NYT Connections bot uses AI to highlight each day’s top mistakes

Connections, the daily word game from The New York Times, isn’t exactly easy. In fact, the solve rate dips below 50 percent some days. To help highlight where players are going wrong, a new bot is employing AI to guess the thought process they’re using for the most common mistakes every day.

For the uninitiated, Connections is a word/logic game that the paper of record debuted last year. Every day, you're presented with a grid of 16 words that you have to split into four categories. There’s only one solution and after four mistakes, the game is over. However, there’s some trickiness afoot. There are often red herrings galore and frequently at least five viable answers for a group.

I’m hooked and I’ve played every day since last July. It’s one of our favorite games of 2023 overall (though it’s perhaps not entirely original).

After you win or lose each day’s game, you can saunter over to the Connections Bot. As with the bot for Wordle, you’ll see how well you did compared with other players and receive a skill score out of 99. This is primarily based on how few mistakes you make, but you’ll get extra credit for solving the more difficult purple and blue categories first.

After you see the skill score and other details (such as whether a red herring caught you out), the AI feature comes into play. This will highlight the most common incorrect guesses from that day. It will also try to guess a description for the group that players had in mind. So, for a failed guess of gutter, bowl, alley and lane, the bot might believe you were looking for a list of bowling-related terms. This is a real example from a recent game in which I made that exact mistake. Alley and lane were actually types of streets.

NYT Connections mistakes
New York Times

Your own failed guesses might not show up in the bot, though. That’s because there are around 2.6 million different ways to group each grid together. Bear in mind that while you don’t need a Times account to play Connections, you’ll need to be logged into one to use the bot and track your scores.

One other interesting thing about the bot is that it marks the first time that the Times’ newsroom will regularly publish AI-generated English text. Before publication, the paper’s editors will review all AI-generated responses and may edit them for style and clarity. That said, the Times notes that “there’s no way to use math or even artificial intelligence to reliably solve the game,” so you can’t really use the likes of ChatGPT to cheat at it quite yet.

This article originally appeared on Engadget at https://www.engadget.com/gaming/nyt-connections-bot-uses-ai-to-highlight-each-days-top-mistakes-145242525.html?src=rss

Sonos made a public Trello board to track its broken app fixes

The Sonos app is broken, and it's been causing not just customers, but also dealers and installers so many problems over the past few months. Now, it's documenting the app issues it has already fixed in a public Trello board that you can view and browse yourself. The company's Social & Community manager, Keith Nieves, has announced the Trello board on Reddit, describing it as a way for people to keep updated about its "progress on the path to a normal core app." While this is the first time Sonos has taken this approach, this isn't a new idea for the tech industry: Fortnite maintains a Trello board packed with information on bugs plaguing different aspects of the game.

Issues with the Sonos app started popping up after the company pushed a major app redesign back in April. The refreshed app was supposed to make it easier to play different kinds of content and was supposed to introduce more new features. Instead, users discovered that the application was lacking basic features like the ability to set alarms. It was also riddled with performance and reliability issues, such as those that prevented them from rearranging speakers, kept speakers from working altogether and made it difficult for them to add new devices to their home setup.

Sonos CEO Patrick Spence acknowledged that people were experiencing "significant problems" with the company's new app in a post published in July. He said that Sonos had released updates that improved the app's performance since May and laid out a timeline for other fixes, including improving alarm consistency and restoring edit mode for playlists, which are rolling out in the next couple of months. 

Nieves is maintaining the Trello board by himself for now, and so far, it contains release notes from previous fixes, as well as a list of patches coming out soon. He said that the cards in the board were sourced from various posts and release notes communications, and he admitted that the board is "not an exhaustive list of every item on [Sonos'] internal roadmap, nor is it a complete view of all known issues." He seemed to be open to collaboration with users, though, and is asking users to PM him if they are experiencing issues not noted on a card. 

This article originally appeared on Engadget at https://www.engadget.com/apps/sonos-made-a-public-trello-board-to-track-its-broken-app-fixes-140013331.html?src=rss

Gemini will soon generate AI images of people again with the upgraded Imagen 3

Google’s generative AI tools are getting some of the boosts the company previewed at Google I/O. Starting this week, the company is rolling out the next-gen version of its Imagen image generator, which reintroduces the ability to generate AI people (after an embarrassing controversy earlier this year). Google’s Gemini chatbot also adds Gems, the company’s take on bots with custom instructions, similar to ChatGPT’s custom GPTs.

