Google has delayed a series of top-secret artificial intelligence (AI) events that were set to showcase the company’s Gemini generative AI tool, according to The Information. If true, it's another blow to Google’s efforts to compete with the likes of ChatGPT in the AI world – although Gemini is still expected to launch soon.
As detailed in a paywalled report from The Information, the Gemini event was due to take place just days from now, with the first one kicking off next week. Yet Google CEO Sundar Pichai has apparently taken the decision to push it back to January 2024, according to “two people with knowledge of the of the decision.”
The reason for the delay? Google was not confident that Gemini was able to “reliably handle some non-English queries,” The Information claims. Google wants to ensure its AI tool works well in a number of languages, and it’s clearly not quite there yet.
The events – due to be held in California, New York and Washington – would apparently have been “Google’s most important product launch of the year” owing to how keen Google is to catch up with OpenAI’s ChatGPT. That means that delaying them could be a sizable blow to Google, even if they do ultimately take place in early 2024.
In ChatGPT’s shadow
Despite their imminent dates, Google hadn’t done much promotion for the events (perhaps due to concerns over Gemini’s abilities). In the end, that meant the company managed to avoid an embarrassing retraction or cancellation of already-announced events.
Yet Google isn’t likely to be very happy with the situation. The delay demonstrates how much the company is struggling to get on level terms with OpenAI, despite its vast wealth and engineering abilities. It’s also the second time in recent weeks that Google has reportedly had to delay its Gemini events.
It comes shortly after OpenAI reportedly had a major breakthrough with its own generative AI efforts. This tool is supposedly able to solve problems it has not been trained on, something that AI has traditionally found difficult, thereby ramping up the pressure on Google to hit back.
Google is likely to weave Gemini’s AI capabilities into its other products, such as Search, Google Assistant, Google Docs and more, The Information believes, so keep an eye out for AI updates if you regularly use the company’s apps. Just don’t expect to see these changes until early next year.
During Microsoft’s Surface Event, the tech giant revealed a ton of new features and updates. One of the most interesting and possibly biggest was the Copilot all-in-one AI assistant, which is coming to Windows 11 on September 26, 2023.
Microsoft Copilot is a more general version of 365 Copilot that uses AI to help users with any Windows 11 programs including popular programs like Paint, Snipping Tool, Photos, and more. And now we have even more information about the standalone app, which just might be the most important update to Windows OS yet.
A new blog post from Yusuf Mehdi, CVP & Consumer Chief Marketing Officer, explains how Copilot works with a multitude of Windows programs. For instance, the Paint Cocreator app (which will be available to Windows Insiders starting September 26) allows users to input a text prompt, select a style, and generate a unique image, which can then be further modified by using the new layers feature for Paint, or by simply drawing on top of the generated image.
Microsoft Clipchamp, meanwhile, is a new companion AI tool for video editing with the use of enhanced AI tools like Auto Compose. By answering a few questions on the type of video you're developing, Clipchamp will provide recommended scenes, edits, and a narrative for you based on your input and creative assets. Then you can upload the video to, for instance, YouTube, TikTok, Instagram, or LinkedIn as well as to OneDrive or Google Drive.
Snipping Tool also received several new upgrades. You can now capture sound using audio and mic support along with visuals, by simply pressing Win + Shift + R or Print Screen to activate. Two new text actions, text extraction and redaction, were also added.
Text extraction is used to scan content from an online article, a video call, or any other source and then copy it text directly into other documents like Word or PowerPoint. Text redaction is a safety feature that completely blacks out any text you don’t want to show, like an email address within your screenshots or a category of information.
Photos app has been upgraded as well with a new feature called Background blur that offers a way to enhance image resolution and apply blur effects. Enhanced search capabilities make it easier to find photos saved in OneDrive by typing in keywords related to objects, locations, or dates. There’s also a new slideshow feature that organizes photos into a presentation to share with family and friends.
Voice also has gotten a boost and now works in more places including during the log-in process. You can now dictate complex and nonstandard words through the new spelling experience, and the corrections functionality will fix words that were recognized incorrectly. Narrator supports even more languages including Spanish, Portuguese, French, German, Mandarin Chinese, Japanese, Korean, and English (UK, India).
