Google Translate can come in handy when you're traveling or communicating with someone who speaks another language, and thanks to a new update, you can now connect with some 614 million more people. Google is adding 110 new languages to its Translate tool using its AI PaLM 2 large language model (LLM), which brings the total of supported languages to nearly 250. This follows the 24 languages added in 2022, including Indigenous languages of the Americas as well as those spoken across Africa and central Asia.

Many of the recently added languages are ones you've probably never heard of, as they're spoken only by small communities or have no native speakers.

Cantonese, which is spoken predominantly in southeastern China, Hong Kong, and Macau as well as communities across the world, may be one of the most recognizable additions with this update. According to Google, the overlap between Cantonese and Mandarin—which was already available—makes it difficult to train LLMs. Punjabi (Shahmukhi), the most spoken language in Pakistan, is also now available.

A quarter of the newly supported languages come from Africa, and include Afar—spoken in Djibouti, Eritrea, and Ethiopia—and Tamazight (Amazigh), a Berber language used across North Africa, as well as NKo, Fon, Kikongo, Luo, Ga, Swati, Venda, and Wolof.

You can also now use Google Translate to communicate in Manx, a Celtic language from the Isle of Man in the Irish Sea. The last native speaker of Manx died in 1974, leading to its near-extinction, but thanks to revitalization efforts, there are now a few dozen first-language speakers, and a couple thousand who speak Manx as a second language.

The update also includes Portuguese (Portugal), Tongan, Tibetan, Tahitian, Venetian, Sicilian, Fijian, and Jamaican Patois.

The Google Translate app is available in the Apple App Store and the Google Play store. It can translate text you paste in, as well as text appearing in photos. It can also translate voice input as well as handwritten characters.

When OpenAI first rolled out the ChatGPT app for Mac, it was exclusive to ChatGPT Plus subscribers. Unless you paid $20 per month, you needed to stick to the web app or the one on your smartphone. As of Tuesday, however, the Mac app is now free for everyone. And, honestly, you should probably give it a go.

At first glance, OpenAI's Mac app offers the usual ChatGPT experience you're used to. When you log in, you'll find all your previous conversations saved to the sidebar, just as they are in the web and mobile apps. You can type your prompts in the text field, use the mic button to ask questions with your voice, and click the headphones icon to enter Voice mode. (Not the "Her" Voice mode, mind you: That feature has been delayed.) You can also use features like Temporary Chats (conversations that don't pull from your chat history), change your GPT model, generate images with DALL-E, and access GPTs.

A better experience than the web app

But there are some Mac-specific features that make this particular app worth using over the web option. First, in addition to uploading files and photos to ChatGPT, you can take a screenshot of any open window on your Mac directly from the app. If you click on the paperclip icon, and select Take Screenshot, you can select an active window from the pop-up list to share with ChatGPT. (The first time you do this, you'll need to grant the ChatGPT app access to screen recording.)

Alternatively, you can take a screenshot of the window manually, then share it to ChatGPT as an image, but this skips a step and makes the bot feel a bit more integrated with macOS.

Credit: Jake Peterson

But what's even more convenient, in my opinion, is the ChatGPT "launcher." This launcher is essentially Spotlight search, but for ChatGPT. Using a keyboard shortcut, you can bring up a ChatGPT text field directly over any window you're currently using on macOS to start a conversation with the bot. You'll then be taken to the app to continue chatting. This basically saves you the step of switching out of the current app you're in and starting a new thread in ChatGPT; if you see something on your Mac you want to know more about, you can hit Option + Spacebar, type your query, and get started.

Credit: Jake Peterson

This launcher also has the same paperclip icon as the app itself, which means you can upload files and take screenshots directly from the shortcut. If you're a ChatGPT power user, this launcher should be a welcome feature. (I don't even use ChatGPT that much, and I really like it.)

Unfortunately, OpenAI is only making the ChatGPT app available on M-series Macs—the machines running Apple silicon. If you have an older Intel-based Mac, you'll still have to head to the web app in order to use ChatGPT on your computer.

If you have a Mac with an M1 chip or newer, you can download the app from OpenAI's download site.

If you or your company pay for Workspace, you may have noticed Google's AI integration with apps like Docs, Sheets, and Drive. The company has been pushing Gemini in its products since their big rebrand from "Bard" back in February, and it appears that train isn't stopping anytime soon: Starting this week, you'll now have access to Gemini via a sidebar panel in some of Google's most-used Workspace apps.

Google announced the change in a blog post on Monday, stating that Gemini's new side panel would be available in Docs, Sheets, Slides, Drive, and Gmail—the latter of which the company announced in a separate post. The side panel sits to the right of the window, and can be called up at any time from the blue Gemini button when working in these apps.

Google says the side panel uses Gemini 1.5 Pro, the LLM the company rolled out back in February, equipped with a "longer context window and more advanced reasoning." That longer context window should be helpful when asking Gemini to analyze long documents or run through large sets of data in Drive, as it allows an LLM to handle more information at once in any given request.

Now, if you've ever used a generative AI experience—especially one from Google—this experience probably won't shock you: You'll see a pretty typical welcome screen when Gemini comes up, in addition to a series of prompt suggestions for you to ask the bot. When you pull up the side panel in a Google Doc, for example, Gemini may immediately offer you a summary of the doc, then present potential prompts, such as "Refine," "Suggest improvements," or "Rephrase." However, the prompt field at the bottom of the panel is always available for you to ask Gemini whatever you want.

Here are some of the uses Google envisions for Gemini in the side panel:

• Docs: Help you write, summarize text, generate writing ideas, come up with content from other Google files

• Slides: Create new slides, create images for slides, summarize existing presentations

• Sheets: Follow and organize your data, create tables, run formulas, ask for help with tasks in the app

• Drive: Summarize "one or two documents," ask for the highlights about a project, request a detailed report based on multiple files

• Gmail: Summarize a thread, suggest replies to an email, advice on writing an email, ask about emails in your inbox or Drive

Credit: Google

None of these features are necessarily groundbreaking (Gemini has been generally available in Workspace since February) but Google's view is they're now available in a convenient location as you use these apps. In fact, Google announced that Gmail for Android and iOS are also getting Gemini—just not as a side panel. But while the company is convinced that adding its generative AI to its apps will have a positive impact on the end user, I'm not quite sold. After all, this is the first big AI development from Google since the company's catastrophic "AI Overviews" rollout. I, for one, am curious if Gemini will suggest that I respond to an email by sharing instructions on adding glue to pizza.