Google’s Imagen 3 is the upgraded version of its image generator, coming to Gemini. The company says the next-gen AI model “sets a new standard for image quality” and is built with guardrails to avoid overcorrecting for diversity, like the bizarre historical AI images that went viral early this year.

“Across a wide range of benchmarks, Imagen 3 performs favorably compared to other image generation models available,” Gemini Product Manager Dave Citron wrote in a press release. The tool allows you to guide the image generation with additional prompts if you don’t like what it spits out the first time.

Citron says Imagen 3 performs “favorably” compared to the competition. It also includes Google’s SynthID tool to watermark images, making it clear that they’re AI-made and not the genuine article.

AI images created with Google's Imagen 3 model. Foxes and balloons.
Google

Citron says the ability to generate people will return in the coming days for paid users, months after Google yanked the feature. He says new guardrails will prevent the generation of “photorealistic, identifiable individuals” — a far cry from the problematic deepfakes generated by Elon Musk’s Grok. Also off-limits are children and (as with other image generators) any gory, violent or sexual scenes. The product manager grounds expectations by saying Gemini’s images won’t be perfect, but he promises the company will continue to listen to user feedback and refine accordingly.

Starting this week, the Imagen 3 model will be available for all users, but reintroducing images featuring people will begin with paid users. English-speaking Gemini Advanced, Business and Enterprise users can expect human image generation to return “over the coming days.”

A Google AI Gem, custom bot, designed to curate cliffghangers.
Google

Initially previewed at Google I/O 2024, Gems are Google’s custom chatbots with user-created instructions. It’s essentially Gemini’s answer to OpenAI’s GPTs, which Google’s competitor rolled out late last year. Gems begin rolling out in the next few days.

“With Gems, you can create a team of experts to help you think through a challenging project, brainstorm ideas for an upcoming event, or write the perfect caption for a social media post,” Citron wrote. “Your Gem can also remember a detailed set of instructions to help you save time on tedious, repetitive or difficult tasks.”

In addition to the blank slate of custom Gems, Gemini will include premade ones “to help you get started” and inspire new ideas. Prebuilt Gems include:

  • Learning coach - to help you understand complex topics

  • Brainstormer - to inspire new ideas

  • Career guide - walk you through skill upgrades, decisions and goals

  • Writing editor - provide constructive feedback on grammar, tone and structure

  • Coding partner - upgrade coding skills for developers and inspire new projects

Gems begin rolling out today on desktop and mobile. However, they’re only available for Gemini Advanced, Business and Enterprise subscribers, so you’ll need a paid plan to check them out.

This article originally appeared on Engadget at https://www.engadget.com/ai/gemini-will-soon-generate-ai-images-of-people-again-with-the-upgraded-imagen-3-161429310.html?src=rss

TikTok is finally rolling out group chats for up to 32 people

TikTok is rolling out some direct message updates, including "highly requested" group chats. Owner ByteDance notes that group chats aren't available everywhere just yet, but those who do have access to the feature will be able to chat in groups of up to 32 people.

You can start a group chat in a couple of ways. From your inbox, you can tap the Chat button at the top of the screen or a name in the messages list and then the "More options..." button. You can then choose which friends you want to include and then tap "Start group chat."

The other method to start a group chat is by sharing a post with a bunch of people. When you have a video you'd like to chat about with more than one person, tap the Share button, then select "Create group chat." Pick the friends you want to send it to, add a message if you like, then start chatting with your buds about the video.

You can accept any group chat invites you receive via your inbox. You can only invite people you mutually follow to a group chat. As with regular DMs, group chats are not available to users aged between 13 and 15. As for 16- and 17-year-olds, they'll only be able to join a group chat if they have at least one mutual friend in there. If they create a group chat, they'll have to review and manually approve anyone new who joins.

TikTok is also bringing stickers to DMs. That gives folks another way to engage with each other visually. You can create and upload custom stickers for anyone to use.