The Passkeys tool creates a unique, unguessable credential and allows you to sign in using your face, fingerprint, or device PIN. It will now be integrated into Windows OS and will work on multiple browsers including Edge, Chrome, Firefox, and more.
Adaptive Dimming is another new feature that allows users to conserve energy. If your PC presence sensor detects you are no longer paying attention, it will slowly dim your screen and save energy. It can also serve as an alert to refocus if your attention has been wandering.
John Cable, Microsoft’s, VP, Program Management, Windows Servicing & Delivery, released his own blog post which outlines how to access all these new features and updates. Windows 11 devices will gradually get access to these updates over the next few weeks. Anyone with Windows 11, version 22H2 running on their device can get access to these tools as soon as they're ready by going to Settings > Windows Update and turning on ‘Get the latest updates as soon as they’re available.’
All of these Copilot updates are expected to be broadly available in the November 2023 security update release. The Windows 11 23H2 update, meanwhile will not be dropping next week but in Q4 2023, which is October at the earliest.
Clippy, that helpful paperclip sprite that used to watch your work in Microsoft Word and do its best to help you, was never that smart, or even that deeply integrated with Microsoft's popular computing platform.
Now imagine if Clippy got a brain and body transplant that made it a true genius and then hooked it into the deepest parts of Windows 11. That's Windows Copilot, which was among the big reveals at Microsoft's 2023 Surface Event.
Microsoft showed off a lot of Copilot demos during its packed AI and Surface launch event, but it wasn't until I got an up-close and personal demo of the Copilot preview in action on a Windows 11 system that I truly understood it and the ramifications for the next generation of Windows 11 users.
When the Windows 11 update arrives on September 26, it will bring with it the Copilot preview. Microsoft tells me that it will work with every PC that supports Windows 11.
To be clear, Copilot is not an app. It's marginally a utility. It's more like the voice inside Windows 11 head, a consciousness that is fully aware of everything Windows 11 can do, and much of what you're doing on Windows 11.
Copilot combines all of Microsoft's best AI work to date. It can bring a large language model (LLM) to understand text, and context, and produce fresh text. It integrates Bing Chat to make it conversational (and also supports voice, though I did not see that in my demos).
Two things, though, make Copilot feel like a true part of the Windows 11 experience. The first is, crucially, that Windows 11 copy and paste triggers Copilot, basically waking it up to the possibility of working directly with you.
Image 1 of 7
Image 2 of 7
Image 3 of 7
Image 4 of 7
Image 5 of 7
Image 6 of 7
Image 7 of 7
In the demo I saw, we opened a Word document full of a massive list of things to do in New York City – there's nothing like a sea of gray text to make the eyes glaze over.
Copilot isn't pushy like Clippy. It didn't pop up immediately asking if it could help. Instead, copying the text set it off.
The fact that Copilot can see that you're performing one of the most fundamental Windows 11 tasks, and using that action to help you, is a big deal. Once Copilot sees the clipboard text, it politely asks if you want to use that text to chat. Once we did that, Copilot's chat asked what we wanted to do with it (Revise, Summarize, Expand, Explain). You can be quite specific in your requests, so we asked for distances between Tribeca and our hotel.
Copilot is deep inside Windows, but it's not shut out from the outside. As with Bing Chat, Copilot sources the web for answers. In fact, it synthesizes the best answers and then, yes, provides citations and links for it all.
Copilot tries to be extra helpful by going beyond the initial request. In this instance, it also quickly served up some local attractions.
The other thing that tells you Copilot isn't simply a plugin or add-on is that it has its own invoke and dismiss key combo: You use the Windows button and 'c.'
In my opinion, you don't get the keys unless you're part of Windows and not just a temporary tenant.
Image 1 of 6
Image 2 of 6
Image 3 of 6
Image 4 of 6
Image 5 of 6
Image 6 of 6
Other demos further solidified my belief that this is not your father's Windows 11.
When we dragged a food photo from Outlook into the chat windows, Copilot asked what we wanted to do with it. We requested instructions on how to make the unidentified dish. It took a moment (Copilot preview isn't always that fast) for it to identify it as Shashuka, and then offer detailed instructions on how to cook it.
Copilot further demonstrated its integration by working seamlessly with Windows Snip (which always sends snipped images to the clipboard). We snipped a math problem image and, naturally, when we asked, Copilot helped us solve it.