As companies like Google continue to add new AI features to their products, we're seeing the weak points in real time: Do you want to trust Gemini's summary of a presentation in Slides, or an important conversation in Gmail, when AI still makes things up and treats them like fact?

Who can try Gemini side panel in Google apps

That said, not everyone will actually see Gemini in their Workspace apps, even as Google rolls it out. As of now, Gemini's new side panel feature is only available to companies who purchase the Business and Enterprise Gemini add-on, schools that purchase the Education and Education Premium Gemini add-on, and Google One AI Premium subscribers. If you don't pay for Google's top tier subscription, and your business or school doesn't pay for Gemini, you're not seeing Google's AI in Gmail. Depending on who you are, that may be a good or bad thing.

Google introduced AI Overviews in search results shortly after Google I/O in May, but it wasn’t first to the AI search game. It had already given Gemini the ability to search the internet, and Meta and other competing AI companies had done similarly with their own models. One of the biggest players in this field was Perplexity, which markets itself as a “conversational search engine”—basically another chatbot with internet access, but with even more of a focus on summaries and current events. Unfortunately, Perplexity is now finding itself in hot water after breaking rules and, like Google, returning wrong answer after wrong answer.

On June 11, Forbes published an article accusing Perplexity of stealing its content for quickly rewriting original articles without sourcing, and passing them off as its own. The AI company went as fair as to adapt Forbes’ reporting to podcast form. Shortly after, Wired ran an exposé on Perplexity, accusing it of “bullshitting” and breaking a widely held internet rule (more on that shortly). Now, we’re learning a lot more about what kind of recent data an AI might be able to train on going forward, and why AIs often make so many mistakes when trying to sum up current events.

Perplexity is accused of breaking a longstanding internet rule

Bots aren’t anything new on the internet. Before AI scraped websites for training material, search engines scraped websites to determine where to place them in search results. This led to a standard called the Robots Exclusion Protocol, which allows developers to lay out which parts of their site they don’t want bots to access. Perplexity says it follows this rule, but, spurred on by the Forbes story and an accusation of rule breaking from developer Robb Knight, Wired conducted its own investigation. What it discovered wasn't flattering to Perplexity.

“Wired provided the Perplexity chatbot with the headlines of dozens of articles published on our website this year, as well as prompts about the subjects of Wired reporting,” Wired’s article reads. According to the investigation, the bot then returned answers “closely paraphrasing Wired stories,” complete with original Wired art. Further, it would summarize stories “inaccurately and with minimal attribution.”

Examples include the chatbot inaccurately accusing a police officer of stealing bicycles, and, in a test, responding to a request to summarize a webpage containing a single sentence with a wholly invented story about a young girl going on a fairy tale adventure. Wired concluded Perplexity’s summaries were the result of the AI flagrantly breaking the Robots Exclusion Protocol, and that its inaccuracies likely stemmed from an attempt to sidestep said rule.

According to both Knight and Wired, when users ask Perplexity questions that would require the bot to summarize an article protected by the Robots Exclusion Protocol, a specific IP address running what is assumed to be an automated web browser would access the websites bots are not supposed to scrape. The IP address couldn’t be tracked back to Perplexity with complete certainty, but its frequent association with the service raised suspicions.

In other cases, Wired recognized traces of its metadata in Perplexity’s responses, which could mean the bot may not be reading articles themselves, but accessing traces of it left in URLs and search engines. These wouldn’t be protected by the Robots Exclusion Protocol, but are so light on information that they’re more likely to lead to AI hallucinations—hence the problem with misinformation in AI search results.

Both of these issues presage a battle for the future of AI in search engines, from both ethical and technical standpoints. Even as artists and other creators argue over AI’s right to scrape older works, accessing writing that is just a few days old puts Perplexity at further legal risk.

Perplexity CEO Aravind Srinivas issued a statement to Wired that said “the questions from Wired reflect a deep and fundamental misunderstanding of how Perplexity and the Internet work.” At the same time, Forbes this week reportedly sent Perplexity a letter threatening legal action over “willful infringement” of its copyrights.

It’s only been a few months since Anthropic debuted Claude 3, but the company is ready to take the next step—at least for one of its models. Enter Claude 3.5 Sonnet. As the middle-ground for Anthropic’s large language models (LLMs), Claude Sonnet is a good option for those who want access to a powerful but affordable AI chatbot, and with Claude 3.5 Sonnet, the company says it's making its middle offering even better.

According to the announcement, Claude 3.5 Sonnet is up to two times faster at processing than Claude 3 Opus, previously the most powerful model the company offered (Opus will be getting an update to take back its top spot). Anthropic claims that Claude 3.5 is “ideal for complex tasks,” and that it shows improvement in writing with nuance, humor, and following complex instructions. Claude 3.5 reportedly solved 64 percent of the problems it was given, outperforming the 38 percent record previously set by Claude 3 Opus. That is, indeed, a marked improvement.

Credit: Anthropic

You can see Anthropic's full list of how Claude 3.5 Sonnet compares to other LLMs across different areas in the image above. Based on the data shown, it appears to outperform OpenAI’s newest model, GPT-4o, in almost every category. However, exactly how well those benchmarks will play out in real-world usage remains to be seen.

Coinciding with the launch of Claude 3.5 Sonnet is Artifacts, a new feature that essentially creates a separate window in your Claude window that can showcase your documents, code, and other AI-generated content in a visual space in real time. Anthropic says that this will make collaborating through Claude much easier for teams. Eventually, it hopes to allow entire organizations to use Claude to securely centralize its knowledge in one shared space and then access it through the chatbot. This will likely be similar to what Google has been doing with its Gemini AI offerings in Google Workspace.