This article originally appeared on Engadget at https://www.engadget.com/social-media/tiktok-is-finally-rolling-out-group-chats-for-up-to-32-people-180056720.html?src=rss

Taskmaster VR doesn’t quite deliver on the promise of the TV show

I’m a huge Taskmaster fan. The UK TV show pits five comedians (or funny-adjacent celebrities) against ridiculous tasks that are then arbitrarily scored by the titular taskmaster.

Given that the tasks involve lateral thinking and physical movement, on paper, the TV show was ripe for a VR spin-off. Think Job Simulator with added jeopardy, in-jokes and British whimsy. Unfortunately, Taskmaster VR doesn’t deliver on the fun and charm of the show. Two points.

After being teased (and available to wishlist) earlier this year, the game launched last week across Steam, Meta Quest 2 and 3, and Pro. I played the game briefly on Quest 2, the lowest-powered option, before shifting to the sharper screens of Quest 3.

You’ll want to play it on a headset like the Quest 3 or better. The basic Quest 2 makes the whole thing very grainy. But that’s not the biggest issue. The problem is it’s not a great VR game. Collision detection is rough, and while it’s not the first VR title guilty of this, the fine motor control demanded of the game’s tasks makes things a pain.

For example, the first game involves you trying to make a sandwich, matching Alex’s arcane sandwich request, plating it up, and putting it on the plinth, ready for judgment. However, you have to do all of that with a pair of spatulas. Opening a fridge with a spatula in real life is challenging enough; in a virtual world, it sometimes felt more like luck than dexterity whether I got that slice of bacon into my sandwich.

In another game, I was to launch an egg as far as I could but land it in a frying pan. I struggled to keep a frying pan in the same place while constructing a poorly-considered ramp of pipes, boxes and wheelie bins. The in-game physics are so simplistic that most object surfaces behave the same. And my pan slid to the floor. Again and again and again. You can crouch in Taskmaster VR – when you drop the pan again – but the physics engine conveys a degree of momentum when you press the button to stand upright again, adding a ‘bounce’ to anything in your hand.

Many of the games are multi-stage and expansive, which I appreciated. Circling back to the sandwich task, some ingredients need to be cooked – and you’ll have to turn on the stove burners to do so. Once you’ve made your sandwich, Greg, the Taskmaster will scrutinize the result. One slice of bread? He’ll go on a tirade about open-face sandwiches. Wrong plate? Your score will be dinged again. Something inedible? He’ll notice.

There are also easter egg references to the show everywhere, from the tutorial onwards. Look out of windows; even the bushes and garden are filled with items and references only a Taskmaster fan would get. Do Americans think all Brits (I’m a Brit) are this whimsical? I hope so.

In another game, you’ll have to rush around the entire Taskmaster estate (a tiny home and garden) searching for dog food, but only after you’ve escaped the garage. The tasks are considered, but I wish there were more. The TV show is famous for comedians’ unusual lateral thinking and solutions that the task makers didn’t plan for. I haven’t seen those possibilities – at least not yet.

Why not mix up the codes for the safe featured in a watermelon smashing task? This would be especially useful if you’re playing with other Taskmaster fans who want to try the tasks themselves. A little bit of random number generation would work wonders. Having said that, there’s a Taskmaster VR Creative mode, where you’ll be able to craft your own levels and tasks, which could add longevity if it finds an audience.

The banter between Taskmaster and his assistant is tonally accurate, benefitting from using the voices of Greg Davies and (Little) Alex Horne, but their in-game models don’t convey the weird, awkward charm of the duo. Maybe I’m just a superfan, and my expectations were too high, but there’s something about the sound quality (with lines recorded in a booth) that pulls you out of the illusion of being on a version of the TV show.

I’m unsure if I’ll ever replay the entire game again. Fortunately, once unlocked, you can choose to play each task standalone, which is the perfect way for me to share the game with visiting friends.

If the team behind Taskmaster VR can offer some new tasks through DLC (how about replicating a task from the next TV series?), this might be more of a draw. Their time starts now.