Image 1 of 5
Image 2 of 5
Image 3 of 5
Image 4 of 5
Image 5 of 5
When we told Copilot we wanted to know how to focus at work, it used its platform integration to help guide us through Windows 11's Focus settings.
Copilot can be used for the most prosaic of Windows tasks, as well. Our demo desktop was getting a little cluttered, so we asked Copilot to “snap my windows.” It quickly organized the desktop and offered advice on how to make adjustments.
Windows Copilot Preview will simply arrive with the Windows 11 update. You won't have to install anything and there's no requirement that you use it. However, based on what I saw, if you ignore Copilot, you may be missing out on an entirely new way of working with the World's most widely used desktop platform.
Windows 11’s Paint app is set for some major upgrades which have been seen coming through in testing right now.
And when we say major changes, we really mean big new avenues of exploration for Microsoft, the most tantalizing of which is the introduction of a layers feature for Paint compositions.
This new addition was revealed in a blog post that notes the Paint app in the Canary and Dev testing channels now has the feature (version 11.2308.18.0 or higher).
Layers mean you can use multiple layers in one image, with different elements placed in different layers. Those layers can be shown or hidden, worked on separately, and indeed merged together if needed, for a more flexible and advanced way of editing any given image.
There’s a Layer button in the Paint app’s toolbar, and you can work with the feature by clicking it (whereby a side panel pops up showing the different layers that you add).
A new transparency effect has been brought in, too, with a checkboard pattern representing parts of an image that are transparent. Paint will also let you open (and save out) transparent PNG image files.
Here’s another big introduction for Paint, too: as spotted by PhantomOfEarth on X (formerly Twitter), that same new version of the app has hidden strings pertaining to an AI-powered image generation feature.
The latest Paint app update in Dev/Canary (11.2308.18.0) includes strings for a new image generation feature:CocreatorDescribe what you’d like to createTry “a cat walking in a wood” or “A house made out of leaves”Choose a styleSeptember 18, 2023
In short, the idea is that Paint will be bolstered with Dall-E image creation chops in a feature apparently called ‘Cocreator’ which allows you to describe an image and have it composed for you (including specifying art styles).
Well, okay, maybe it’s a little bit of a surprise in terms of Paint getting fleshed out considerably more than we ever thought it would, with the introduction of layers certainly coming out of left field.
In fairness, the ability to work with layers is something that has been hacked into Paint in the past, and that did prompt a whole lot of badgering of Microsoft about why it wouldn’t officially introduce the capability.
So, fair play to Microsoft, the software giant has listened and is doing just that – assuming that the layers and transparency elements make it through testing to the release version of Paint (which seems very likely).
Maybe we’ll see other hacks become official Paint features – you may recall that simple animations were also unofficially introduced to Microsoft’s creative app in the past.
Apple might have found a wild solution to VR’s prescription lens problem; liquid lenses.
VR headsets and glasses don’t usually mix well. Often they sit too close to your face for glasses to fit in front of your eyes, and the solutions deployed by headset designers are a mixed bag – some package in optional spacers that make room for specs like the Oculus Quest 2, while others include a prescription lens attachment (but you need to buy lenses for it at an added cost) like the Apple Vision Pro, and a few do nothing at all.
This has resulted in some glasses wearers feeling like VR isn’t accessible to them, but that might change if Apple’s latest patent comes to one of its headsets.
According to the patent granted in the US (via Apple Insider) Apple has created a design for an “electronic device with liquid lenses.” The document describes a “head-mounted device” (sounds like a VR headset) with “tunable liquid lenses.” You can read the patent for the full details, but the TL;DR is that electronic signals sent to the lenses will deform the liquid in them and alter the refractive index of the lenses.
This should allow the liquid lenses to correct a wide range of eyesight issues without the need for any accessories. What’s more, the correction is calibrated by the headset’s eye-tracking system.
Apple’s patent also states that it could apply to a “pair of glasses.” We can’t read too much into patent wordings, but this could hint at the Apple AR glasses that Apple apparently also has in development.
When will we get liquid lenses?
As with all patents we need to note that there’s no guarantee that we’ll ever see these liquid lenses appear in a real headset – one that’s made by Apple or otherwise. While the design exists in theory, it might be less than practical to add liquid lenses to a commercially available headset – either from a design or cost perspective. Alternatively, Apple might develop a different solution to VR’s prescription problem.
