Canada bans Huawei equipment from 5G networks, orders removal by 2024

Canada has banned the use of Huawei and fellow Chinese tech giant ZTE’s equipment in its 5G networks, its government has announced. In a statement, it cited national security concerns for the move, saying that the suppliers could be forced to comply with “extrajudicial directions from foreign governments” in ways that could “conflict with Canadian laws or would be detrimental to Canadian interests.”

Telcos will be prevented from procuring new 4G or 5G equipment from the companies by September this year, and must remove all ZTE- and Huawei-branded 5G equipment from their networks by June 28th, 2024. Equipment must also be removed from 4G networks by the end of 2027. “The Government is committed to maximizing the social and economic benefits of 5G and access to telecommunications services writ large, but not at the expense of security,” the Canadian government wrote in its statement.

The move makes Canada the latest member of the Five Eyes intelligence alliance to have placed restrictions on the use of Huawei and ZTE equipment in their communication networks. US telcos are spending billions removing and replacing the equipment in their networks, while the UK banned the use of Huawei’s equipment in 2020, and ordered its removal by 2027. Australia and New Zealand have also restricted the use of their equipment on national security grounds.

At the core of these concerns is China’s National Intelligence Law, which critics claim can be used to make Chinese organizations and citizens cooperate with state intelligence work, CBC News reports. The fear is this could be used to force Chinese tech companies like Huawei and ZTE to hand over sensitive information from foreign networks to the Chinese government.

Huawei disputes the claim and says its based on a “misreading” of China’s law. “China will comprehensively and seriously evaluate this incident and take all necessary measures to safeguard the legitimate rights and interests of Chinese companies,” China’s Canadian embassy said in a statement in response to Canada’s ban. In a statement emailed to The Verge, Alykhan Avelshi, a vice president at Huawei Canada called the policy “an unfortunate political decision that has nothing to do with cyber security or any of the technologies in question.”

Canada has taken around three years to come to its decision about the use of Huawei and ZTE equipment in its telecoms networks, a period which Bloomberg notes has coincided with worsening relations between it and China. In December 2018, Canada arrested Huawei’s Chief Financial Officer Meng Wanzhou on suspicion of violating US sanctions. Days later, China imprisoned two Canadian nationals: former diplomat Michael Spavor and entrepreneur Michael Kovrig. After the US came to a deferred-prosecution deal with Meng that allowed her to return to China last year, the Canadians were released.

Opposition politicians criticized the Canadian government’s delay. “In the years of delay, Canadian telecommunications companies purchased hundreds of millions of dollars of Huawei equipment which will now need to be removed from their networks at enormous expense,” Conservative MP Raquel Dancho said in a statement reported by the Toronto Sun. But Bloomberg reports that the likes of BCE and Telus have already been winding down their use of Huawei’s equipment over fears of an eventual ban.

Update May 20th, 6:04PM ET: Added statement from Huawei.

Repost: Original Source and Author Link


Huawei App Store Bug Gives Anyone A Free Pass At Paid Apps

Android app developer Dylan Roussel discovered a bug that, while non-trivial to exploit, isn’t impossible either. In a nutshell, Huawei’s AppGallery exposed certain details about an app, including the download link for the Android package (APK). While that may be normal, the bug is that the same link can be used to directly download a paid app without having to pay for it or even having to verify anything.

This bug has two damaging consequences for Huawei’s app marketplace. The first is more obvious in that anyone with a bit of technical know-how can easily bypass restrictions and download paid apps for free. The bigger threat, however, is that the AppGallery makes it too easy to download apps, both paid and free, outside of official channels, which in turn makes it too easy to pirate apps on that platform. This creates a very large deterrent for developers who may not bother putting in the work needed to offer their apps for Huawei’s ecosystem.

This vulnerability was discovered and reported back in February 2022, but it took Huawei 90 days to send a response. The company did apologize for the miscommunication and delay, citing logistics problems in fixing AppGallery across different regions since it apparently works very differently, too. A fix is promised to arrive by May 25, but the bug’s existence still raises concerns about similar issues that may be lurking in the shadows still undiscovered.

