What actually changed. DALL-E was one of the original ways to generate images in ChatGPT — a separate, named tool you could open from the GPT menu. OpenAI has replaced it with ChatGPT Images, built on its newer gpt-image-1 models, which is now simply part of the regular ChatGPT chat. You no longer pick a separate "DALL-E" option — you just ask ChatGPT for an image in the same conversation you'd use for anything else, and it generates one using the newer technology. This is available on every plan, including the free tier, not just paid subscriptions.

Why this matters. OpenAI has not said what happens to images you previously created and saved through the old DALL-E tool once the deadline passes. To be safe, if you have images stored there that you want to keep, download them to your device rather than assuming they'll stay available in ChatGPT indefinitely.

Read the full article on AI for Regulars

There's no single right length, but most funeral guides converge on a similar range: somewhere around 3 to 7 minutes spoken aloud, which usually works out to roughly 400 to 900 words, depending on how quickly you naturally speak. If you're not a confident public speaker, or you suspect you'll be emotional, aim for the shorter end — a short eulogy delivered steadily leaves a stronger impression than a long one that trails off.

When you ask ChatGPT to draft or shape your eulogy, tell it the target length up front ("about 500 words" or "around 4 minutes read aloud") so you're not cutting a finished draft down afterward.

Read the full article on AI for Regulars

What free accounts get. Free and Go accounts can now create up to 3 active scheduled tasks. Each task can run once, or repeat no more than once a day — so you can set up a morning news summary or a weekly reminder, but not something that checks in every hour. This is the same 3-task limit Go subscribers already had; Free now matches it.

What's still paid-only. Two things did not come to the Free plan. First, event-based triggers — having ChatGPT automatically run a task when something happens in Gmail, Slack, or GitHub — remain exclusive to Plus, Pro, Business, and Enterprise accounts. Free and Go can only schedule tasks by time, not by event. Second, the higher task limits are unchanged: Plus still gets 5 active tasks, and Pro and Enterprise still get 15.

Read the full article on AI for Regulars

None of these on their own means anything is wrong. But if a few of them feel familiar, it's worth a closer look.

You open it for a quick question and lose an hour. You meant to ask one thing and look up later, well past when you meant to stop, having followed the conversation somewhere you didn't plan to go.

Read the full article on AI for Regulars

Who actually sees ads. People on the Free and Go plans. If you pay for Plus, Pro, Business, Enterprise, or Edu, you won't see ads — that's part of what your subscription covers. Users under 18 are excluded regardless of plan.

What's different in Europe. In the US, UK, and other countries where ads launched earlier this year, OpenAI could personalize ads based on your account and settings. In the EU, Iceland, Liechtenstein, Norway, and Switzerland, that's not allowed to start — European privacy law (GDPR) requires clear consent before a company can personalize ads around you, and OpenAI hasn't asked for that yet. So for now, the ads you see are based only on what you're currently asking ChatGPT about, your rough location, and your language — not your chat history or saved memories. OpenAI says it may introduce personalization later, and if it does, you should get a separate notice asking for your consent — you're not supposed to be opted in automatically.

Read the full article on AI for Regulars

The law, known as HF 1606, doesn't target the person who types a request into one of these tools. Instead, it goes after the companies and websites that offer the tool in the first place — banning any online service that lets someone turn a real photo into a convincing fake nude image of an identifiable person without that person's consent. Violators face civil penalties of up to $500,000 per violation, pursued by the state Attorney General. Separately, the law also gives victims themselves the right to sue an app's owner directly for damages.

The law hasn't gone unchallenged. Elon Musk's AI company, xAI, whose Grok chatbot can generate and edit images, sued to block it just before it took effect, arguing it's written so broadly that it violates free speech. A federal judge, Donovan Frank, denied xAI's request for an emergency pause on July 31 — his reasoning was that xAI had waited nearly three months after the law was signed before asking for emergency relief, which undercut its claim that it needed the court to act immediately. That let the law take effect on schedule. A full hearing on whether to block the law for longer took place on August 19; as of this writing, the judge hasn't yet issued his decision.