What’s more, even if liquid lenses do appear in an Apple headset you or I could pick up off the shelf there’s no telling when that will happen.
It’s probably an impossibility for the first-generation Vision Pro to launch in early 2024, and we’d be surprised if it appeared in the second-generation headset that rumors predict will appear sometime in the next few years. Instead, it seems far more likely we’d see liquid lenses in the third-generation model (assuming we see them at all) in half a decade or so – as this would give Apple plenty of time to hone the design.
But despite being the leader in the AI arms race – and powering Microsoft’s Bing AI – it looks like ChatGPT might be losing momentum. According to SimilarWeb, traffic to OpenAI’s ChatGPT site dropped by almost 10% compared to last month, while metrics from Sensor Tower also demonstrated that downloads of the iOS app are in decline too.
As reported by Insider, paying users of the more powerful GPT-4 model (access to which is included in ChatGPT Plus) have been complaining on social media and OpenAI’s own forums about a dip in output quality from the chatbot.
A common consensus was that GPT-4 was able to generate outputs faster, but at a lower level of quality. Peter Yang, a product lead for Roblox, took to Twitter to decry the bot’s recent work, claiming that “the quality seems worse”. One forum user said the recent GPT-4 experience felt “like driving a Ferrari for a month then suddenly it turns into a beaten up old pickup”.
GPT4’s output has changed recently.It generates faster, but the quality seems worse.Perhaps OpenAI is trying to save costs.Has anyone else noticed this?May 21, 2023
Why is GPT-4 suddenly struggling?
Some users were even harsher, calling the bot “dumber” and “lazier” than before, with a lengthy thread on OpenAI’s forums filled with all manner of complaints. One user, ‘bitbytebit’, described it as “totally horrible now” and “braindead vs. before”.
According to users, there was a point a few weeks ago where GPT-4 became massively faster – but at a cost of performance. The AI community has speculated that this could be due to a shift in OpenAI’s design ethos behind the more powerful machine learning model – namely, breaking it up into multiple smaller models trained in specific areas, which can act in tandem to provide the same end result while being cheaper for OpenAI to run.
OpenAI has yet to officially confirm this is the case, as there has been no mention of such a major change to the way GPT-4 works. It’s a credible explanation according to industry experts like Sharon Zhou, CEO of AI-building company Lamini, who described the multi-model idea as the “natural next step” in developing GPT-4.
AIs eating AIs
However, there’s another pressing problem with ChatGPT that some users suspect could be the cause of the recent drop in performance – an issue that the AI industry seems largely unprepared to tackle.
If you’re not familiar with the term ‘AI cannibalism’, let me break it down in brief: large language models (LLMs) like ChatGPT and Google Bard scrape the public internet for data to be used when generating responses. In recent months, a veritable boom in AI-generated content online – including an unwanted torrent of AI-authored novels on Kindle Unlimited – means that LLMs are increasingly likely to scoop up materials that were already produced by an AI when hunting through the web for information.
This runs the risk of creating a feedback loop, where AI models ‘learn’ from content that was itself AI-generated, resulting in a gradual decline in output coherence and quality. With numerous LLMs now available both to professionals and the wider public, the risk of AI cannibalism is becoming increasingly prevalent – especially since there’s yet to be any meaningful demonstration of how AI models might accurately differentiate between ‘real’ information and AI-generated content.
Discussions around AI have largely focused on the risks it poses to society – for example, Facebook owner Meta recently declined to open up its new speech-generating AI to the public after it was deemed ‘too dangerous’ to be released. But content cannibalization is more of a risk to the future of AI itself; something that threatens to ruin the functionality of tools such as ChatGPT, which depend upon original human-made materials in order to learn and generate content.
Do you use ChatGPT or GPT-4? If you do, have you felt that there’s been a drop in quality recently, or have you simply lost interest in the chatbot? I’d love to hear from you on Twitter. With so many competitors now springing up, is it possible that OpenAI’s dominance might be coming to an end?
Adobe Photoshop has been the gold standard in photo editing for over three decades, so it was only a matter of time until it embraced the tricks seen in thebest AI art generators – and it's now done just that in the form of a new tool called Generative Fill.