Anthropic isn’t stopping here, either. The AI company says it plans to release updated versions of the Opus and Haiku LLMs later this year. The company also noted that it is exploring features like Memory, which would allow Claude to remember a user’s specific preferences and interaction history to help make their experiences even more personal. ChatGPT already utilizes a memory system, so it isn’t surprising to see Anthropic leaning that way with Claude, too.

If you’re interested in trying Claude 3.5 Sonnet for yourself, you can access it directly from the Claude website or through the Claude iOS app. Claude Pro and Team subscribers will still get access to higher rates with the new model. Developers can also utilize Anthropic’s API directly, though they’ll need to pay for tokens to do so.

Microsoft's AI chatbot, Copilot, has been steadily growing and adding new features since its introduction last year. (At that time, Microsoft called it Bing Chat.) As with all things AI, it can be difficult to keep up with the updates, changes, and new features, but Microsoft is adding them to Copilot at a steady clip. Here are some of the best features and changes Microsoft has made to Copilot this year.

Copilot has an app now

If you're still using the Copilot web app, feel free to keep doing so. However, since the beginning of this year, Microsoft has offered Copilot as a dedicated mobile app as well. You can choose to use the experience signed in or signed out, but signing into your Microsoft account gives you access to more features (including bypassing the very strict prompt limit).

Everyone can use Copilot in Microsoft 365 (if you pay)

One of Copilot's flagship features is its integration with Microsoft 365. Microsoft turned the bot into an AI Clippy, adding AI assistant options to apps like Word, Excel, PowerPoint, and OneNote. However, Copilot in 365 was only available to business users—the rest of us that use these apps outside of work were out of luck.

That changed early this year, when Microsoft rolled out Copilot support in Microsoft 365 to all Copilot Pro users. As long as you subscribe to the plan for $20 per month, you can try out Copilot in this suite of apps. While it's a pricey subscription, if you're interested in Copilot, it might be worth the price, since Microsoft is adding most of Copilot's new features to Microsoft 365 apps.

You can use Copilot in Outlook

Previously, if you wanted to use Copilot in Outlook, you needed to head to the web app or go the long way through Microsoft Teams. Since last month, however, Microsoft has offered Copilot support in the Outlook app itself. That makes it easier to use some of the new Copilot features in Outlook, like email draft coaching, choosing the tone of a draft (e.g. neutral, casual, formal).

Reference files when prompting Copilot

Since last month, you've been able to pull in files from your device, SharePoint, and OneDrive when prompting Copilot. If you want the bot to summarize a Word doc, or to have the context of a Powerpoint presentation when responding to your prompt, just type a / when prompting to pull up the file locator.

New options in Word with Copilot

Personally, if there's one app that could benefit most from Copilot, I feel it's Word. Generative AI's main strength in my opinion is text-based, so having an assistant to help you manage your word processing could be a big help.

This year, Microsoft has given Copilot in Word a boost. Here are some of the highlights:

• Use Rewrite on specific sections of a document.

• Highlight a portion of text to summarize and share.

• Create tables from your text.

• Make new tables based on the format of previous tables in your doc.

• Confidential docs are labeled as confidential when referencing them in new docs.

New features for Copilot in Excel

Microsoft has been adding new Copilot features to Excel, as well. Since the beginning of this year, here's what you've been able to do:

• Request a chart of your data.

• Ask Copilot follow-up questions, including requesting clarifications to previous responses.

• Generate formula column options with one prompt.

• Use Copilot to figure out why you're running into issues with a task.

Copilot in OneNote

OneNote actually has had quite a few new Copilot features since January. If you have access to Copilot in OneNote and frequently use the app, here's what you can expect:

• Create notes from audio recordings and transcriptions, then ask Copilot to summarize the notes and arrange them in different ways.

• Create to-do lists with Copilot.

• Copilot can search through information within your organization for added context to your requests.

• Ask Copilot to organize your notes for you.

Copilot for Teams got an upgrade

If you use Copilot in Teams, you may notice now that the bot can now automatically take notes during meetings. If you head to Recap the meeting, you can get a summary of what your team or the call just talked about.

You may also see a new Copilot option attached to the top of your Teams chats. This lets you quickly prompt Copilot inside chats, pulling in documents with the / key. You'll also see that Teams will alert you when AI is being used in a meeting, such as when Copilot is in use without transcriptions.

Let the AI do the prompting for you

Soon, Copilot will start autocompleting your prompts for you. When you start typing, the bot will offer suggestions for what it thinks you might want to do. If you say "Summarize," before you can say what you want summarized, Copilot will guess what you want to round up, including things like your "last 10 emails."

This post is part of Lifehacker’s “Living With AI” series: We investigate the current state of AI, walk through how it can be useful (and how it can’t), and evaluate where this revolutionary tech is heading next. Read more here.

AI-generated art isn’t a concept: It’s here. Thanks to numerous tools with simple and approachable interfaces, anyone can hop on their computer can start generating whatever image ideas pop into their minds. However, as more people have started experimenting with these tools, serious ethical and legal issues have cropped up, and just about everyone online seems to have an opinion on this divisive technology.

As part of our series on living with AI, we put together this guide to demystify how AI art tools work, explain the controversies around them, and show how they impact everyone, from professional artists to curious casuals.

Where and how to make AI art

Before we get too far in the weeds on the tech and ethics of AI art, let’s quickly overview the tools themselves.

There are many AI art generators out there, but the major players would be Midjourney, Stable Diffusion, Copilot, DALL-E 3, and Craiyon. All of these tools are accessible via the web or desktop, and some also have mobile apps.

Midjourney is one of the most powerful options, but it requires a subscription starting at $10 per month. Midjourney also requires a Discord account, since it operates entirely through a dedicate Discord chat server. (Although this is changing.) That means you’re working alongside other users, and all your images are posted publicly on Midjourney’s web gallery, unless you pay $60 per month for the “stealth mode” feature included with the Pro plan.

DALL-E 3 is another powerful option, and is easy to use, since OpenAI now bundles it with ChatGPT Plus. However, you don't need to pay $20 per month to use it: Copilot has free access to DALL-E 3, so as long as you have a Microsoft account, you have DALL-E.

Stable Diffusion is also free and let you make as many images as you want, but image generation takes longer, especially if the servers are busy. Craiyon is also free, but subject to longer generation times, and image quality is lower.