This article originally appeared on Engadget at https://www.engadget.com/gaming/taskmaster-vr-doesnt-quite-deliver-on-the-promise-of-the-tv-show-130005486.html?src=rss

Apple is fighting Tencent and ByteDance over in-app payments in China

Apple has been pressuring ByteDance and Tencent to close loopholes in China that funnel customers to external payment systems for making in-app purchases, according to a report by Bloomberg. Anyone living in the West knows the drill here. Apple wants that 30 percent commission.

Reporting indicates this pressure campaign began in May. Apple allegedly warned Tencent that it would reject crucial WeChat updates if it didn’t eliminate the ability for users to make payments outside of Apple’s ecosystem. Tencent complied with the original request, issuing an update in July, but Apple went one step further.

WeChat is home to thousands of third-party mini-games and experiences. Apple asked Tencent to disable in-game chat between creators and players, as that's another theoretical loophole that could funnel users to third-party payment systems. Tencent has yet to agree to this request.

Back in June, Apple reportedly did something similar with ByteDance. It threatened to withhold updates of Douyin, which is basically the Chinese version of TikTok, unless it plugged any gaps that steer users away from that much-coveted 30 percent commission. According to Bloomberg, ByteDance has yet to issue a formal response.

These are aggressive moves on the part of Apple. China is the world’s largest smartphone market, sure, but the iPhone isn’t the dominant brand throughout the country. As a matter of fact, the phone failed to crack the top five in sales last quarter and the company recently experienced a 6.5 percent decline in profits.

It’s also worth noting that both ByteDance and Tencent aren’t happy corporate warriors looking out for the little guy. These massive companies levy their own commissions on creators and likely didn’t want Apple cutting into their bottom lines.

An Apple spokesperson was unusually blunt in a statement to Bloomberg, simply saying that company guidelines dictate that the sale of all digital goods must go through its system and that the review team has the power to reject app submissions that violate that policy. Neither Tencent or ByteDance issued a comment to Bloomberg

China, like the rest of the world, has been cracking down on walled gardens like Apple’s App Store. Despite the country’s hesitance to continue allowing closed ecosystems controlled by a single entity, Apple CEO Tim Cook is bullish about its prospects in China. “We continue to be confident in the long-term opportunity in China,” he said during a recent earnings call. “I don’t know how every chapter of the book reads, but we’re very confident in the long term.”

Apple is facing numerous legal hurdles all over the world regarding its Hungry Hungry Hippos approach to gobbling up commission fees. The European Commission issued a ruling that dictates it must allow app developers to steer users to payment systems and offers outside of the App Store. The company also faces potential fines from the EU, to the tune of ten percent of global annual revenue. As for the US, Epic sued Apple over its developer transaction fee policy and many other companies have expressed their own concerns. It's also worth nothing that Tencent owns a 40 percent stake in Epic Games. 

This article originally appeared on Engadget at https://www.engadget.com/apple-is-fighting-tencent-and-bytedance-over-in-app-payments-in-china-155949462.html?src=rss

Meta explains why its AI claimed Trump’s assassination attempt didn’t happen

Meta has explained why its AI chatbot didn't want to respond to inquiries about the assassination attempt on Trump and then, in some cases, denied that the event took place. The company said it programmed Meta AI to not answer questions about an event right after it happens, because there's typically "an enormous amount of confusion, conflicting information, or outright conspiracy theories in the public domain." As for why Meta AI eventually started asserting that the attempt didn't happen "in a small number of cases," it was apparently due to hallucinations. 

An AI "hallucinates" when it generates false or misleading responses to questions that require factual replies due to various factors like inaccurate training data and AI models struggling to parse multiple sources of information. Meta says it has updated its AI's responses and admits that it should have done so sooner. It's still working to address its hallucination issue, though, so its chatbot could still be telling people that there was no attempt on the former president's life. 

In addition, Meta has also explained why its social media platforms had been incorrectly applying the fact check label to the photo of Trump with his fist in the air taken right after the assassination attempt. A doctored version of that image made it look like his Secret Service agents were smiling, and the company applied a fact check label to it. Because the original and doctored photos were almost identical, Meta's systems applied the label to the real image, as well. The company has since corrected the mistake. 