The new tool, which lets you extend images or add objects to them using text prompts, certainly isn't the first AI-powered feature we've seen in Photoshop. Generative Fill is also a user-friendly development of existing Adobe tools, like Content Aware Fill, but it's also one of the most significant new Photoshop features we’ve seen for years.
That’s because it leans on the power of Adobe Firefly, the company’s new generative AI engine, to help you fix big compositional mistakes or completely reinvent an image’s contents. In Adobe’s demos, images in portrait orientation are instantly turned into ones in landscape – with Photoshop simply inventing the sides of the photo based on the original image.
While some of those examples are quite subtle, others have a very obvious art aesthetic. For example, a photo of a corgi is turned into one with very obviously fake bubbles and a van in the background.
Adobe clearly sees Generative Fill as a tool for both beginners and pros, but the new text-to-image prompt box is certainly a useful touch for those who don’t know Photoshop’s existing tools. You can use this to add small details to an image or completely change its background – in another demo, a deer is moved from its forest background to a city thanks to the prompt ‘wet alley at night’.
Of course, none of this will be new to fans of Midjourney or Dall-E, which have helped spark this year’s boom in text-to-image generation.
But Adobe is keen to stress that AI tools like Generative Fill model have only been trained on Adobe Stock images, openly-licensed content, and public domain content where the copyright has expired. This means they can be used for commercial use without the threat of class-action lawsuits from artists who claim some AI models have stolen their work.
While Generative Fill is only rolling out to the full Photoshop app in the “second half of 2023”, there are a couple of ways you can try it out now. First, it’s available in Photoshop’s desktop beta app, which you can get by going to the Creative Cloud desktop app, choosing Beta apps in the left sidebar, then installing it from there.
The feature is also available as a module within the web-only Adobe Firefly, which was also recently added to Google Bard. To use Firefly in Bard, you can simply write your image request (for example, 'make an image of a unicorn and a cake at a kid's party') and it'll do the rest. What a time to be alive.
Analysis: Photoshop battles its new AI rivals
Like Google, Adobe is a giant incumbent that's under attack from AI upstarts like OpenAI and Midjourney. While Firefly and Photoshop's Generative Fill aren't doing things we haven't seen before, they are doing them in a measured way that sidesteps any copyright issues and helps maintains its reputation.
Photoshop's embrace of generative AI also brings these tools fully into the mainstream. The image editor may not be the dominant force it was before the likes of Canva, Affinity Photo and GIMP arrived to offer more affordable alternatives, but it remains one of the best photo editors around and certainly one of the most widely used.
From Adobe's early demos, it looks like Generative Fill is in its early days and produces mixed results, depending on your tastes. In some images, the effects are subtle and realistic, while in others – particularly images where large parts of entirely AI-generated – the results are clearly AI-generated and may not date very well.
Still, the arrival of Generative AI alongside other new features like the Remove Tool –another development of the Photoshop's existing ability to let you eliminate unwanted objects – is only a good thing for those who aren't familiar with the app's sometimes arcane interface.
That’s pretty crummy, of course, but now the free version has been upped to include unlimited exporting to Full HD or 1080p resolution, which previously required you to subscribe to the Creator plan (at the price of $ 9 / £8 / AU$ 10 monthly).
There are still benefits for paying a Creator subscription, of course, namely access to stock audio clips and unlimited cloud storage for your projects. Meanwhile, business plans furnish you with extra goodies including stock video clips and the ability to fully brand your creations.
Analysis: A quick change following Windows 11 debut, and a necessary one
It’s good to see 1080p exports coming to the freebie incarnation of Clipchamp, though really, 480p was a very low bar to set, and so this was a move that needed to be made.
It’s obvious enough that Clipchamp being introduced as a Windows 11 app in testing – presumably set to come to the OS later this year in the big 2022 update – brought a lot more attention to the program. And with the spotlight shining more intensely on that export limitation, and perhaps given initial tester feedback, Microsoft decided that the 480p situation had to change (we can but guess).
Clipchamp is, in effect, the return of Windows Movie Maker, giving casual users a built-in and convenient option to quickly edit video clips within Windows 11. However, having to pay for a decent resolution with the end result would’ve severely limited Clipchamp’s usefulness in terms of that positioning.