Hasta La Gadget!

The demise of Humane is perhaps the most predictable tech story of 2025. The company tried to build some buzz around its AI Pin in late 2023, marketing the device as a tiny replacement for smartphones and playing up the fact that Humane’s co-founders were former Apple employees. The problem was that it wasn’t really clear what the AI Pin would do to justify its $700 asking price (plus a $24/month subscription).

It didn’t take long for things to spiral out of control. The AI Pin was released in April of 2024 to some of the worst reviews I’ve ever seen for a consumer tech product. Just a month after launch, reports surfaced that the company was already trying to be acquired — for the positively ludicrous sum of $750 million to $1 billion dollars. At the same time, it was rumored that Humane sold only 10,000 Pins, a far cry from the 100,000 they had planned for. HP was named as a potential suitor last May, and the company smartly waited until they could pluck Humane for a comparatively paltry $116 million.

As it turned out, waving your hands and shouting about the promise of AI doesn’t make it any easier to build compelling hardware — we slammed its high price, terrible battery life, slow performance, excessive heat and hard-to-use projected display. (And Engadget was far from the only publication to eviscerate this device.)

Hayato Huseman for Engadget

For a device whose main interface was conversational, the challenges the AI Pin had answering questions or executing commands made it a non-starter. Even when it did what was asked, it did it in some strange ways, like sending generic texts instead of letting you dictate what exactly you wanted to say. Its camera rarely worked as intended; after taking photos and viewing them on the projector, the Pin would get extremely warm and sometimes just shut down entirely. Speaking of that projection screen, it was nearly impossible to actually see it outdoors, even on a cloudy day. And interacting with it made our reviewer Cherlynn Low want to “rip [her] eyes out.”

Oh, let’s not forget that its extended battery case was recalled because it was a full-on fire hazard!

The whole debacle is an illustrative example of how most consumer-grade AI isn’t ready for prime time. Google and Apple may be trying to shove Gemini and Apple Intelligence down our throats on nearly every product they make, but those tools are additive, built on top of the strong foundations of each company’s existing platforms. In Humane’s case, there was nothing to fall back on. And the combo of terrible voice responses and recognition paired with a projector display that was not at all ready for prime time (not to mention the other hardware failings) was far too much to overcome.

There was simply no intelligence to be found here, artificial or otherwise.

To be fair to Humane, building hardware is notoriously difficult; first-generation products often have glaring flaws, even when you’re talking about massive companies like Apple. The first iPhone, iPad and Apple Watch may have shown huge promise, but they also had strange omissions and performance issues that made them hard to recommend to everyone.

Some companies manage to figure that out; Pebble is a great example of a hardware startup that survived some early issues to find some success. It beat Apple, Samsung and Google to the smartwatch market, and did it in a way that has yet to be imitated (maybe that’s why the company is being resurrected). The Pebble wasn’t exactly the most elegant piece of hardware, but both the watch and its software worked well enough that it paved the way for the more advanced smartwatches we have today. Of course, that wasn’t enough to keep Pebble alive, as the company eventually filed for insolvency and had its assets picked up by Fitbit (which was later purchased by Google, if you’re keeping track).

Humane’s situation at launch wasn’t entirely different — it was trying to build a new type of hardware altogether, and history tells us that the first products in a new space are going to be far from perfect. But, if there had at least been a glimmer of useful software, Humane might have survived to improve on those hardware problems with a future version. But its assistant was so bad that it killed any potential that the AI Pin had. There was simply no intelligence to be found here, artificial or otherwise.

My takeaway from the Humane disaster is that it’s too soon to spend your hard-earned money on the promise of AI — the marketplace isn’t solidified at this point, and trusting a brand-new company like Humane to get this sort of thing right is several bridges too far. (If you’re not convinced, look at the similarly flawed Rabbit R1.) Apple Intelligence is still half-baked at best, but at least you can turn it off and ignore it. But the AI Pin, well, relied completely on AI, and it wasn’t just “not ready” — it was one of the worst devices we’ve ever tried in our nearly 21 years as a publication. For some thousands of early adopters, that means their Pin will be a brick in just a few days, with no financial compensation coming their way. But hey, at least it will still be able to tell you its battery level.