Read the full article on AI for Regulars

Much of the alarm traces back to a 2024 estimate from a UC Riverside researcher, widely reported as: writing a short, roughly 100-word response with ChatGPT could consume around 500 milliliters of water — about a standard bottle. That number spread fast because it's vivid and easy to picture.

There's an important update to that story that gets shared far less often. In 2026, the same researcher, Shaolei Ren, revised his own estimate down significantly — to roughly 15 milliliters for a typical modern prompt — pointing to more efficient chips and cooling systems, and noting that the original figure was based on specific 2024 assumptions that don't reflect how today's systems run. That's a more than 30-fold drop in the estimate, from the same source.

Read the full article on AI for Regulars

What the video claims. Multiple AI-generated videos spreading on YouTube say Congress passed a law giving homeowners 65 and older a federal property tax exemption, freeze, or refund. Some tell viewers to request a form called the "Senior Homeowner Tax Review Request"; others point to a fake "Form 1098-SR," dressed up to look like an official IRS document. Neither form exists, and there is no federal property tax program — in the United States, property tax is set and collected entirely at the state and local level, never by the federal government or the IRS.

How officials found out. County assessor and auditor offices didn't spot the videos first — they got flooded with phone calls and emails from residents asking for the "federal form." Douglas County, Colorado, and Franklin and Hamilton Counties in Ohio each issued public consumer alerts in early August after the calls piled up. St. Louis County, Minnesota, and Kern County, California, warned residents about the same videos. By mid-August, the financial site Moneywise counted more than 20 assessor offices across several states that had put out warnings. One Colorado official pointed to a simple motive: the videos run 15 to 20 minutes with heavy ad breaks, which pays out regardless of whether anything in them is true.

Read the full article on AI for Regulars

In the US, that means Google AI Pro, which normally costs about $20 a month (roughly $200 a year). Students in more than 140 other countries get Google AI Plus, a lower-priced tier. Both include higher usage limits on the Gemini app, Gemini built into Gmail and Google Docs, and extra Google storage.

To sign up, a student needs a personal Google account (not a school-issued one) and must verify their student status. Google also asks for a payment card at sign-up, even though the first year is free. The offer can be claimed through December 31, 2026.

Read the full article on AI for Regulars

What changed. On August 18, 2026, OpenAI launched ChatGPT for Teens, a separate version of ChatGPT built for users aged 13 to 17. Accounts get switched into it automatically — either because the person entered a birth date between 13 and 17 when they signed up, or because OpenAI's own age-prediction system estimates, from how the account is used, that it likely belongs to someone under 18. OpenAI has not said how accurate that automatic guess is, so it's worth treating it as an extra layer, not a guarantee.

What's off by default. In the teen version, ChatGPT won't use romantic language or terms of endearment, and it's instructed not to suggest it has feelings of its own. It also has stronger built-in protections around self-harm, eating disorders, violence, and sexually explicit content — topics where the standard version of ChatGPT is more permissive.

Read the full article on AI for Regulars

What changed. Gemini (and Google's video/music tool, Flow) now has a "Media watermark" switch in settings — On or Off. With it off, images made with Gemini's image tool, videos made with its video tool, and music made with its music tool no longer carry the small visible AI mark in the corner. You'll find it in the web version of Gemini by clicking the gear icon in the bottom left, then "Media watermark." Google has said it plans to bring the same option to Search.

What did NOT change. Turning the switch off only removes the mark you can see. Every file still carries an invisible digital watermark (called SynthID) and metadata (called C2PA) embedded in the file itself, regardless of the setting. Tools built to check for that hidden marker — like Google Search's "About this image" feature — can still identify the content as AI-generated even with the visible mark switched off.

Read the full article on AI for Regulars

Group Chat becomes 1:1 only. If you've been using Copilot in a group conversation with other people, that group format disappears on August 18. Microsoft says existing group chats will be turned into individual chats, so your own messages stay visible in your history — but anything other participants shared or created inside the group, like their prompts or generated images, will no longer be accessible to you. If there's something from a group chat you want to keep, save it before the 18th.