In terms of approachability, DALL-E 3 via Copilot by far the best option if you’re just curious about these tools. It’s free and you can access it from Copilot's web app, Microsoft Edge, or from the Copilot mobile app.

That said, despite the differences in quality and and interface, these tools all work the same way: You type a prompt into a text box describing the image you want to see, press enter, then wait a few moments for the AI to generate the picture based on your description.

The quality of the final product will depend on which tool you’re using and how detailed your prompt is. Some tools, like Midjourney and Stable Diffusion, have guides on coming up with better prompts, and even extra features that can help the AI get closer to your intended result. But even with these extra steps, the process only takes a few moments, and every tool is simple to start using.

How do these tools learn how to draw? And why are there too many fingers?

The images you get from these tools can be impressive, but it’s not because the software actually knows how to draw.

As I pointed out earlier this year, calling these products “AI” is a misnomer. Unlike the common conception of artificial intelligence as seen in science fiction media, these tools are not alive, sentient, or aware in any way, and they do not reason or learn. This is true of both text-based chatbots and generative art tools. In the simplest terms, they work like your phone’s predictive text, pulling from a list of possible solutions to your prompt. When it comes to generative art tools specifically, the tool simply searches for images that match the keywords or descriptions in your prompt, then mashes the elements together.

This is entirely different from the actual process of drawing—in fact, the AI never “draws” anything at all, which is why these tools are notorious for “not knowing how to draw hands.”

As video game designer Doc Burford explains, “If I tell a machine ‘show me Nic Cage dressed as Superman,’ the machine may have images tagged with ‘Nic Cage,’ and it may have images tagged ‘Superman,’ but where the thinking mind of an actual intelligence will put those ideas together and fill in the blank spots with things it knows—like an artist who has also memorized human anatomy—the AI is still gonna give me an imperfect S-Shield on Superman’s chest, it’s gonna mess up the fingers.”

Ethical and legal concerns of AI art

These tools are easy to use and can often spit back compelling results—discounting the occasional extra digits and wonky faces—but there are major ethical concerns around making and distributing AI-generated art that go beyond quality and accuracy.

The main issue with generative AI art tools is they’re built on the backs of uncredited, unpaid artists whose art is used without consent. Every image you generate only exists because of the artists it’s copying from, even if those works aren’t copyrighted. Some AI evangelists like to claim these tools work “just like the human brain” and that “human artists are inspired by or reference other artists the same way,” but this is untrue for multiple reasons.

First, there is no being or mind in an AI, and therefore no memory, no intention, and no skill. Stable Diffusion isn’t “learning” how to draw or taking inspiration from another piece of art: It’s just an algorithm that searches and auto-fills data in the ways it’s programmed to. Humans, on the other hand, think, feel, and act with intention. Their works come from their memorized skill and lived experience. Even using another person’s art as a reference or inspiration is a deliberate choice informed by the artist’s goals.

To help explain the distinction, I reached out to Nicholas Kole, an illustrator and character designer who works with major film and video game studios like Disney, Activision, and DreamWorks. “The work I do, as a concept artist and illustrator, begins with digging deeply into the context of each project,” he says. “I ask pointed questions, tease out ideas about worldbuilding, story, gameplay [and] the process from start to finish is extremely specific, bespoke, and tailored to the precise needs of my colleagues and clients. Every single cufflink, belt buckle, prop, and motif—the art we make is the careful work of thoughtful design, done lovingly and with attention to detail.”

“The intrusion of an algorithm into that process that doesn’t care for context, that doesn’t know whether people have five or 17 fingers, that mashes up visual guesswork based on stolen data, and functions essentially like a million monkeys at a million typewriters is anathema to me.”

Kole says AI art “flies in the face of everything I stand for creatively, and everything I’ve wanted to do with my life’s work. It’s an insult to the reason I make and engage with art—I want to see thoughtful human craft that is expressive, and express myself with thoughtful human craft.” A quick glance through portfolio sites like Art Station shows Kole is not alone in those sentiments, with many professional artists taking a hard stand against AI art.

This hardline stance isn’t just for ideological or aesthetic reasons, either. AI automatization poses a threat to job security for many industries. The threat to working artists is just as real.

AI art also poses a risk to the companies that employ these artists. There have already been major legal battles over AI art infringing on copyrighted materials, and the system is starting to favor original artists. As such, some companies outright ban the use of AI art and reject any applications from artists with AI-generated works in their portfolios to avoid any copyright issues.

Are there ethical uses of AI art?

Despite the ethical and legal issues, some argue there is a place for these tools, and that they can even be helpful to professional artists. In an interview with Kotaku, visual artist RJ Palmers says artists could use AI to “come up with loose compositions, color patterns, lighting, etc.,” for example, and that the tools “can all be very cool for getting inspiration.”

Similarly, author and animator Scott Sullivan argues in his blog that AI is helpful for ideation and iteration while brainstorming, and that “it’s all about the artist’s intention and how they use the tool.”

AI art generators also aren't strictly a "prompt to image" creator, either. Some options, like Microsoft Designer, have multiple other purposes as well. You can use the AI photo editing tools to remove a subject from a photo, or swap out the background entirely; you can create messaging stickers from AI to plop relevant images into conversations; you can even create social media posts from prompts, in case you don't know how to use designer tools yourself. So these tools can have a purpose above simply "generating art."

But while AI art is contentious among professional artists, casual users may wonder whether any of this matters for the layperson or hobbyist that just wants to play with them once in a while. And, sure, AI tools could be used as toys, but it’s important to note that’s not how the creators of these products treat them.

Almost all AI art generators are commercial products in some way. Some are paid products, while free services may earn money through ad revenue. Some are also used as “proof of concept” examples to entice commercial clients to pay for the more powerful version of the tool.

In all cases, the people that make these tools earn money off the work of the artists whose work is used to create the image you’re generating, even if it’s just for fun. As Kole explains, “The generative system can’t function without the stolen life's work of countless passionate people just like me. They each brought their lived experiences, opinions, fixations, and points of view to their bodies of work, only now to have them smashed together inexpertly and touted as original art.” Even if you don’t share or sell images you make, many of these tools keep a public record of all generated content that other users can download and distribute.