Repost: Original Source and Author Link

Tech News

Nokia X60 rumored to run Huawei HarmonyOS instead of Android

HMD Global has made great strides in reviving the Nokia brand in the Android world. Whether it’s successful or not is still debatable but, so far, it has definitely played the part of a good Android maker, especially when it comes to its commitment to Android updates. Now there are whispers that its honeymoon phase with Android might be over and is courting another partner, one that would see an upcoming Nokia X60 and X60 Pro run Huawei’s shiny new HarmonyOS instead.

It’s really not that easy to qualify what HarmonyOS really is beyond being Huawei’s multi-platform OS. The company might vehemently deny the association and define what HarmonyOS is in buzzword-filled marketing language, but, at least based on some developers’ analyses, its phone version still has many hooks into Android. Calling it an Android fork probably comes the closest, which means it is both Android yet also not Android.

That is what makes this new rumor rather surprising if true. According to IThome, HMD Global has decided to put this very same HarmonyOS on one of its upcoming phones, called the Nokia X60 and Nokia X60 Pro. HMD Global does have experience with other mobile platforms, like the Symbian-based S60 on its Nokia feature phones, but this rumor comes rather out of the blue even for the company.

One potential reason for this sudden change is that HMD may want to target Chinese markets specifically with these two phones. The fact that HarmonyOS has no access to Google Play services and apps is not a big deal in China since those have not been available there anyway. In global markets, it’s a deal-breaker, as Huawei can attest to.

As for the Nokia X60 itself, there is still some uncertainty regarding its specs but features like a 200MP camera, 6,000 mAh battery, and curved edge screens have been mentioned. Those are features that definitely sound almost too good to be true, which only makes this rumored HarmonyOS Nokia phone sound even more fantastical.

Repost: Original Source and Author Link

Tech News

Huawei shipped only 14.19 million smartphones last quarter

Recent market figures have shown Huawei’s fall from grace in the smartphone market, falling off most market analysts’ Top 5 vendors. Just how much it fell isn’t always evident but a new infographic reveals the troubles that Huawei experienced in the first quarter of this year. According to Counterpoint Research, Huawei only made up 4% of the global smartphone shipments of the year’s first three months, a massive loss considering it was once the world’s number one smartphone brand even for a brief moment.

The numbers come in the context of the mobile market’s Q1 2021 performance which is good or bad news, depending on what you compare it with. The smartphone market grew by as much as 20% in the first quarter compared to the same period in 2020 but, compared to the last quarter of 2020. All in all, the industry shipped 354.94 million phones.

If Counterpoint Research’s data is correct, Huawei only sold 4% of that total last quarter, falling way behind Vivo in fifth place with 10% of the global phone market’s pie. To be fair to Huawei, it was also the first quarter after it sold off its Honor business which would partially explain the steep decline in sales. That, however, still doesn’t bode well for the company.

That makes its recent HarmonyOS 2.0 launch all the more critical to the survival of its smartphone business. The homegrown operating system reportedly has more than 10 million users already, hinting at both the interest in the software and the great number of Huawei phones out there. Whether it is able to maintain and grow these numbers will depend on how well is OS performs on its own.

As for the other major brands, Apple saw some decline in its share globally and in some regions. In Europe, that was enough to cede the top position to Samsung, though Apple retains its lead in North America.

Repost: Original Source and Author Link

Tech News

Huawei Watch 3 vs Samsung Galaxy Watch 3: The better option

The smartwatch industry is gaining pace with more OEMs joining the party. After Samsung, which has been a fiercer competitor for the Apple Watch, the likes of OnePlus and now Huawei is trying to make inroads. To present a viable option, Chinese tech giant Huawei has announced its first smartwatch running the in-house HarmonyOS. Dubbed Huawei Watch 3, it has launched alongside a Pro variant.

Given that the new operating system and an excellent set of features will make this Huawei wearable a worthy contender, we thought it would make sense to compare it with the existing, top-of-the-line Android-based smartwatch – the Galaxy Watch 3. Of course, HarmonyOS is a different operating system but it draws many comparisons to Android, and it’s fair therefore, to understand which wearable is worth your wrist.

We have already pitted the Galaxy Watch 3 against the Apple Watch Series 6 and the OnePlus Watch, here we will compare vital features to give you an idea of what’s what when it comes to the Huawei wearable and the Galaxy watch.

