ChatGPT gets a big new rival as Anthropic claims its Claude 3 AIs beat it

AI company Anthropic is previewing its new “family” of Claude 3 models it claims can outperform Google’s Gemini and OpenAI’s ChatGPT across multiple benchmarks.

This group consists of three AIs with varying degrees of “capability”. You have Claude 3 Haiku down at the bottom, followed by Claude 3 Sonnet, and then there’s Claude 3 Opus as the top dog. Anthropic claims the trio delivers “powerful performance” across the board due to their multimodality, improved level of accuracy, better understanding of context, and speed. What’s also notable about the trio is they’ll be more willing to answer tough questions. 

Anthropic explains older versions of Claude would sometimes refuse to answer prompts that pushed the boundaries of the safety guardrails. Now, the Claude 3 family will have a more nuanced approach with its responses allowing them to answer those tricky questions.

Despite the all-around performance boost, much of the announcement is focused on Opus as being the best in all of these areas. They go so far as to say the model “exhibits near-human levels of comprehension… [for] complex tasks”.

Specialized AIs

To test it, Anthropic put Opus through a “Needle In a Haystack” or NIAH evaluation to see how well it’s able to recall data. As it turns out, it’s pretty good since the AI could remember information with almost perfect detail. The company goes on to claim that Opus is quite the smart cookie able to solve math problems, generate computer code, and display better reasoning than GPT-4

The technology isn’t without its quirks. Even though Anthropic states their AIs have improved accuracy, there is still the problem of hallucinations. The responses the models churn out may contain wrong information, although they are greatly reduced compared to Claude 2.1. Plus, Opus is a little slow when it comes to answering a question with speeds comparable to Claude 2.

Of course, this isn’t to say Haiku or Sonnet are lesser than Opus as they have specific use cases. Haiku, for example, is great at giving quick replies and grabbing information “from unstructured data”. Also, it’s not as good at answering math questions as Opus. Sonnet is a larger-scale model meant to help people save time at menial tasks and even parse lines of “text from images”, while Opus is ideal for large-scale operations.

Changing the internet

Both Sonnet and Opus are currently available for purchase although there is a free version of Claude on the company website. A launch date was not given for Haiku, but Anthropic states it’ll be released soon. 

As you can probably guess, the Claude 3 trio is meant more for businesses looking to automate certain workloads. Your experience with the group will likely come in the form of an online chatbot. Amazon recently announced it’s going to be implementing Anthropic’s new AIs into AWS (Amazon Web Services) giving websites on the platform a way to create a customized Claude 3 model to suit the needs of brands and their customers.

If you're looking for a model suited for everyday use, check out TechRadar's list of the best AI content generators for 2024.

You might also like

TechRadar – All the latest technology news

Read More

Forget completing Zelda as fast as possible – the latest speedrun to beat is an installation of Windows 10

Speed running has long been a pursuit undertaken by gamers, but in more recent times we’ve seen some more left-field speedruns, and here’s another one: a super-quick installation of Windows 10.

Yes, you read that right, the challenge of installing the Microsoft operating system as fast as possible was taken on by NTDev, who is the developer of the lightweight version of Windows 10 known as Tiny10 (and its sibling version for Windows 11, named Tiny11 as you might guess).

NTDev managed to install their own version of Windows 10, meaning Tiny10, in just over 100 seconds, so not far off a minute and a half.

Pretty impressive? Yes, but there is a caveat, and it’s not a tiny one: the Tiny10 version used was an old and further modified installer which was optimized with speed running in mind.

In fact, the attempt was made using a Tiny10 install based on Windows 10 1809, which is the October 2018 Update.


Analysis: Rules of the game?

To be fair, a working version of Windows 10 was still installed – well, we assume – and the caveat of it being an old Tiny10 build, further tinkered with and streamlined to set up at lightning speed, isn’t really a criticism as such. Depending on how you look at things, anyway…

After all, a speed run which was just the standard install process, based on how fast you could click, would be deadly dull and pointless, of course. For us, the hacking away at the OS to run faster in setup is the speed running, in the same way that, for example, bouncing off walls or strafe running and so on is for gamers (and often finding weird glitches to exploit in one way or another).

As long as the Windows version that’s up and running actually functions, the speed run should count in our books. But, if we’re going to get serious for a moment, this does bring up complicated questions about what could be legitimately cut out, and what features must remain, if you wanted to standardize OS speed running rules in some way.

The other (perhaps simpler, but less fun arguably) route to go would be having a standard installation mandated, with no tweaking, so the skill would be in the hardware setup. However, even then, there would need to be rules on what setups and components were fair game. (Naturally NTDev tells us they were using the fastest storage and RAM they could get their hands on).