Trump's supporters have been crying foul over Meta AI's actions and have been accusing the company of suppressing the story. Google had to issue a response of its own after Elon Musk claimed that the company's search engine imposed a "search ban" on the former president. Musk shared an image that showed Google's autocomplete suggesting "president donald duck" when someone types in "president donald." Google explained that it was due to a bug affecting its autocomplete feature and said that users can search for whatever they want anytime. 

This article originally appeared on Engadget at https://www.engadget.com/meta-explains-why-its-ai-claimed-trumps-assassination-attempt-didnt-happen-120002196.html?src=rss

Overwatch 2 may test a return to six-player teams

Blizzard has provided an update that will be music to the ears of many Overwatch 2 fans. The publisher plans to experiment with a range of team compositions beyond the role-locked 5v5 format that the game currently has. That includes a potential revival of six-player teams from the original Overwatch.

"The community has, juuuust once or twice, suggested a test," Overwatch 2 game director Aaron Keller wrote in a blog post. "Why not put various forms of 6v6 in the game in order to gauge the results? We agree, and based on your feedback, we’re exploring how we can test different forms of 6v6 in the game to gauge the results."

Keller is quick to point out that it may take some time before the team runs a 6v6 trial. Overwatch 2 was designed from the ground up for five-player teams, with hero balancing and map designs (and redesigns) taking into account the reduced damage mitigation from having one fewer tank on each side.

There are technical considerations here too, as features such as visual upgrades, more technically demanding hero kits and the ability to see outlines of allies through walls impact Overwatch 2's performance. As such, shifting from 10-player to 12-player lobbies could have a negative impact on performance, particularly on older or less-powerful systems.

"Overwatch is a fast-paced game, and maintaining a game that runs smoothly across all our platforms is important for the player’s experience," Keller wrote. "While a limited-time test could arrive sooner, the team is still investigating exactly how long it would take to permanently increase performance across the game. This would be a large effort that would most likely take at least several seasons to accomplish."

Sombra in Overwatch 2
Blizzard Entertainment

The possibility of lengthier queue times is also a concern if 6v6 tests are a success and the format sticks around in some way for the long haul. Keller's blog post is a lengthy read, but well worth taking the time to check out if you're interested in how we ended up at 5v5 with players locked to specific roles in the core modes. One of the reasons why Blizzard ditched a tank from each team with the launch of Overwatch 2 two years ago was that it was the least popular of the three roles, with damage in first place and support in between. Having one fewer tank has helped to reduce queue times across the board.

Keller notes that his team does have some ideas about how to tackle potentially longer queue times if 6v6 returns for the long haul, but it wasn't a problem that was solved in the past and it's not a sure thing that the fresh strategies will work. "Is there a world where people are willing to live with long queue times to play this format? Maybe, but that’s a pretty risky move to make," Keller wrote. "We also have tens of millions of new players that have only played 5v5. We want to be openly mindful not to frustrate those who like the game as it is today."

There are other reasons why Blizzard changed to a 5v5 format including that it wasn't necessarily fun to play against a team that used two shield-based tanks. Chokepoints could feel oppressive with two tanks blocking the path. Dropping one of them and placing more of an emphasis on things like flanking, individual impact and a little thing called first-person shooting (with more opportunities to actually shoot at opponents instead of barriers) was designed to make gameplay more fluid and fun. Team fights with multiple tanks on each side also were often a war of attrition until ultimate abilities became available, especially during the infamous GOATS meta.

Still, many fans spent years playing the game in groups of six, and having to lose one of those from the gang with the shift to 5v5 sucked for a lot of people. Plus, the synergy of two tanks could make for really fun moments — the Earthshatter/D.Va Bomb combo was always a personal favorite. Given that Blizzard has always prided itself on taking player feedback into account and fostering the concept of building Overwatch 2 together with the fans, it's great to see that the team is willing to at least test how 6v6 could work in the modern landscape.

However, 6v6 isn't the only alternate approach to the current team composition that Blizzard plans to try. "Besides running experiments with 6v6, we’d like to run some that re-examine the ways we tried solving previous problems, specifically with the goal of bringing some of the freedom back to an Overwatch match without the severity of issues that accompanied it," Keller wrote. "For instance, we think there could be other ways of putting a team together that aren’t quite as rigid as a set composition, but not as loose as Open Queue." Blizzard will try at least one version of this idea in an experimental Quick Play mode during Season 13, which will take place this fall.