Huawei is known for offering watches that are very capable fitness trackers, priced competitively. Huawei Watch 3 is built on this winning formula of its precursors along with an emphasis on further accuracy of fitness data and good battery backup. Let’s then learn how the two full-fledged smartwatches packed with fitness functions and personalization options compare against each other.

Design and display

Staying true to its image of a more traditional watch, the Huawei Watch 3 comes in circular stainless steel and ceramic case (Watch 3 Pro has a titanium case). With a 1.43-inch AMOLED display, which supports an impressive 60Hz refresh rate, the watch has a bezel-less design with the screen stretching from edge to edge.

Since the rotating bezel has gone, Huawei has included an Apple Watch-style digital crown, which can rotate to scroll through the features on the screen. Additional pusher at 4 o’clock launches workout mode with a click, so you can instantly launch your running or cycling mode and begin tracking your regime. The watch has 2GB RAM and 16GB of internal memory.

Galaxy Watch 3 on the other hand has a little more premium feel to it with a choice of either stainless steel or titanium case (available in larger display size only). It also has a rounded face and comes with a precision rotatable bezel to toggle between features when you don’t want to interact with the touchscreen. The watch has 41mm and 45mm Super AMOLED display covered in Corning Gorilla Glass DX with only a GB of RAM and 8 gigabytes of onboard storage.

Samsung Galaxy Watch 3 comes with the option of three watchbands – leather, quick-change sport & hybrid – there are plenty more on offer through third-party vendors. The choice of straps is available with the Huawei wearable as well, it comes paired with either leather, silicone, metal link or nylon straps, which can be swapped according to your mood.

Software and features

Huawei Watch 3 comes with a new operating system, which provides the watch a fresh UI. If you have used a Huawei wearable before, it wouldn’t really feel alien. The OS is now called the HarmonyOS but reportedly it’s based on the LiteOS, which has been used in the Huawei smartwatches since it was forced out of Google Wear due to US trade restrictions.

That said, you would instantly latch on to some additional capabilities of the Huawei smartwatch, which now supports 4G eSIM and can be used independently to receive call and message notifications. In addition to allowing voice calls, Huawei has made video calling through MeeTime Service possible directly from the watch.

Huawei Watch 3 has a redesigned home screen with grid-style app layout as opposed to the list format and allows Huawei Music subscribers to stream music on the go. Moreover, the watch now has Huawei AppGallery pre-installed with the hope of attracting more third-party developers to provide content so that it can give competition to Apple and Android watches on the market.

Despite the option, the third-party support on the Huawei Watch 3 is pretty limited at the moment. Music streaming is also confined to only Huawei Music. Not that Galaxy Watch 3 has spectacular third-party support, but comparatively it’s better on the Sammy wearable that touts software, which surpasses the current version of the WearOS as a go-to option for Android users. Both watches are compatible with Android and iOS.

Powered by Tizen OS, the Galaxy Watch 3 allows you to receive smartphone notifications and even view messages from select messaging apps right on the writs. The watch comes in Wi-Fi and LTE variants and supports offline streaming of music locally and through Spotify. You can store music, photos and more on the 8GB internal memory of the watch for on-the-go entertainment. Interestingly, Huawei Watch 3 will not allow contactless payments outside of China, where it’s currently available. Galaxy watch comes with Samsung Pay to make contactless payments from the wrist.

Health and fitness

For health monitoring and activity tracking, the Huawei Watch 3 comes with some of the features and sensors party to its predecessors. This includes heart rate monitoring, SpO2 and sleep tracking. Huawei has not stopped at that, it has now added a temperature sensor to the Watch 3. This sensor can continuously track body temperature by skin contact through the course of the day.

Of course, this not precise for medical purposes, but it definitely alerts when the body temperature is rising or falling. The 50-meter water-resistant watch is pre-installed with Huawei’s own Health app and reportedly has hundreds of different preset workout modes for running, swimming, cycling and more. This list of activities also includes skateboarding and dancing.

Coming on to the Galaxy Watch 3, it has a unique automatic fitness tracking feature that kicks in without you having to set it, so when your running, cycling or swimming, you can take care of your regime and the watch will start tracking on its own. Like the Huawei wearable, this also features heart rate, SpO2 monitoring and sleep tracking.

Samsung wearable is water-resistant up to 50 meters and can also record ECG (electrocardiogram) for 30 seconds. Unlike the other watches, Galaxy Watch 3 has a detailed sleep tracking function to monitor deep sleep, REM and even the wake after sleep onset.

