lawful technical tools which users may legitimately use cannot be regarded as also having communicated the work to the public." I.e. if a VPN is a lawful technical tool, then the VPN provider isn't liable as the publisher of works accessible through it. That leaves open the possibility for VPNs that are not lawful or not mere technical tools to be liable.

For Alibaba, it's definitely all about shopping. That's where their money is. They haven't had much luck investing in entertainment so far.

Giving it numerical data in an LLM-generated prompt doesn't seem to help much: https://imgur.com/a/KFhczOd

It included the table verbatim and even managed to hallucinate a reasonable heading for it, but then the graph doesn't even manage to align the data points with the time axis, leading to an unfortunate collision in the middle.

I guess you should use a traditional graphing library for your presentation slides for now.

GPT Image 1 ended up with a yellow tint without training on another image-generation model's output. It's just that humans like pictures with a soft sunset glow, and this is a very easy global signal for a preference model to pick up on, and for a image-generation model to imitate. So optimizing for aesthetic appeal makes everything slightly tinted by default, unless you make sure to countersteer.

The idea that Chinese models cost less to train seems to be based on that one time DeepSeek estimated the training cost for their V3 model at GPU rental rates as $5 million, and comparing this to other companies' entire R&D budgets. Yet DeepSeek raised $7 billion of fresh money last month, enough to train more than 1000 such models. What gives?

- You need to train lots of experimental models to dial in the training process just right for the one model that actually gets released in the end. Fortunately, these can be smaller.

- However, everyone is training much bigger models now, and doing a lot of RL rollouts on top.

- You can't get the GPUs for this piecemeal at rental rates because they need to be wired together using high-bandwidth interconnects.

- Nvidia GPUs are much more expensive in China, and local alternatives are still immature and not as efficient. Some companies have gotten around this using data centers in Singapore, which should tell you that electricity prices are not the primary consideration.

- The one line item where Chinese companies can probably save quite a bit of money is salaries for rank-and-file researchers.

In any case, they need to make back that money somehow. Giving away freebies isn't going to cut it.

There's a table in the article breaking down the rate by field.

They use a "threshold calibrated so pre-ChatGPT papers flag at 0.4%", so these things do work most of the time. It also means that there are known false positives, so for any given paper, scoring above the threshold isn't irrefutable proof of AI usage. But for things like estimating the overall proportion of AI writing, you only need to be correct on average, so individual false positives don't matter much.

Guayaquil is in Ecuador.

All the coefficients and evaluation points are rational, so it's a counterexample in all fields where 2 ≠ 0 and 3 ≠ 0, doesn't matter whether that field is the complex numbers, real numbers, rational numbers or even a finite field.

Not without first filling the oceans and atmosphere with toxic amounts of oxygen, driving all slimy things that couldn't adapt fast enough into extinction: https://en.wikipedia.org/wiki/Great_Oxidation_Event

On the plus side, since we descend from slimy things that survived the carnage, we have fairly good oxygen tolerance and completely removing all CO₂ might not even hurt all that much.

FWIW someone wrote their PhD thesis on the Iu Mien language: https://opal.latrobe.edu.au/articles/thesis/An_Iu_Mien_gramm... It's not a textbook, but that doesn't mean you can't try to use it as a textbook. For example, there are some thoroughly-analyzed example texts. Also, the acknowledgments section mentions the Iu Mien Literacy Project, which has published a short course on the very basics, as it turns out: https://iumienliteracyprojects.com/resources/ I suspect the 925-page PDF contains some more hints to additional resources like that. :)

I think they're probably more concerned about their Chinese competition, considering that despite all that spending, the Qwen app still trails Bytedance's Doubao in terms of monthly active users: https://www.aicpb.com/ai-rankings/products/china-ai-rankings Though Quark in third place is also made by Alibaba, so put together they're almost caught up with Doubao + Jimeng (place 7, also ByteDance).

If you want a tutorial, http://designwithfontforge.com might be up your alley.

You can read his speech here: https://www.xinhuanet.com/politics/leaders/20260717/72728b6f... He mentioned open source as one way to stimulate innovation and development, that's all. Also pay attention to the part where he says that misuse needs to be prevented. If unsupervised access to LLMs becomes perceived as undermining state control, no more open weights for you.

Well, if you look at Alibaba's financials for FY 2026 https://data.alibabagroup.com/ecms-files/1514443390/5b9061ed... their sales and marketing expenses rose by about 100 billion RMB (10% of revenue), "primarily attributable to the investment in user experiences of Alibaba China E-commerce Group and user acquisition of Qwen app."