At any rate, this is an entertaining feat to watch, especially given that a typical Windows installation will probably eat around half an hour of your life (and most definitely isn’t something you’d want to watch). All that’s missing here, frankly, is a ‘Ride of the Valkyries’ soundtrack which surely should have been the choice of background music (not the feeble electronic beat supplied).

There’s also a recent Windows 11 speedrun from NTDev which is completed in a somewhat longer, but still impressive, three minutes (see above).

How long does it take you to install Windows 11? For us, it’s two-and-a-third years and counting (sorry Microsoft, couldn’t resist).

Via Tom's Hardware

You might also like…

TechRadar – All the latest technology news

Read More

The Meta Quest 3 doesn’t beat my 4K TV for Xbox gaming, but I don’t care – I love it

After Thursday’s surprise Xbox Cloud Gaming launch on the Meta Quest 3, I've spent most (read: too much) of my evenings trying the service out. And while it doesn’t hold a candle to my more traditional 4K TV and console setup from a technical perspective, its portability more than makes up for that.

For the uninitiated, Xbox Cloud Gaming is basically Netflix for video games. For a monthly fee of $ 16.99 / £12.99 / AU$ 18.95, you can stream titles from a massive catalog of content to your phone, PC, Xbox console, and now Meta Quest 3, Quest 2, and Quest Pro. The advantage, of course, is you don’t need super powerful hardware to play the latest games – they’re run on high-end machines many hundreds (maybe thousands) of miles away and just use your device as a screen and a relay for your controller inputs.

When playing in the real world, you’re limited to the size of your TV, phone screen, or computer monitor. In VR, you can enjoy playing these games on a gigantic virtual display – with the size becoming especially apparent when using the Quest 3’s mixed-reality mode. The Large and Extra-large screen options were bigger than any TV I’ve seen before – even the ridiculous displays shown off at tech trade shows – and it made me feel like I was gaming in my own private movie theater.

My view as I play Xbox games on my Quest 3 while all cozy in bed. The screen floats in front of me while I hold a white Xbox controller.

This screenshot doesn’t do the virtual screen size justice (Image credit: Future / Hamish Hector)

The trade-off is the graphics quality leaves something to be desired. Xbox Cloud Gaming can apparently stream 1080p (full-HD) at 60fps gameplay – but I’m certain the quality I experienced wasn’t this high. That's most likely due to a combination of the Quest 3’s display specs, my internet connection throttling the app’s abilities, and the gigantic virtual screen not giving visual blemishes anywhere to hide; instead blowing them up to make them more noticeable than ever.

Head in the cloud

Yet, when I lay in bed wearing my Quest 3 with the virtual display floating on the ceiling above me using mixed reality, I was still utterly lost in Starfield until the early hours of the morning. I only stopped when my headset alerted me that its charge was low and I realized it was well past 2 am.

This portability – to be able to play anywhere with a strong enough internet connection – is why cloud gaming in VR succeeds. You can play in bed, during your commute, at a coffee shop while waiting for your friends, or pretty much anywhere you can think of. Yes, you can stream Xbox titles to your phone, too, but the display is small, and the experience just isn’t as immersive as the virtual screen that wraps around you.

To this end, the VR headset is starting to borrow elements of the many AR glasses I’ve tested over the past year – the likes of the Xreal Air 2 or Rokid Max – and I’m pretty darn excited about it. These AR specs connect to a compatible phone, laptop, or games console using a USB-C adapter and virtually project the screen in front of you.

The Xreal Air 2 Pro AR smart glasses next to the Xreal Beam hub, they're both on a wooden table in front of a brick wall

The Xreal Air 2 Pro AR smart glasses and Beam (Image credit: Future)

The clear advantage of the Quest hardware is it’s more than just a wearable projector – it’s a whole spatial computer in its own right that can do incredible things without any external hardware. What’s more, while the Quest 3 is pricier on paper – by about $ 100 / £100 / AU$ 150 depending on the glasses – once you factor in the need to buy a few not-so-optional add-ons to get the most out of AR specs, the cost difference is negligible (the Quest 3 might even end up being cheaper). Not to mention that you get far more bang for your buck from a VR headset.

While going fully wireless has advantages, having tried Xbox Cloud Gaming I’m now even more desperate for Meta’s VR hardware to start supporting wired connections to a greater range of devices. I’d love to use a Steam Deck, Nintendo Switch, PS5, and other gadgets in a similarly immersive way without needing to splash out on AR specs. Hopefully, such features will be added to the Quest platform in the not-too-distant future.

Some work is needed 

Before Meta works on adding these capabilities, though, I’d like it and Microsoft to make a few updates to the Xbox Cloud Gaming app. Considering this app was announced over a year ago at Meta Connect 2022, I’m surprised it’s so basic and lacking a few features that feel like no-brainers.