Battery life

Smartwatches are battery guzzlers by virtue. This is because they have to manage continuous health and fitness tracking, display notifications and do a lot more. The trend is changing with new models wherein companies are working on better optimization. Huawei has tried to make some alterations and its new wearable has three days of battery life in regular smartwatch mode, with always-on display deactivated that is.

By deactivating Wi-Fi, cellular connection and GPS, Huawei says the Watch 3 can give up to two weeks of backup. In the low power mode, you can still continue to track activities and workouts while the watch can be instantly juiced up with Qi wireless charging. Galaxy Watch 3, however, disappoints on the battery front. In normal usage, its 340mAh Li-Ion battery can only last for up to two days.


Huawei Watch 3 is available for preorder in China for CNY 2,600 (roughly $410), the sale will begin from June 11. The watch is expected to start retailing globally in select markets from June 18. 41mm Galaxy Watch 3 started at $399 and the 45mm model retailed for $429.99 while for the LTE option you had to shell out $449. The watch can now be bought for as low as $190.

Repost: Original Source and Author Link

Tech News

Huawei HarmonyOS will be coming to these phones next week

Wednesday will be a big day for Huawei and not because it’s announcing new phones. Actually, it might tease some new phones but the focus of its event will be on the software that will run on those phones. It will be HarmonyOS that will be taking center stage, figuratively and literally, and it will finally show to the world Huawei’s vision for a Google-less world. Of course, its success will depend on how many phones it will be available on and it seems that the company will be pushing it hard even on Huawei phones already out in the wild.

HarmonyOS is, of course, expected to be pre-installed in future Huawei devices. These include the still-absent Huawei P50 series as well as the recently rumored MatePad Pro 2. Those, however, will only be a fraction of the phones under Huawei’s command and it seems that the phone maker really wants to push its new OS to everyone.

According to a leak on Weibo, Huawei will be making HarmonyOS available on some existing phones as well. It includes the entire Huawei Mate 40 and P40 series as well as the older MatePad Pro. The foldable Huawei Mate X2 will also be updated and the mid-range Nova 8 and Nova 8 Pro will also be able to taste the new OS.

The leak says that these updates will be available on June 2, right on the day that Huawei debuts HarmonyOS for phones, a.k.a. HarmonyOS 2.0. More interestingly, the source says that phones are also being upgraded in offline stores, suggesting that any new phone from this list that you buy will be running HarmonyOS instead of Android.

That said, there’s still some confusion on whether HarmonyOS 2.0 isn’t just Android with a custom skin, something that Huawei strongly denies. We will at least finally get our answer after June 2, especially when Huawei’s Google-free OS starts rolling out to existing phones.

Repost: Original Source and Author Link

Tech News

Huawei eavesdropped on 6.5 million Dutch mobile users — here’s how

Chinese technology provider Huawei was recently accused of being able to monitor all calls made using Dutch mobile operator KPN. The revelations are from a secret 2010 report made by consultancy firm Capgemini, which KPN commissioned to evaluate the risks of working with Huawei infrastructure.

While the full report on the issue has not been made public, journalists reporting on the story have outlined specific concerns that Huawei personnel in the Netherlands and China had access to security-essential parts of KPN’s network – including the call data of millions of Dutch citizens – and that a lack of records meant KPN couldn’t establish how often this happened.

Both KPN and Huawei have denied any impropriety, though in the years since the 2010 report, Huawei has increasingly found itself labeled a high-risk vendor for telecoms companies to work with, including by the UK’s National Cyber Security Centre.

To better understand this story, and to consider whether other telecoms networks may have had similar security vulnerabilities to KPN’s, we need to look at how complex mobile networks are run. KPN essentially granted Huawei “administrator rights” to its mobile network by outsourcing work to the Chinese firm. Legislation is only now catching up to prevent similar vulnerabilities in telecoms security.

Commercial pressures

Huawei is one of the three dominant radio equipment providers in the world, alongside Ericsson and Nokia. These giant technology companies provide the base stations and equipment that deliver mobile phone signals. Operators like KPN increasingly pay these companies not only to buy the equipment, but also for them to support and maintain it.