Any permanent changes to the current format, whether or not that means a shift back to 6v6, won't happen overnight. Blizzard would run any 6v6 tests for a few weeks at a time to gauge feedback from players, and then see how that format impacts the game beyond core gameplay, such as in arcade modes. 

"We’d take the lessons from this playtest to see what we can learn about the mode within the current game ecosystem and for the future of Overwatch," Keller wrote. "We would reflect carefully on the learnings from whatever test we run and explore how to best give players what's being asked for. Whether that’s a world of 5v5, 6v6 or even both, is for future us to figure out."

This article originally appeared on Engadget at https://www.engadget.com/overwatch-2-may-test-a-return-to-six-player-teams-180021753.html?src=rss

Google gives free Gemini users access to its faster, lighter 1.5 Flash AI model

Google is making its Gemini AI faster and more efficient across the board. You now have access to 1.5 Flash, its generative AI model designed to be able to generate responses more quickly and efficiently, even if you're not paying for Gemini Advanced. The company says you'll notice improvements in latency, as well as the tool's reasoning and image understanding, on both the web and mobile. 

In addition, it's expanding the AI assistant's context window, so that you can have longer conversations with it and ask it more complex questions. In the near future, Google will also give you the ability to upload files to Gemini from Google Drive or from your device. If you give it access to your notes, for instance, it will be able to create a study guide or a practice exam for you. Plus, the assistant will be able to analyze data and make it easier to digest with graphics and charts. 

As part of its work to reduce hallucinations, Google is now displaying links to related content if you ask it questions that require factual answers. It will display a "gray chip" at the end of a paragraph in its response that links to websites where you can read more about the topic. Those chips could even lead to your emails, if you've linked Gemini to your Gmail account. The feature is currently limited to select locations for English prompts only. 

The company is making Gemini more accessible overall, as well. It has started gradually rolling out Gemini in Google Messages for Android devices in the European Economic Area (EU, Iceland, Liechtenstein and Norway), the UK and Switzerland. You'll now also be able to chat with Gemini in French, Polish and Spanish within the Messages app. Finally, Google is expanding access to Gemini's mobile app to more regions and is giving more teenagers the ability to use the AI tool. As long as you meet its minimum age requirement of 13, you'll be able to chat with the assistant. Google has even introduced a teen-specific onboarding process and an AI literacy guide, so you can get an idea on how to use the tool to accomplish your tasks. 

This article originally appeared on Engadget at https://www.engadget.com/google-gives-free-gemini-users-access-to-its-faster-lighter-15-flash-ai-model-160006705.html?src=rss

Microsoft is adding AI-powered summaries to Bing search results

The race to bring more AI features to search is escalating, with Microsoft moving forward with additional tools for Bing. Today, the company began previews for Bing generative search, where the top result for a user's query will be an original response compiled by AI.

The blog post about Bing generative search showed a few sample results. In addition to the overview statement, Microsoft will provide links to the main sources that the large-language models and small-language models used to create their answer. It will also have a section of related information. For instance, with the sample query "how long do elephants live," the main summary is followed by videos with information about factors that impact elephant longevity. After the generative results, the usual list of search hits will be displayed.

Preview of Bing generative search results
Microsoft

If you're curious and want to try out generative search right away, you may be out of luck. This isn't an opt-in choice yet for Bing users and it is only being applied to "a small percentage of user queries" at this stage. "We are slowly rolling this out and will take our time, garner feedback, test and learn, and work to create a great experience before making this more broadly available," Microsoft said.

Google rolled out a similar tool earlier this year called AI Overview in an effort to retain users who might go directly to an AI chatbot for answers to their questions. AI Overview had a rough, gluey pizza start, so it will be interesting to see how well Microsoft's counterpart performs in comparison. And while Microsoft did emphasize in the blog post that it developed generative search without the intent of ruining web publishers' business, it's worth repeating that this kind of AI tool is not a replacement for actual news.

This article originally appeared on Engadget at https://www.engadget.com/microsoft-is-adding-ai-powered-summaries-to-bing-search-results-203053790.html?src=rss