Given all these concerns, it’s hard to recommend AI art generators, even if the intent to use them is innocent. Nevertheless, these tools are here, and unless some future regulations force them to change, we can’t stop folks from giving them a try. But, if you do, please keep in mind the legal and ethical issues associated with making and sharing AI art, think twice about sharing it, and never claim an AI-generated image as your own work.

This post is part of Lifehacker’s “Living With AI” series: We investigate the current state of AI, walk through how it can be useful (and how it can’t), and evaluate where this revolutionary tech is heading next. Read more here.

Almost as soon as ChatGPT launched in late 2022, the world started talking about how and when to use it. Is it ethical to use generative AI at work? Is that “cheating?” Or are we simply witnessing the next big technological innovation, one that everyone will either have to embrace, or fall behind dragging their feet?

AI is now a part of work, whether you like it or not

AI, like anything else, is a tool first and foremost, and tools help us get more done than we can on our own. (My job would literally not be possible without my computer.) In that regard, there’s nothing wrong, in theory, with using AI to be more productive. In fact, some work apps have fully embraced the AI bandwagon. Just look at Microsoft: The company basically conquered the meaning of “computing at work,” and it's adding AI functionality directly into its products.

Since last year, the entire Microsoft 365 suite—including Word, PowerPoint, Excel, Teams, and more—has adopted “Copilot,” the company’s AI assist tool. Think of it like Clippy from back in the day, only now way more useful. In Teams, you can ask the bot to summarize your meeting notes; in Word, you can ask the AI to draft a work proposal based on your bullet list, then request it tighten up specific paragraphs you aren’t thrilled with; in Excel, you can ask Copilot to analyze and model your data; in PowerPoint, you can ask for an entire slideshow to be created for you based on a prompt.

These tools don’t just exist: They’re being actively created by the companies that make our work products, and their use is encouraged. It reminds me of how Microsoft advertised Excel itself back in 1990: The ad presents spreadsheets as time consuming, rigid, and featureless, but with Excel, you can create a working presentation in an elevator ride. We don’t see that as “cheating” work: This is work.

Intelligently relying on AI is the same thing: Just as 1990's Excel extrapolates data into cells you didn’t create yourself, 2023's Excel will answer questions you have about your data, and will execute commands you give it in normal language, rather than formulas and functions. It’s a tool.

What work shouldn’t you use AI for?

Of course, there’s still an ethical line you can cross here. Tools can be used to make work better, but they can also be used to cheat. If you use the internet to hire someone else to do your job, then pass that work off as your own, that’s not using the tool to do your work better. That’s wrong. If you simply ask Copilot or ChatGPT to do your job for you in its entirety, same deal.

You also have to consider your own company’s guidelines when it comes to AI and the use of outside technology. It’s possible your organization has already established these rules, given AI’s prominence over the past year and a half or so: Maybe your company is giving you the green light to use AI tools within reason. If so, great! But if your company decides you can’t use AI for any purpose as far as work in concerned, you might want to log out of ChatGPT during business hours.

But, let’s be real: Your company probably isn’t going to know whether or not you use AI tools if you’re using them responsibly. The bigger issue here is privacy and confidentiality, and it’s something not enough people think about when using AI in general.

In brief, generative AI tools work because they are trained on huge sets of data. But AI is far from perfect, and the more data the system has to work with, the more it can improve. You train AI systems with every prompt you give them, unless the service allows you to specifically opt out of this training. When you ask Copilot for help writing an email, it takes in the entire exchange, from how you reacted to its responses, to the contents of the email itself.

As such, it’s a good rule of thumb to never give confidential or sensitive information to AI. An easy way to avoid trouble is to treat AI like you would you work email: Only share information with something like ChatGPT you’d be comfortable emailing a colleague. After all, your emails could very well be made public someday: Would you be OK with the world seeing what you said? If so, you should be fine sharing with AI. If not, keep it away from the robots.

If the service offers you the choice, opt out of this training. By doing so, your interactions with the AI will not be used to improve the service, and your previous chats will likely be deleted from the servers after a set period of time. Even so, always refrain from sharing private or corporate data with an AI chatbot: If the developer keeps more data than we realize, and they're ever hacked, you could put your work data in a precarious place.

This post is part of Lifehacker’s “Living With AI” series: We investigate the current state of AI, walk through how it can be useful (and how it can’t), and evaluate where this revolutionary tech is heading next. Read more here.

There’s a lot of talk about how AI is going to change your life. But unless you know how to code and are deeply aware of the latest advancements in AI tech, you likely assume you have no part to play here. (I know I did.) But as it turns out, there are companies out there designing programs to help you build AI tools without needing a lick of code.

What is the no-code movement?

The idea behind “no-code” is simple: Everyone should have the accessibility to build programs, tools, and other digital services regardless of their level of coding experience. While some take a “low-code” approach, which still requires some coding knowledge, the services on this list are strictly “no-code.” Specifically, they’re no-code solutions to building AI tools.

You don’t need to be a computer scientist to build your own AI tools. You don’t even need to know how to code. You can train a neural network to identify a specific type of plant, or build a simple chatbot to help customers solve issues on your website.

That being said, keep your expectations in check here: The best AI tools are going to require extensive knowledge of both computer science and coding. But it’s good to know there are utilities out there ready to help you build practical AI tools from scratch, without needing to know much about coding (or tech) in the first place.

Train simple machine-learning models for free with Lobe

If training a machine learning model sounds like something reserved for the AI experts, think again. While it’s true that machine learning is a complicated practice, there’s a way to build you own model for free with as few tools as a laptop and a webcam.

That’s thanks to a program called Lobe: The free app, owned by Microsoft, makes it easy to build your own machine learning model to recognize whatever you want. Need your app to differentiate between colors? You can train it to do that. Want to make a program that can identify different types of plants? Train away.

You can see from the example video that you can train a model to identify when someone is drinking from a cup in only a few minutes. While you can include any images you may have previously taken, you can also simply snap some photos of you drinking from a cup from your webcam. Once you take enough sample photos of you drinking and not drinking, you can use those photos to train the model.