The telecoms market in which KPN operates is one of the most price-competitive in the world. European mobile operators saw average revenues per user in 2019 of €14.90 (£12.85) a month, compared with €36.90 a month in the USA. European spend on telecoms services are also reducingyear-on-year as operators compete to offer the best deals to consumers.

Lower revenues force operators to carefully manage costs. This means that operators have been keen to outsource parts of their businesses to third parties, especially since the late 2000s.

Large numbers of highly skilled engineers are an expensive liability to have on the balance sheet, and can often appear underused when things are running smoothly. Such jobs are often outsourced, with personnel transferring to the outsourced provider, to help operators to cut their payroll costs.

Outsourcing gone too far

When everything is working, very few people notice outsourcing. But when things go wrong, outsourcing can often significantly complicate recovery, or create a large “single point of failure” or security issue.

In the UK, for instance, mobile operator O2 has seen at least one outage which has been linked to the use of outsourced functions. Where large numbers of operators rely on the same outsourcing partner, any issue or security breach affecting the outsourced provider can have a widespread impact.

Still, outsourcing by mobile operators is widespread. And firms in the UK and across Europe have often turned to Huawei to provide IT services and to help build core networks. In 2010, Huawei was managing the security-critical functions of KPN’s core network.

Administrator access

At the same time, equipment suppliers like Huawei are trying to move away from merely selling equipment and towards providing a managed service, including installation, maintenance, and support. This helps them create recurring revenue in an industry that has generally been dominated by large five-year or ten-year purchasing cycles.

But as these vendors add services to their repertoire, they gain wider access to the mobile networks they work with. This could include certain security-critical parts of telecoms networks, which are often designed to work in trusted, secure environments.

In the scenario where a vendor like Huawei also provides a managed service, they find themselves sitting in a uniquely privileged position, with inside knowledge of their own equipment, and with direct access to trusted management interfaces.

This creates the high-tech equivalent of putting all your eggs in one basket. It’s akin to giving the combinations of the bank vault to the same security guard in charge of the CCTV camera footage. It’s difficult to reliably monitor operations carried out by the vendor without relying on that vendor’s own software.

In cases where a vendor has been designated as high-risk as a result of their own product security practices, it’s very difficult to know whether that vendor didn’t do anything untoward. This is the situation KPN apparently found themselves in with Huawei back in 2010.

Are changes needed?

With at least one operator aiming to reduce European operating expenditure by €1.2 billion, and 5G deployments bringing new opportunities for managed services and software-based solutions to be used in networks, decisions around outsourcing will continue to play an important role for mobile operators going forwards.

But legislation is rapidly catching up. The UK has proposed a telecoms security bill, and associated draft secondary legislation includes requirements for network operators to monitor all activity carried out by third-party providers, to identify and manage the risks of using them, and to have a plan in place to maintain normal network operations if their supplier’s service is disrupted.

For some operators, it’s conceivable this might mean bringing key skills back in-house to ensure there’s someone watching the (outsourced) watchmen. In the case of KPN, these measures would likely have prevented Huawei from having seemingly unchecked and privileged access to its customers’ mobile data.

This article by Greig Paul, Lead Mobile Networks and Security Engineer, University of Strathclyde is republished from The Conversation under a Creative Commons license. Read the original article.

Repost: Original Source and Author Link

Tech News

Huawei Harmony OS might be adopted by Xiaomi, OPPO, and Vivo

In response to the US’ sanctions and export bans, Huawei has been waging a campaign to rid itself of its dependence on Google’s mobile operating system, or at least its proprietary parts. Although it sounds like an epic endeavor to replace Android, Huawei’s Harmony OS definitely has big ambitions to be present in almost all smart markets. Going just beyond its own smartphones, however, there are now rumors that Huawei wants to spread its new platform to other phones as well, including those from its competitors and compatriots in China.

This sounds almost like a no-go if the initial impressions of Harmony OS 2.0 Beta are anything to go by. While Huawei defends that it isn’t really the final vision, the version that developers were given access to looked a lot like Android underneath a thin custom skin. That does make the transition a bit easier for both developers and smartphone makers but it pales in comparison to the grand and brand new experience that Huawei is promising.

Another limitation, at least as far as phones go, is that Harmony OS is supposedly developed to work primarily on Huawei’s own HiSilicon Kirin processors. Android itself has parts that are developed specifically for certain chipsets, like Qualcomm and MediaTek, among others, which is part of what makes it possible to run it on the majority of the world’s phones.