The first is an easy way to position the screen. From what I can tell, the only virtual display controls within the app are the size options. If you want to move the screen to a different position – which is essential for playing lying down – you need to first press the Oculus button to open up your quick menu bar. Then, when you grab the bar to move it, the Xbox screen will move with it.

This workaround is effective but not intuitive; I stumbled into it completely by accident.

The second is an emulator so the Quest handsets can double as an Xbox controller. Admittedly, this may be tough, as the controllers are almost identical, but the Quest controllers lack the D-Pad and third menu button. Even if this emulated controller is only compatible with a small selection of titles, it would offer users a great way to test out Cloud Gaming before they invest in a wireless Xbox controller (they aren’t massively expensive, but they aren’t cheap either – they’re pricier than most VR games).

Lastly, it would be helpful if there was an in-app way to see how stable your connection is, see what resolution and framerate you’re getting, and choose if you want to optimize for graphics or performance. 

All that said, despite its deficiencies, the Xbox Cloud Gaming app is a must-try – especially for those with an Xbox Game Pass Ultimate subscription and compatible controller already. While I’ll still spend a lot of time gaming on my TV, I can honestly see this VR app becoming one of my most used in 2024. It might even convince me to start taking my Quest 3 with me everywhere so I can game on the go.

You Might Also Like

TechRadar – All the latest technology news

Read More

Meta was late to the AI party – and now it’ll never beat ChatGPT

Meta – the tech titan formerly known as Facebook – desperately wants to take pole position at the forefront of AI research, but things aren’t exactly going to plan.

As reported by Gizmochina, Meta lost a third of its AI research staff in 2022, many of whom cited burnout or lack of faith in the company’s leadership as their reasons for departing. An internal survey from earlier this year showed that just 26% of employees expressed confidence in Meta’s direction as a business.

CEO Mark Zuckerberg hired French computer scientist and roboticist Yann LeCun to lead Meta’s AI efforts back in 2013, but in more recent times Meta has visibly struggled to keep up with the rapid pace of AI expansion demonstrated by competing platforms like ChatGPT and Google Bard. LeCun was notably not among the invitees to the White House’s recent Companies at the Frontier of Artificial Intelligence Innovation summit.

That’s not to say that Meta is failing completely in the AI sphere; recent reveals like a powerful AI music creator and a speech-generation tool too dangerous to release to the public show that the Facebook owner isn’t exactly sitting on its hands when it comes to AI development. So why is it still lagging behind?

Abstract artwork promoting Meta's new Voicebox AI.

Meta’s AI ‘Voicebox’ tool is almost terrifyingly powerful – so terrifying, in fact, that Meta isn’t releasing it to the public (Image credit: Meta)

Change of direction

The clue’s in the name: remember back in 2021, when the then-ubiquitous Facebook underwent a total rebrand to become ‘Meta’? At the time, it was supposed to herald a new era of technology, led by our reptilian overlord Zuckerberg. Enter the metaverse, he said, where your wildest dreams can come true.

Two years down the line, it’s become pretty clear that his Ready Player Zuck fantasies aren’t going to materialize; at least, not for quite a while. AI, on the other hand, really is the new technology frontier – but Meta’s previous obsession with the metaverse has left it on the back foot in the AI gold rush.

Even though Meta has now shifted to AI as its prime area of investment and has maintained an AI research department for years, it’s fair to say that the Facebook owner failed to capitalize on the AI boom late last year. According to Gizmochina, employees have been urging management to shift focus back towards generative AI, which fell by the wayside in favor of the company’s metaverse push.

Meta commentary

A female-presenting person works at her desk in Meta's Horizons VR

Meta’s virtual Horizon workspace was never going to take off, let’s be honest (Image credit: Meta)

Perhaps Meta is simply spread too thin. Back in February, Zuckerberg described 2023 as the company’s “year of efficiency” – a thin cover for Meta’s mass layoffs and project closures back in November 2022, which have seen internal morale fall to an all-time low. Meta is still trying to push ahead in the VR market with products like the Meta Quest Pro, and recently announced it would be releasing a Twitter rival, supposedly called ‘Threads’.

In any case, it seems that Meta might have simply missed the boat. ChatGPT and Microsoft’s Bing AI are already making huge waves in the wider public sphere, along with the best AI art generators such as Midjourney.

It’s hard to see where Meta’s AI projects will fit in the current lineup; perhaps Zuckerberg should just stick to social media instead. Or maybe we'll see Meta pull another hasty name-change to become 'MetAI' or something equally ridiculous. The possibilities are endless!

TechRadar – All the latest technology news

Read More