You can then test the model to see how well (or not) it can predict if you’re drinking from a cup. In this example, it does a great job whenever it sees the cup in hand, but it incorrectly identifies holding a hand to your face as drinking as well. You can use feedback buttons to tell the model when it gets something wrong, so it can quickly retrain itself based on this information and hopefully make more accurate predictions going forward.

Google also has a similar tool for training simple machine-learning models called Teachable Machine, if you’d like to compare its offering to Microsoft’s.

Build your own AI chatbot with Juji Studio

AI chatbots are all the rage lately. ChatGPT, of course, kicked off the modern AI craze because of its accessible yet powerful chat features, but everything from Facebook Messenger to healthcare sites have used chatbots for years. While OpenAI built ChatGPT with years of expertise, you can make your own chatbot without typing a single line of code.

Juji Studio wants to make building a light version of ChatGPT, in the company’s words, as easy as making PowerPoint slides. The program gives you the tools to build a working chatbot you can implement into your site or Facebook Messenger. That includes controlling the flow of the chatbot, adjusting its personality, and feeding it a Q&A list so it can accurately answer specific questions users might have.

Juji lets you start with a blank canvas, or base your chatbot on one of its existing templates. Templates include customer service bots, job interview bots, teaching assistant bots, and bots that can issue user experience surveys. No matter what you choose, you’ll see the “brains” of your bot in a column on the left side of the screen.

It really does resemble PowerPoint slides: Each “slide” corresponds to a different task for the chatbot to follow. For example, with the customer service chatbot, you have an “invite user questions until done” slide, which is pre-programmed to listen to user questions until the user gives a “done” signal. You can go in and customize the prompts the chatbot will ask the user, such as asking for an account number or email address, or even more personal questions, like asking about a bad experience the user had, or the best part of their day.

You can, of course, customize the entire experience to your needs. You can build a bot that changes its approach based on whether or not the user responds positively or negatively to an opinion-based question:

Build custom versions of Copilot or ChatGPT

Chatbots like Copilot and ChatGPT can be useful for a variety of tasks, but when you want to use AI for a specific function, you'll want to turn to GPTs. GPTs, not to be confused with OpenAI's GPT AI models, are custom chatbots that can be built to serve virtually any purpose. Best of all, there's no coding necessary. Instead, you simply tell the bot what you want, and the service walks you through the process to set up your GPT.

You can build a GPT that helps the user learn a language, plans a meal and teaches you how to make it, or generates logos for different purposes. Really, whatever you want your chatbot to do, you can build a GPT to accomplish it. (Or, at least create a chatbot that's more focused on your task than ChatGPT or Copilot in general.)

You can access Copilot GPTs if you subscribe to Copilot Pro. OpenAI used to lock its GPTs behind a subscription, but the company is making them free for all users. Plus, OpenAI lets users put their custom-built GPTs on the GPT Store. If you don't want to make your own, you can browse other users' creations and try them out for yourself.

Create anything you want with Bubble

For the ultimate no-code experience, you’ll want to use a tool like Bubble. You use an interface similar to something like Photoshop to build your app or service, dragging and dropping new UI elements and functions as necessary.

But while Bubble is a no-brainer for us code-illiterates to build things, it’s also integrated with AI. There are tons of AI applications you can include in your programs using Bubble: You can connect your builds to OpenAI products like GPT and DALL-E, while at the same time taking advantage of plugins make by other Bubble members. All of these tools allow you to build a useful AI program by yourself—something that uses the power of GPT without needing to know how it works in the first place.

One of the best ways to get started here is by taking advantage of OpenAI Playground. Playground is similar to ChatGPT, in that it’s based on OpenAI’s large language models, but it isn’t a chatbot. As such, you can use Playground to create different kinds of products and functions that you can then easily move to a Bubble project using the “View Code” button.

This post is part of Lifehacker’s “Living With AI” series: We investigate the current state of AI, walk through how it can be useful (and how it can’t), and evaluate where this revolutionary tech is heading next. Read more here.

Artificial intelligence (AI) is the latest tech revolution. Just as the cryptocurrency boom introduced the world to a whole bunch of new jargon, the AI hype train has brought with it a set of terms that are frequently used, but not always explained. If you’re wondering about the difference between a chatbot and a LLM, or between deep learning and machine learning, you’re in the right place: Here is a glossary of 20 AI-related terms, along with newbie-friendly explanations of what it all means.

Artificial intelligence (AI)

In simple terms, AI is intelligence in computers or machines, especially that which mimics human intelligence. AI is a broad term that covers many different types of machine intelligence, but the discourse around AI right now mostly centers around tools that create art, content, and summarize or transcribe content. To call these tools “intelligent” is up for debate, but AI is the term that has stuck.

Algorithm

An algorithm is a set of instructions that a program follows to give you a result. Common examples of algorithms include search engines, which show you a set of results based on your queries, or social media apps, which show content based on your interests. Algorithms allow AI tools to create predictive models, or create content or art based on your inputs.

Bias

In the context of AI, bias refers to erroneous results produced because the algorithm makes incorrect assumptions or lacks sufficient data. For example, speech recognition tools may not be able to understand certain English accents correctly because the tools were trained only with an American accent.

Conversational AI

AI tools that you can talk to, such as chatbots or voice assistants, are called conversational AI. If you're asking the assistant something yourself, it's conversational AI.

Data mining

The process of combing through large sets of data to find patterns or trends. Some AI tools use data mining to help you understand what makes people buy more items in a store or on a website, or how to optimize a business to cater to increased demand during peak hours.

Deep learning

Deep learning attempts to recreate the way the human brain learns, by utilizing three or more neural network “layers” to process large volumes of data and learn by example. These layers each process their own view of the given data and come together to reach a final conclusion.

Software for self-driving cars uses deep learning to identify stop signs, lane markers, and traffic lights, through object recognition: This is achieved by showing the AI tool many examples of what a certain object looks like (e.g., a stop sign), and through repeated training, the AI tool will eventually be able to identify that object with as close to 100% accuracy as possible.