There is now a report that Harmony OS will undergo development to make it run on Qualcomm’s and MediaTek’s platforms, which means it will be able to run on phones beyond Huawei’s flagships. Even more interesting, however, is the rumor that Chinese smartphone makers are actually interested in using it for their phones. That’s not exactly outside the realm of possibility because Xiaomi, OPPO, and Vivo already run versions of Android without Google in China.

It will really all boil down to what Harmony OS on phones will turn out to be. If it’s simply a skin on top of Android, the change for smartphone makers won’t be that big and might not even make sense at all. If, on the other hand, it will be a truly new mobile OS from the ground up, Google stands to lose whatever small mind share it has in China through Android, even if it doesn’t officially have a presence there through Google Play Store.

Repost: Original Source and Author Link

Tech News

Huawei Mate X2 foldable phone has some durability surprises

Although it has the lion’s share of the attention, Samsung is hardly the only one making foldable phones. There’s, for example, Royole’s FlexPai and, despite troubles with supplies, Huawei’s Mate X. The latter just had its true second-gen model, one that goes in the opposite direction from the first Huawei Mate X’s “outie” fold. The design, however, isn’t the only thing that has changed with the Mate X2 and, fortunately, the changes that come with the redesign also made it a lot more durable in the process.

To be fair, the first Huawei Mate X didn’t exactly have the same disastrous problems as the first Galaxy Fold exactly because of its outie fold. The screen was, of course, still more fragile than your regular phone screen and that design also carried its own problems. Dropping the Huawei Mate X, whether opened or even closed, was equally catastrophic.

With the Huawei Mate X2, the company switched to an innie fold just like the Galaxy Fold and Galaxy Z Fold 2. While more expensive, the new foldable did more than just upgrade the specs but also added some measure of redundancy and reliability. After all, you still have a screen on the outside even in the worst-case scenario that the main foldable screen breaks.

That flexible screen does get scratched with sharp nails but sand and rocks surprisingly do no damage. The Mate X2’s hinge design notably didn’t let sand get out once folded, and JerryRigEverything speculates that it wouldn’t let dust in either. Most importantly, the phone doesn’t even flex when forcibly bent the other way, an effect of the hinge and the two halves locking against each other.

The Huawei Mate X2 turns out to be quite a durable phone despite being a usually fragile foldable one. There’s still a lot of room for improvement on the screen, but every year seems to push technology and materials towards that direction. It will be interesting to see if the Galaxy Z Fold 3, which may be announced in July, has more to offer in this regard and whether it will be more accessible than a $3,000 foldable phone without Google Play Store that you can’t easily buy outside of China.

Repost: Original Source and Author Link


Huawei trained the Chinese-language equivalent of GPT-3

Join Transform 2021 this July 12-16. Register for the AI event of the year.

For the better part of a year, OpenAI’s GPT-3 has remained among the largest AI language models ever created, if not the largest of its kind. Via an API, people have used it to automatically write emails and articles, summarize text, compose poetry and recipes, create website layouts, and generate code for deep learning in Python. But GPT-3 has key limitations, chief among them that it’s only available in English. The 45-terabyte dataset the model was trained on drew exclusively from English-language sources.

This week, a research team at Chinese company Huawei quietly detailed what might be the Chinese-language equivalent of GPT-3. Called PanGu-Alpha (stylized PanGu-α), the 750-gigabyte model contains up to 200 billion parameters — 25 million more than GPT-3 — and was trained on 1.1 terabytes of Chinese-language ebooks, encyclopedias, news, social media, and web pages.

The team claims that the model achieves “superior” performance in Chinese-language tasks spanning text summarization, question answering, and dialogue generation. Huawei says it’s seeking a way to let nonprofit research institutes and companies gain access to pretrained PanGu-α models, either by releasing the code, model, and dataset or via APIs.

Familiar architecture

In machine learning, parameters are the part of the model that’s learned from historical training data. Generally speaking, in the language domain, the correlation between the number of parameters and sophistication has held up remarkably well.

Large language models like OpenAI’s GPT-3 learn to write humanlike text by internalizing billions of examples from the public web. Drawing on sources like ebooks, Wikipedia, and social media platforms like Reddit, they make inferences to complete sentences and even whole paragraphs.