This article originally appeared on Engadget at https://www.engadget.com/ai/the-humane-ai-pin-debacle-is-a-reminder-that-ai-alone-doesnt-make-a-compelling-product-190119112.html?src=rss

The embattled EV truck maker Nikola is going kaput. The company just announced a filing for Chapter 11 bankruptcy protection and a plan to sell off assets. These assets will be sold at an auction, pending court approval. Nikola noted it has $47 million in cash on hand to fund the bankruptcy proceedings and begin the sale process. The company has reported between $500 million to $1 billion in assets, but liabilities totaling $1 billion to $10 billion, according to a court filing.

“Like other companies in the electric vehicle industry, we have faced various market and macroeconomic factors that have impacted our ability to operate,” said Steve Girsky, President and CEO of Nikola.

Those market and macroeconomic factors are real, as EV adoption growth rates have slowed a bit. However, Nikola is a special case. The company has been embroiled in numerous scandals in recent years, so this news caps off a particularly dire fall from grace for the once-buzzy EV maker.

Nikola was founded in 2015 with a mission to create zero-emissions heavy trucks. The company secured a lucrative partnership with GM in 2020, but things fell apart soon thereafter. It was credibly accused of fraud, thanks to a report by short-selling firm Hindenburg Research. This report even included a video showing a Nikola truck rolling down a hill to simulate driving.

This “elaborate ruse” caused the SEC to begin an investigation which led to founder Trevor Milton stepping down as board chair and CEO. Later, he was arrested and indicted on fraud charges. Milton was found guilty and sentenced to four years in prison. The company settled with the SEC for $125 million. It goes without saying that GM backed out of that partnership.

The company did go public in 2020 and began shipping its first trucks in 2021. However, reports indicated that Nikola was losing hundreds of thousands of dollars on every truck it sold. All told, the company only made around 600 vehicles, many of which were recalled due to a variety of defects.

This subpar performance caused the stock to crater, stumbling from a high of over $1,000 per share to, as of this writing, $0.47 per share. I guess only one company named after the guy who invented alternating current (AC) energy is allowed to consistently fall below expectations without making a dent on its valuation.

This article originally appeared on Engadget at https://www.engadget.com/transportation/evs/ev-truck-maker-nikola-has-finally-called-it-quits-172516625.html?src=rss

Despite the obvious benefits of electric cars, Toyota spent the last decade insisting hydrogen would win out in the end. But, as the company announces its third generation fuel cell system, you can tell it’s finally ready to tacitly admit defeat: the new cell is designed for industrial applications, where hydrogen clearly always made more sense. 

The new cell is designed to meet the “particular needs of the commercial sector,” focusing on durability equal to a diesel engine. It’s a lot more fuel efficient, cheaper to make and outputs twice as much power while sitting in the same footprint as the second-generation model. Given Toyota’s love of shrinking its engine technology, that size wasn’t a factor here is enormously telling of where it envisions these cells being used.

Toyota could never make the economic or technological argument for hydrogen cars as a better option than electricity (the Mirai, Toyota’s flagship hydrogen EV, has managed to sell just 28,000 models since its 2014 birth). But for heavy duty vehicles, where battery weight and power are more pressing concerns, hydrogen’s flaws turn into assets. Trucks, construction vehicles, trains, ships and backup generators — less at risk from the lack of general-purpose hydrogen infrastructure — are welcome homes for fuel cells.

This article originally appeared on Engadget at https://www.engadget.com/transportation/toyota-kinda-sorta-gives-up-on-hydrogen-cars-151059624.html?src=rss

It’s honestly difficult to remember the simpler days of video card shopping, before crypto fanatics, supply chain issues and pandemic demand pushed GPUs far beyond rising manufacturer prices. Ideally, I’d like to tell you that NVIDIA’s $549 RTX 5070 and $749 RTX 5070 Ti are more reasonable alternatives to the $2,000 RTX 5090 and $1,000 5080. But card makers and retailers have already pushed RTX 5070 prices far beyond those MSRPs. Our review unit, the ASUS 5070 Ti Prime, is currently selling for $900 at Best Buy and $750 at Newegg (we’ll see how long that lasts). And of course, it’s out of stock at both stores.

While I can’t guarantee the actual cost for any RTX 5070 Ti card, I can say this: they’ll definitely be solid 4K performers for far less than the RTX 5080 and 5090. But if you’re not desperate for an upgrade, it’s worth waiting a few months for inventory and prices to stabilize.

