cyp0633

This is my direct translation from the subtitle of the Chinese article. Apologies if there's any inaccuracy.

No translation yet

It's GPT. Tried and reproduced some polluted single-token Chinese phrases from 4o era.

Z-Image is trained on Ascend though. I believe there'll be a news article from Huawei if so does GLM-5.

It performs well on Mandarin audio transcription, considering it's an European company. It's weird though that it keeps adding spaces between single Chinese characters, and mixing traditional & simplified characters.

At least it's better than sending peasants into the mountains and building solar panels on the flat field that has been growing crops for thousands of years.

US-designed iPhones have at least 2 cameras, some microphones, and biometric sensors. From this point, everyone outside the US should stop using iPhones to prevent surveillance from the American empire.

From another angle, the iPhones are primarily made in China AND India via third-party factories, so no one should ever use iPhones any more.

You have the right to concern about privacy, but that's not how it works.

Human always have thoughts, ideas, and experiences. Writing them down is already a cool thing. If AI has those too, then it should start its own blog :)

In a figure: Model size (B tokens)?

Anthropic itself is a company full of political bias. The metrics simply don't mean anything outside USA.

My Gitea instance also encountered aggressive scraping some days ago, but with highly distributed IP & ASN & geolocation, each of which is well below the rate of a human visitor. I assume Anubis will not stop the massively funded AI companies, so I'm considering poisoning the scrapers with garbage code, only targeting blind scrapers, of course.

The freedom of installing whatever you want indeed brings more opportunity to come across malware, but as long as you lose the freedom, it's up to Google to decide which apps are "safe", which are not. Google will be the only, sole source of apps, they control everything.

It's not about immediate safety, it's about safety in the long run.

That's you, not the common Chinese crowd - compared to a known problem of another good-selling car, I can say that Tesla problem is indeed rare.

A brief search in Chinese social media turns out:

- the cases are rare, once every several months

- the affected battery pack is usually replaced free of charge

- many of the cases are from Chinese in North America, and even one case is about imported Model 3

Considering Chinese media prefer reporting Tesla defects to gain more clicks, I would assume this is not a Shanghai-specific, broadly affecting problem.

I loved typesetting my bachelor thesis with Typst (but with LaTeX math formula), and now it's even more promising after being able to embed PDF figures this July (see issue #145).

Someone just copy-pasted an implementation from a random Chinese blog, completely unaware of what the key means.

Chinese coordinates definitely can be converted to WGS-84 - it's Google that did not do that. Look at Shenzhen River in OpenStreetMap, the streets of Hong Kong and Shenzhen align with each other perfectly.

Really great job, been happy so far. Would be better if the text is searchable via OCR.

With a little search you can find it's a laboratory within the CS department of THU. It's a fairly large lab though, not those led by just one or two professors.

The runner of Compiler Explorer tried to collect the public shortlinks and do the redirection themselves:

Compiler Explorer and the Promise of URLs That Last Forever (May 2025, 357 points, 189 comments)

https://news.ycombinator.com/item?id=44117722

That is not the case here - I never encountered this with whisper-large-v3 or similar ASR models. Part of the reason, I guess, is that those subs are burnt into the movie, which makes them hard to extract. Standalone subs need the corresponding video resource to match the audio and text. So nothing is better than YouTube videos which are already aligned.

The same happens with whisper-large-v3 on Chinese transcription: silence is transcribed to something like "please upvote, share and favourite this video". I suspect they trained the model on some random YouTube video without carefully picking really useful data.

Stellantis brands sales in China have been in decline long before the emergence of EVs. There is indeed some existence of Stellantis in China via Leapmotor (laugh).

I use fprintd and it works well with GNOME + builtin Elan sensor. It indeed needs more complex configuration than Touch ID or Windows Hello though.

Update: this news was denied by BYD

Even privacy unfriendly methods may not solve the problem.

Buying large concert tickets in China has required the buyer's personal ID since several years ago, and the personal ID associated with the ticket cannot be changed after the order is placed. At the entrance of those concerts they do verification based on ticket, ID card and facial recognition. There are still many scalpers who offers tickets even after they're "sold out", though. What they do is to let buyers lie they forgot their ID cards so that there's a higher possibility of entering with the ID card provided by scalpers (which they used to buy tickets). And by the way, the scalper don't have to provide their own ID card - personal information leakage is so common here.

Many phones/chargers with propietary protocols are also equipped with USB PD compatibility. For example, you can pick up some random Xiaomi phone labelled at 90W and expect it to charge at more than 20W with USB PD, or even ~90W on some vivo models.

I don't see why the best charger is the old feature phone's charger - the point in fast charging is that you don't have to wait too long if in a hurry. That charger is the "best" if only you have almost unlimited time to charge, like during the night.

I still can't get MS Office 365 working on Linux over Wine, while no alternatives make me comfortable. Comparing Linux and Win32 ABI on Linux is nonsense without talking about Wine compatibility.

I supposed "modern Xiaomi" would be something like Mi 10 or newer

Existing solutions like Tesseract already can embed text into the image, but I'm wondering if there's a way to combine LLM with Tesseract, so that LLMs can help correcting results and finding unidentified text, and finally still embed text back to the image