So it seems like it's very important to them that people use the Qwen app and they're willing to pay a lot of money for that. Presumably someone thought that keeping their best models closed would drive more business to them (as the sole provider) but then they discovered that closed releases mostly get ignored unless they're really good. (See also: People who think that Chinese AI companies are required to release weights as a matter of policy, because the closed ones hardly ever show up in the news.) Releasing weights for Qwen 3.8 at least lets them get some of that "pretty good for the price of free" media buzz.

Speech-to-text models predict the next token of text from the preceding tokens of text and the current tokens of speech.

Do the parts before you hit enter still get updated if later context indicates you said something else?

That's the error rate for STT, not TTS. TTS is generally easier than STT because you only need to produce one valid pronunciation and don't need to handle variation within and between individuals.

You mean the fall to a thousand questions per month. Now that the volume is low enough someone has a chance of looking at every single one of them, maybe the StackOverflow community can finally collaborate in peace, safe from the onslaught of questions that could be answered by reading the documentation.

OpenAI's own logo description was also written by you-know who: https://openai.com/brand/ The part I find funny is "right angles introduce the precision and structure that technology demands" while the logo is hexagonal and has no right angles at all. Precision and structure not included!

I think the article is mostly just confusing throughput and latency. Releasing on a fixed schedule adds some latency, but that doesn't mean there's a bottleneck constraining throughput. And long-term, development speed is determined by throughput, not whether you shipped a feature a few days earlier.

High latency might become a problem if you need user feedback to decide what to work on next, but you don't necessarily need to push everything to production immediately to get feedback. You can have a separate beta/staging system and ask users to check out your new changes there before they get rolled out to everyone.

China's labor costs are low for most labor-intensive manufactured goods that higher-income countries import from them. They're high for most labor-intensive manufactured goods that China imports from lower-income countries. It's all relative. Not all manufacturing is super polluting (nor is all manufacturing labor-intensive), but if you tried to replicate a non-polluting labor-intensive Chinese factory's cost structure in a high-income country, you'd still run into the problem that the workers you'd need to hire have better options available.

In his speech https://www.xinhuanet.com/politics/leaders/20260717/72728b6f... Xi Jinping mentioned "open source" (开源) exactly once in a list of buzzwords. Deriving from that an "officially stated national strategy" that ensures the free beer won't stop flowing is motivated reasoning. The most popular LLM chatbot in China is Bytedance's Doubao. Do you think this national strategy will involve forcing them to release their weights?

Another buzzword mentioned by Xi is "safety and controllability" (安全可控), which is generally in tension with open source, since that involves giving up control. Historically, when there's a conflict between openness and controllability, the Chinese government tends to choose control, see the Great Firewall.

Two piers were built in the 1940s, side by side

As far as I can tell, only one pier was built (using stainless steel rebar) and the presentation is comparing it to a hypothetical alternative as a theoretical exercise: "What if the Progreso Pier was built using carbon steel rebar?"

The context in which the author wants to know this information is that they enjoy reading fiction, but not low-effort AI fiction, so they built a tool to filter out some of the low-effort AI fiction. Not every application is such a high-stakes affair that anything less than perfect isn't good enough.

When you go, I recommend you look at a map of how the local public transportation network is laid out physically (e.g. https://commons.wikimedia.org/wiki/File:Shanghai_Metro_Linem... ) notice how it gets sparser further from the center, and explore one of the few-kilometer gaps on foot or by bike. Lest you end up thinking the future is evenly distributed. I used to live in the south near where the purple line 5 forks and went on a lot of walks. Half an hour from the university campus and technology park, there were people living in dilapidated houses, heating with coal in winter, a pipe under each house emptying into an open ditch, with children playing in the water in summer. That's China, too.

Indeed you cannot rule out astrology on information-density grounds. Astrology involves quite a lot of information, the problem is that it's mostly unrelated to the outcomes of interest. To get back to the information-density of text, "I love you" doesn't contain a lot of information, but it does contain the one bit you care about, because someone who loves you is more likely to say it than someone who doesn't.

So if you want to determine whether something was written by a human or by AI, to do better than chance it's enough for there to be a difference in the probabilites of a human writing it and AI writing it, respectively. Whether the resulting accuracy is good enough for a particular use case is another matter. 99% is pretty good odds for love and pretty bad odds for "am I going to survive today?" Hopefully there won't be a death penalty for posting AI-generated content.

Whether a text was written by a human or not is just a single bit of information. So you can't rule out its detectability a priori, since even the shortest text contains more information than that.

As long as LLMs are used to write texts humans wouldn't want to write if they could help it (that's why they're getting an LLM to do it, after all), they'll remain detectable. Even if the reasoning might end up equivalent to "This looks like spam; no human in their right mind would write this spam by hand if they could get an LLM to write it, therefore it's most likely written by an LLM."