Hardware

Based on its specs and (hopeful) pricing, the RTX 5070 Ti currently offers the best balance between performance and value in NVIDIA’s lineup. It features 8,960 CUDA cores and 16GB of GDDR7 VRAM, well below the 5080’s 10,752 CUDA cores, but at least those cards have the same amount of memory. The cheaper 5070 comes with just 12GB of VRAM, which could be a problem when gaming in 4K.

Our ASUS 5070 Ti card is fairly nondescript, with three fans, a plastic frame and a standard heatsink design. You can choose between performance and quiet BIOS modes, which only changes how aggressive the fans are. Its 2.5-slot design makes it tiny enough for small form-factor cases, though I noticed it was actually slightly larger than the RTX 5090 Founders card.

The 5070 Ti could also easily fit into more gaming rigs without requiring a power supply upgrade. It has a peak power draw of 300 watts, compared to the 5080’s 360W and the 5090’s whopping 575W. That means the 5070 Ti should be able to run comfortably with an 850W PSU, without needing to make the leap to a massive 1,000W unit.

Devindra Hardawar for Engadget

What really makes this GPU special, though, is that it fully supports multi-frame generation in DLSS 4, NVIDIA’s AI upscaling technology. That allows the GPU to generate up to 3 frames with AI for every frame rendered in real-time. It also lets NVIDIA claim that the 5070 can match the speeds of the $1,599 RTX 4090. While you could argue those frames are just “fake” to make benchmarks look better, my time with the RTX 5070 Ti and 5090 has shown that they do lead to a smoother gameplay experience.

On top of multi-frame generation, other DLSS 4 features are also trickling down to earlier NVIDIA cards. As I noted in my 5090 review, “RTX 40 cards will be more efficient with their single-frame generation, while RTX 30 and 20 cards will also see an upgrade from AI transformer models used for ray reconstruction (leading to more stable ray tracing), Super Resolution (higher quality textures) and Deep Learning Anti-Aliasing (DLAA).”

Devindra Hardawar for Engadget

In use: A capable 4K gaming GPU

First things first: The RTX 5070 Ti is only slightly faster than the 4070 and 4070 Ti in most benchmarks. The new card is 17 percent ahead of the 4070 Ti in the 3DMark Timespy Extreme test, and 21 percent faster than the 4070 Ti Super in Speedway bench. The difference is even smaller in raw computing and rendering tasks: The 5070 Ti scored a mere 8 percent more than the 4070 Ti in the Geekbench 6 GPU benchmark.

But, of course, actual gaming performance matters more than benchmarks. And if you’re playing something with support for DLSS 4, you’ll certainly notice some improvements. Dragon Age: The Veilguard held a steady 200fps in 4K with 4X multi-frame generation, ray tracing and graphics settings maxed out, On the 4070 Ti, I typically saw between 90fps and 100fps with those same graphics settings and DLSS 3.5’s single frame generation.

Now, I can’t actually say the game looked twice as smooth on my Alienware 32-inch QD-OLED monitor, but it definitely looked silky over the hours I’ve tested. There weren’t any weird upscaling artifacts, those frames felt real. It’s also worth noting the RTX 5090 clocked 240fps in Dragon Age with the same graphics settings. Perhaps my CPU held it back a bit (I’m running a Ryzen 9 7900X), but the 5070 Ti’s performance was still remarkably close while being a much cheaper GPU.

Cyberpunk 2077 also played like a dream in 4K in ray tracing overdrive mode with multi-frame gen, reaching 150fps on average. That’s well below the 5090’s stunning 250fps figure, but it’s still impressive for a game that used to bring powerful rigs to their knees. Cyberpunk also hit 230fps in 1,440p with those same settings, which also upscales beautifully to 4K screens.

For games without DLSS 4, like Halo Infinite, the 5070 Ti was still a solid performer, reaching an average of 140fps with maxed-out graphics and ray tracing. In comparison, the 5090 hit 180fps on average. Even if you’re lucky enough to have a 240Hz 4K monitor, I’d bet even demanding gamers would be just fine with the 5070 Ti’s speeds. But if you care more about framerates than resolution, it’ll still have you covered. I saw 220fps in Halo Infinite in 1,440p, and 320fps in 1080p.

The ASUS 5070 Ti typically idled between 30C and 35C, and it quickly reached up to 65C under load. Its fan array isn’t as sophisticated as the 5090 Founder’s card, but it still managed to cool down the card below 40C in around 15 seconds.

Devindra Hardawar for Engadget

Should you buy the RTX 5070 Ti?