Large language model (LLM)

A large language model (LLM) is a deep-learning algorithm that is trained on a massive data set to generate, translate, and process text. LLMs (like OpenAI’s GPT-4) allow AI tools to understand your queries and to generate text inputs based on them. LLMs also power AI tools that can identify the important parts of text or videos and summarize them for you.

Generative AI

Generative AI can generate art, images, text, or other results from your inputs, which are often powered by an LLM. It has become the catch-all term for the current AI tech many companies are now adding to their products. For example, a generative AI model can generate an image with a few text prompts, or turn a vertical photo into a wide-screen wallpaper.

Hallucination

When AI presents fiction as fact, we call that hallucinating. Hallucinations can happen when an AI’s data set isn’t accurate or its training is flawed, so it outputs an answer it’s confident on based on its available knowledge. That said, because AI is based on a complex web of networks, we don’t necessarily understand each example of hallucination. Lifehacker writer Stephen Johnson has great advice for spotting AI hallucinations.

Image recognition

The ability to identify specific subjects in an image. Computer programs can use image recognition to spot flowers in an image and name them, or to identify different species of birds in a photo.

Machine learning

When algorithms can improve themselves by learning from experience or data, it’s referred to as machine learning. Machine learning is the general practice that other AI terms we’ve discuss stem from: Deep learning is a form of machine learning, and large language models are trained through machine learning.

Natural language processing

When a program can understand inputs written in human languages, it falls under natural language processing. It’s how your calendar app understands what to do when you write, “I have a meeting at 8 p.m. at the coffee shop on Fifth Avenue tomorrow,” or when you ask Siri, “What’s the weather like today?”

Neural networks

The human brain has layers upon layers of neurons constantly processing information and learning from it. An AI neural network mimics this structure of neurons to learn from data sets. A neural network is the system that allows for machine learning and deep learning, and, at the end, allows machines to perform complex tasks such as image recognition and text generation.

Optical character recognition (OCR)

The process of extracting text from images is done via OCR. Programs that support OCR can identify handwritten or typed text, and let you copy and paste it as well.

Prompt engineering

A prompt is any series of words that you use to get a response from a program, such as generative AI. In the context of AI, prompt engineering is the art of writing prompts to get chatbots to give you the most useful responses. It’s also a field where people are hired to come up with creative prompts to test AI tools and identify its limits and weaknesses.

Reinforcement learning from human feedback (RLHF)

RLHF is the process of training AI with feedback from people. When the AI delivers incorrect results, a human shows it what the correct response should be. This allows the AI to deliver accurate and useful results a lot faster than it would otherwise.

Speech recognition

A program’s ability to understand human speech. Speech recognition can be used for conversational AI to understand your queries and deliver responses, or for speech-to-text tools to understand spoken words and convert them to text.

Token

When you feed a text query into an AI tool, it breaks down this text into tokens, common sequences of characters in text, which are then processed by the AI program. If you use a GPT model, for example, the pricing is based on the number of tokens it processes: You can calculate this number using the company’s tokenizer tool, which also shows you how words are broken down into tokens. OpenAI says one token is roughly four characters of text.

Training data

A training set or training data is the information that an algorithm or machine learning tool uses to learn and execute its function. For example, large language models may use training data by scraping some of the world’s most popular websites to pick up text, queries, and human expressions.

Turing Test

Alan Turing was the British mathematician known as the “father of theoretical computer science and artificial intelligence.” His Turing Test (or “The Imitation Game”) is designed to identify if a computer’s intelligence is identical to that of a human. A computer is said to have passed the Turing Test when a human is tricked into thinking the machine’s responses were written by a human.

This post is part of Lifehacker’s “Living With AI” series: We investigate the current state of AI, walk through how it can be useful (and how it can’t), and evaluate where this revolutionary tech is heading next. Read more here.

You wouldn’t be blamed for thinking AI really kicked off in the past couple years. But AI has been a long time in the making, including most of the 20th century. It's difficult to pick up a phone or laptop today without seeing some type of AI feature, but that's only because of working going back nearly one hundred years.

AI’s conceptual beginnings

Of course, people have been wondering if we could make machines that think for as long as we’ve had machines. The modern concept came from Alan Turing, a renowned mathematician well known for his work in deciphering Nazi Germany’s “unbreakable” code produced by their Enigma machine during World War II. As the New York Times highlights, Turing essentially predicted what the computer could—and would—become, imagining it as “one machine for all possible tasks.”

But it was what Turing wrote in “Computing Machinery and Intelligence” that changed things forever: The computer scientist posed the question, “Can machines think?” but also argued this framing was the wrong approach to take. Instead, he proposed a thought-experiment called “The Imitation Game.” Imagine you have three people: a man (A), a woman (B), and an interrogator, separated into three rooms. The interrogator’s goal is to determine which player is the man and which is the woman using only text-based communication. If both players were truthful in their answers, it’s not such a difficult task. But if one or both decides to lie, it becomes much more challenging.

But the point of the Imitation Game isn’t to test a human’s deduction ability. Rather, Turing asks you to imagine a machine taking the place of player A or B. Could the machine effectively trick the interrogator into thinking it was human?

Kick-starting the idea of neural networks

Turing was the most influential spark for the concept of AI, but it was Frank Rosenblatt who actually kick-started the technology’s practice, even if he never saw it come to fruition. Rosenblatt created the “Perceptron,” a computer modeled after how neurons work in the brain, with the ability to teach itself new skills. The computer has a single layer neural network, and it works like this: You have the machine make a prediction about something—say, whether a punch card is marked on the left or the right. If the computer is wrong, it adjusts to be more accurate. Over thousands or even millions of attempts, it “learns” the right answers instead of having to predict them.

That design is based on neurons: You have an input, such as a piece of information you want the computer to recognize. The neuron takes the data and, based on its previous knowledge, produces a corresponding output. If that output is wrong, you tell the computer, and adjust the “weight” of the neuron to produce an outcome you hope is closer to the desired output. Over time, you find the right weight, and the computer will have successfully “learned.”

Unfortunately, despite some promising attempts, the Perceptron simply couldn’t follow through on Rosenblatt’s theories and claims, and interest in both it and the practice of artificial intelligence dried up. As we know today, however, Rosenblatt wasn’t wrong: His machine was just too simple. The perceptron’s neural network had only one layer, which isn’t enough to enable machine learning on any meaningful level.