Huawei language model

Above: PanGu-α generating dialog for a video game.

Akin to GPT-3, PanGu-α is what’s called a generative pretrained transformer (GPT), a language model that is first pretrained on unlabeled text and then fine-tuned for tasks. Using Huawei’s MindSpore framework for development and testing, the researchers trained the model on a cluster of 2,048 Huawei Ascend 910 AI processors, each delivering 256 teraflops of computing power.

To build the training dataset for PanGu-α, the Huawei team collected nearly 80 terabytes of raw data from public datasets, including the popular Common Crawl dataset, as well as the open web. They then filtered the data, removing documents containing fewer than 60% Chinese characters, less than 150 characters, or only titles, advertisements, or navigation bars. Chinese text was converted into simplified Chinese, and 724 potentially offensive words, spam, and “low-quality” samples were filtered out.

One crucial difference between GPT-3 and PanGu-α is the number of tokens on which the models trained. Tokens, a way of separating pieces of text into smaller units in natural language, can be either words, characters, or parts of words. While GPT-3 trained on 499 billion tokens, PanGu-α trained on only 40 billion, suggesting it’s comparatively undertrained.


Huawei language model

Above: PanGu-α writing fiction.

Image Credit: Huawei

In experiments, the researchers say that PanGu-α was particularly adept at writing poetry, fiction, and dialog as well as summarizing text. Absent fine-tuning on examples, PanGu-α could generate poems in the Chinese forms of gushi and duilian. And given a brief conversation as prompt, the model could brainstorm rounds of “plausible” follow-up dialog.

This isn’t to suggest that PanGu-α solves all of the problems plaguing language models of its size. A focus group tasked with evaluating the model’s outputs found 10% of them to be “unacceptable” in terms of quality. And the researchers observed that some of PanGu-α’s creations contained irrelevant, repetitive, or illogical sentences.

Huawei language model

Above: PanGu-α summarizing text from news articles.

The PanGu-α team also didn’t address some of the longstanding challenges in natural language generation, including the tendency of models to contradict themselves. Like GPT-3, PanGu-α can’t remember earlier conversations, and it lacks the ability to learn concepts through further conversation and to ground entities and actions to experiences in the real world.

“The main point of excitement is the extension of these large models to Chinese,” Maria Antoniak, a natural language processing researcher and data scientist at Cornell University, told VentureBeat via email. “In other ways, it’s similar to GPT-3 in both its benefits and risks. Like GPT-3, it’s a huge model and can generate plausible outputs in a variety of scenarios, and so it’s exciting that we can extend this to non-English scenarios … By constructing this huge dataset, [Huawei is] able to train a model in Chinese at a similar scale to English models like GPT-3. So in sum, I’d point to the dataset and the Chinese domain as the most interesting factors, rather than the model architecture, though training a big model like this is always an engineering feat.”


Indeed, many experts believe that while PanGu-α and similarly large models are impressive with respect to their performance, they don’t move the ball forward on the research side of the equation. They’re prestige projects that demonstrate the scalability of existing techniques, rather, or that serve as a showcase for a company’s products.

“I think the best analogy is with some oil-rich country being able to build a very tall skyscraper,” Guy Van den Broeck, an assistant professor of computer science at UCLA, said in a previous interview with VentureBeat. “Sure, a lot of money and engineering effort goes into building these things. And you do get the ‘state of the art’ in building tall buildings. But there is no scientific advancement per se … I’m sure academics and other companies will be happy to use these large language models in downstream tasks, but I don’t think they fundamentally change progress in AI.”

Huawei language model

Above: PanGu-α writing articles.

Even OpenAI’s GPT-3 paper hinted at the limitations of merely throwing more compute at problems in natural language. While GPT-3 completes tasks from generating sentences to translating between languages with ease, it fails to perform much better than chance on a test — adversarial natural language inference — that tasks it with discovering relationships between sentences.

The PanGu-α team makes no claim that the model overcomes other blockers in natural language, like answering math problems correctly or responding to questions without paraphrasing training data. More problematically, their experiments didn’t probe PanGu-α for the types of bias and toxicity found to exist in models like GPT-3. OpenAI itself notes that GPT-3 places words like “naughty” or “sucked” near female pronouns and “Islam” near terms like “terrorism.” A separate paper by Stanford University Ph.D. candidate and Gradio founder Abubakar Abid details the inequitable tendencies of text generated by GPT-3, like associating the word “Jews” with “money.”