Simply put, the RTX 5070 Ti handled just about everything I threw at it, and I didn’t find myself missing the 5090 too much (aside from bragging rights). Unfortunately, I haven’t had a chance to test the RTX 5080, but given its high cost, it’s still something I’d have trouble recommending to anyone.

The real question for gamers right now is: Do you need the RTX 5070 Ti’s 16GB of VRAM and higher CUDA count? If you’re aiming to play in 4K most of the time, it’ll be worth having more than just the 5070’s 12GB of RAM. Games are becoming more complex every year, so it likely won’t be long before you’ll actually need 16GB of VRAM to play 4K games comfortably. But if you’re living the 1,440p life, then 12GB will likely be enough for years to come.

DLSS 4’s multi-frame gen is the biggest draw for NVIDIA’s 50-series cards, and it’s mostly useful for 4K gaming. So if you’re happy with your 40-series GPU and don’t need to push a 4K 240Hz monitor to its limit, there’s not much reason to upgrade. For 30- and 20-series owners though, your patience will be rewarded.

As I mentioned before, it’s still worth waiting a few months to see how prices settle. If you’re lucky enough to score the RTX 5070 Ti for $750, go for it. But it’s far less compelling at $900 or above. At that point, you’re just way too close to the 5090’s $1,000 MSRP.

We’re still waiting to see how AMD’s upcoming RDNA 4 Radeon 9070 and 9070 XT GPUs will perform, but they’re being positioned as direct competitors to the 5070 and 5070 Ti. AMD finally has DLSS-like AI-powered upscaling coming this year, so the difference between its cards and NVIDIA’s may be slimmer than usual. But NVIDIA also has a dramatic head start, and it’ll likely take a while for AMD’s Fluid Motion Frames technology to catch up on multi-frame generation.

Devindra Hardawar for Engadget

Wrap-up: A great 4K card… if you can get it close to $750

The RTX 5070 Ti won me over in ways I didn’t expect. I knew it would be a tad faster than the 4070 Ti Super, but with the addition of multi-frame generation, it’s also a far more capable 4K card. And it’s definitely more future-proof than the 5070, since it has 16GB of VRAM like the 5090.

While I think the $549 5070 remains the most intriguing entry of NVIDIA’s new family, it’s nice to see that there’s something for sensible enthusiasts between that and the $1,000 5090. And yes, it’s still strange to call a $750 video card “sensible.”

This article originally appeared on Engadget at https://www.engadget.com/gaming/pc/nvidia-geforce-5070-ti-review-a-sensible-4k-powerhouse-for-749-140023082.html?src=rss

xAI has launched its Grok 3 models during a livestream with Elon Musk, who said they were “an order of magnitude more capable than Grok 2.” The Grok 3 mini model can answer questions quickly, but it’s not as accurate as the other models in the family. Meanwhile, the Grok 3 Reasoning and Grok 3 mini Reasoning models are capable of mimicking human-like reasoning when it comes to analyzing information the user needs. 

Other examples of AI models capable of reasoning tasks are DeepSeek’s R1 and OpenAI’s o3-mini. According to TechCrunch, xAI claimed during the event that Grok 3 Reasoning performed better than the best version of o3-mini on several benchmarks. Grok 3’s features will initially be available to subscribers paying for X’s Premium+ tier, which now costs $40 a month in the US. (X raised the Premium+ tier’s pricing from $16 to $22 in December — now, less than two months later, it’s almost twice as expensive.) They will also be available through an upcoming separate subscription option for the standalone Grok app and Grok on the web. Based on leaked information, the subscription option will be called SuperGrok and will cost $30 a month. 

With the Grok 3 models enabled, users will be able to ask the chatbot to “Think” if they want to tap its reasoning capabilities for mathematics, science and programming questions. For even more complex queries, they can use the “Big Brain” function that requires additional computing. The models’ reasoning capabilities power a new Grok feature called DeepSearch, which xAI describes as the “next generation search engine.” DeepSearch will scan the internet and X, formerly Twitter, to conjure a brief summary for research inquiries. 

In addition to launching the Grok 3 models, xAI also revealed during the event that the Grok app will get a “voice mode” within a week, giving it synthesized voices to converse with users. Grok 2, the company’s older models, will be open sourced in the coming months. 

This article originally appeared on Engadget at https://www.engadget.com/ai/xai-launches-grok-3-ai-claiming-it-is-capable-of-human-reasoning-140007172.html?src=rss