Many layers makes machine learning work

That’s what Geoffrey Hinton discovered in the 1980s: Where Turing posited the idea, and Rosenblatt created the first machines, Hinton pushed AI into its current iteration by theorizing that nature had cracked neural network-based AI already in the human brain. He and other researchers, like Yann LeCun and Yoshua Bengio, proved that neural networks built upon multiple layers and a huge number of connections can enable machine learning.

Through the 1990s and 2000s, researchers would slowly prove neural networks’ potential. LeCun, for example, created a neural net that could recognize handwritten characters. But it was still slow going: While the theories were right on the money, computers weren’t powerful enough to handle the amount of data necessary to see AI’s full potential. Moore’s Law finds a way, of course, and around 2012, both hardware and data sets had advanced to the point that machine learning took off: Suddenly, researchers could train neural nets to do things they never could before, and we started to see AI in action in everything from smart assistants to self-driving cars.

And then, in late 2022, ChatGPT blew up, showing both professionals, enthusiasts, and the general public what AI could really do, and we’ve been on a wild ride ever since. We don’t know what the future of AI actually has in store: All we can do is look at how far the tech has come, what we can do with it now, and imagine where we go from here.

Living with AI

To that end, take a look through our collection of articles all about living with AI. We define AI terms you need to know, walk you through building AI tools without needing to know how to code, talk about how to use AI responsibly for work, and discuss the ethics of generating AI art.

It’s been a tough few weeks for Microsoft’s headlining Copilot+ feature, and it hasn't even launched yet. After being called out for security concerns before being made opt-in by default, Recall is now being outright delayed.

In a blog post on the Windows website on Thursday, Windows+ Devices corporate vice president Pavan Davuliri wrote that Recall will no longer launch with Copilot+ AI laptops on June 18th, and is instead being relegated to a Windows Insider preview “in the coming weeks.”

“We are adjusting the release model for Recall to leverage the expertise of the Windows Insider Community to ensure the experience meets our high standards for quality and security,” Davuluri explained.

The AI feature was plagued by security concerns

That’s a big blow for Microsoft, as Recall was supposed to be the star feature for its big push into AI laptops. The idea was for it to act like a sort of rewind button for your PC, taking constant screenshots and allowing you to search through previous activity to get caught up on anything you did in the past, from reviewing your browsing habits to tracking down old school notes. But the feature also raised concerns over who has access to that data.

Davuliri explains in his post that screenshots are stored locally and that Recall does not send snapshots to Microsoft. He also says that snapshots have “per-user encryption” that keeps administrators and others logged into the same device from viewing them.

At the same time, security researchers have been able to uncover and extract the text file that a pre-release version of Recall uses for storage, which they claimed was unencrypted. This puts things like passwords and financial information at risk of being stolen by hackers, or even just a nosy roommate.

Davuliri wasn’t clear about when exactly Windows Insiders would get their hands on Recall, but thanked the community for giving a “clear signal” that Microsoft needed to do more. Specifically, he attributed the choice to disable Recall by default and to enforce Windows Hello (which requires either biometric identification or a PIN) for Recall before users can access it.

Generously, limiting access to the Windows Insider program, which anyone can join for free, gives Microsoft more time to collect and weigh this kind of feedback. But it also takes the wind out of Copilot+’s sails just a week before launch, leaving the base experience nearly identical to current versions of Windows (outside of a few creative apps).

It also puts Qualcomm, which will be providing the chips for Microsoft’s first Copilot+ PCs, on a more even playing field with AMD and Intel, which won’t get Copilot+ features until later this year.

Despite explaining away issues with its AI Overviews while promising to make them better, Google is still apparently telling people to put glue in their pizza. And in fact, articles like this are only making the situation worse.

When they launched to everyone in the U.S. shortly after Google I/O, AI Overviews immediately became the laughing stock of search, telling people to eat rocks, use butt plugs while squatting, and, perhaps most famously, to add glue to their homemade pizza.

Most of these offending answers were quickly scrubbed from the web, and Google issued a somewhat defensive apology. Unfortunately, if you use the right phrasing, you can reportedly still get these blatantly incorrect "answers" to pop up.

In a post on June 11, Bluesky user Colin McMillen said he was still able to get AI Overviews to tell him to add “1/8 cup, or 2 tablespoons, of white, nontoxic glue to pizza sauce” when asking “how much glue to add to pizza.”

The question seems purposefully designed to mess with AI Overviews, sure—although given the recent discourse, a well-meaning person who’s not so terminally online might legitimately be curious what all the hubbub is about. At any rate, Google did promise to address even leading questions like these (as it probably doesn’t want its AI to appear to be endorsing anything that could make people sick), and it clearly hasn’t.

Perhaps more frustrating is the fact that Google’s AI Overview sourced the recent pizza claim to Katie Notopoulus of Business Insider, who most certainly did not tell people to put glue in their pizza. Rather, Notopoulus was reporting on AI Overview’s initial mistake; Google’s AI just decided to attribute that mistake to her because of it. 

“Google’s AI is eating itself already,” McMillen said, in response to the situation.

I wasn’t able to reproduce the response myself, but The Verge did, though with different wording: The AI Overview still cited Business Insider, but rightly attributed the initial advice to to Google’s own AI. Which means Google AI’s source for its ongoing hallucination is...itself.

What’s likely going on here is that Google stopped its AI from using sarcastic Reddit posts as sources, but it’s now turning to news articles reporting on its mistakes to fill in the gaps. In other words, as Google messes up, and as people report on it, Google will then use that reporting to back its initial claims. The Verge compared it to Google bombing, an old tactic where people would link the words “miserable failure” to a photo of George W. Bush so often that Google images would return a photo of the president when you searched for the phrase.

Google is likely to fix this latest AI hiccup soon, but it’s all bit of a “laying the train tracks as you go situation,” and certainly not likely to do anything to improve AI search's reputation.

Anyway, just in case Google attaches my name to a future AI Overview as a source, I want to make it clear: Do not put glue in your pizza (and leave out the pineapple while you’re at it).