Carbon impact

Among others, leading AI researcher Timnit Gebru has questioned the wisdom of building large language models, examining who benefits from them and who’s disadvantaged. A paper coauthored by Gebru earlier this year spotlights the impact of large language models’ carbon footprint on minority communities and such models’ tendency to perpetuate abusive language, hate speech, microaggressions, stereotypes, and other dehumanizing language aimed at specific groups of people.

In particular, the effects of AI and machine learning model training on the environment have been brought into relief. In June 2020, researchers at the University of Massachusetts at Amherst released a report estimating that the amount of power required for training and searching a certain model involves the emissions of roughly 626,000 pounds of carbon dioxide, equivalent to nearly 5 times the lifetime emissions of the average U.S. car.

Huawei language model

Above: PanGu-α creating poetry.

While the environmental impact of training PanGu-α is unclear, it’s likely that the model’s footprint is substantial — at least compared with language models a fraction of its size. As the coauthors of a recent MIT paper wrote, evidence suggests that deep learning is approaching computational limits. “We do not anticipate that the computational requirements implied by the targets … The hardware, environmental, and monetary costs would be prohibitive,” the researchers said. “Hitting this in an economical way will require more efficient hardware, more efficient algorithms, or other improvements such that the net impact is this large a gain.”

Antoniak says that it’s an open question as to whether larger models are the right approach in natural language. While the best performance scores on tasks currently come from large datasets and models, whether the pattern of dumping enormous amounts of data into models will pay off is uncertain. “The current structure of the field is task-focused, where the community gathers together to try to solve specific problems on specific datasets,” she said. “These tasks are usually very structured and can have their own weaknesses, so while they help our field move forward in some ways, they can also constrain us. Large models perform well on these tasks, but whether these tasks can ultimately lead us to any true language understanding is up for debate.”

Future directions

The PanGu-α team’s choices aside, they might not have long to set standards that address the language model’s potential impact on society. A paper published by researchers from OpenAI and Stanford University found that large language model developers like Huawei, OpenAI, and others may only have a six- to nine-month advantage until others can reproduce their work. EleutherAI, a community of machine learning researchers and data scientists, expects to release an open source implementation of GPT-3 in August.

The coauthors of the OpenAI and Stanford paper suggest ways to address the negative consequences of large language models, such as enacting laws that require companies to acknowledge when text is generated by AI — perhaps along the lines of California’s bot law. Other recommendations include:

  • Training a separate model that acts as a filter for content generated by a language model
  • Deploying a suite of bias tests to run models through before allowing people to use the model
  • Avoiding some specific use cases

The consequences of failing to take any of these steps could be catastrophic over the long term. In recent research, the Middlebury Institute of International Studies’ Center on Terrorism, Extremism, and Counterterrorism claims that GPT-3 could reliably generate “informational” and “influential” text that might radicalize people into violent far-right extremist ideologies and behaviors. And toxic language models deployed into production might struggle to understand aspects of minority languages and dialects. This could force people using the models to switch to “white-aligned English,” for example, to ensure that the models work better for them, which could discourage minority speakers from engaging with the models to begin with.

Given Huawei’s ties with the Chinese government, there’s also a concern that models like PanGu-α could be used to discriminate against marginalized peoples including Uyghurs living in China. A Washington Post report revealed that Huawei tested facial recognition software that could send automated “Uighur alarms” to government authorities when its camera systems identified members of the minority group.

We’ve reached out to Huawei for comment and will update this article once we hear back.

“With PanGu-α, like with GPT-3, there are risks of memorization, biases, and toxicity in the outputs,” Antoniak said. “This suggests that perhaps we should try to better model how humans learn language.”


VentureBeat’s mission is to be a digital town square for technical decision-makers to gain knowledge about transformative technology and transact.

Our site delivers essential information on data technologies and strategies to guide you as you lead your organizations. We invite you to become a member of our community, to access:

  • up-to-date information on the subjects of interest to you
  • our newsletters
  • gated thought-leader content and discounted access to our prized events, such as Transform 2021: Learn More
  • networking features, and more

Become a member

Repost: Original Source and Author Link