Podcasts are shutting down completely. Copilot's feature for turning documents or web pages into an AI-generated audio "podcast" is being removed, not just changed. After August 18 you won't be able to create new ones or open ones you already made, and any links you shared to a Copilot podcast will stop working. You can still download individual podcasts from your podcast library before the deadline — look for the download option in the podcast's menu.

Read the full article on AI for Regulars

Who actually sees ads. Only logged-in adults on the Free and Go plans. If you pay for Plus, Pro, Business, Enterprise, or Edu, you won't see ads at all — that's one of the things your subscription covers. Ads also never appear in temporary (incognito-style) chats, regardless of your plan.

What the ads look like. OpenAI says ads are visually separated from ChatGPT's actual answer and clearly marked as sponsored, and that they don't influence what the AI tells you — the answer itself is generated the same way either way. You can tap an ad to see why it was shown to you, dismiss it, leave feedback, or delete your ad data with one tap.

Read the full article on AI for Regulars

If you want to get better at spotting fakes before they fool you or someone you love, how to spot a deepfake video covers the visual signs, and is that really your family on the video call walks through verifying a live call. If the fake involves a cloned voice rather than an image, how AI voice cloning scams work explains how those are made and reported.

Read the full article on AI for Regulars

What's actually changing. The rollout starts on the web this week and reaches phones and tablets on August 17. Once it's live, eligible students can open a Gemini tab inside Classroom and turn their own class materials — a reading, a set of notes, a study guide the teacher posted — into flashcards or practice quizzes tailored to that specific assignment. Google calls these "contextualized starter prompts": instead of a blank chat box, students pick a class and an assignment first, and Gemini builds the study aid around it.

Your school has to turn it on — it doesn't switch on by itself. This is the part worth knowing before you worry: students only get access if their school's IT administrator has already enabled Gemini in Classroom, Gemini, and Gemini Notebook for that account. If your school's admin hasn't turned those on, nothing changes for your child on August 10 or 17 — the tab simply won't appear. Schools can also set up a separate group for students under 18 and switch access off for just that group, independent of older students or staff.

Read the full article on AI for Regulars

What's actually changing: Free and Go accounts are moving to GPT-5.6 Luna as their default model this week. The daily rate limit on text chats is being removed in the same rollout wave — OpenAI says that part lands the following week, starting around August 10. Once it does, you'll be able to keep typing and asking questions all day without hitting a wall. Both changes are rolling out gradually, so they may show up for you a few days before or after they show up for someone else — that's normal for this kind of update, not a sign anything is broken.

What's still limited: This change covers text only. Uploading files, generating images, using voice mode, and other tools still have daily limits on Free and Go. If you're doing a lot of image generation or document uploads, you'll still run into caps there — Plus and Pro remain the way to raise those.

Read the full article on AI for Regulars

Nobody sits down and decides to pay for three AI subscriptions at once. It happens one signup at a time: you tried ChatGPT Plus for a work project, added Gemini because it came bundled with Google storage you needed anyway, then picked up Perplexity during a free trial you meant to cancel. Each decision made sense on its own. The problem is that nobody ever goes back and asks whether all three are still earning their keep together — because there's no single screen that shows you every AI subscription in one place. Your bank statement lists them as separate line items, weeks apart, and it's easy to stop noticing a charge once it's routine.

Annual plans lock you in. Several services offer a cheaper annual rate — Claude Pro, for instance, drops to about $17/mo if you pay $200 upfront instead of $20/mo month to month. That's a real saving if you're confident you'll use it all year, but it also means an audit that finds you're barely using it won't get your money back until renewal.

Read the full article on AI for Regulars

Here's what was supposed to start on August 2, 2026. Under the EU AI Act, companies and public bodies that use "high-risk" AI systems — software that helps decide who gets a job interview, a loan, a school place, or how police and border agencies use AI — were meant to have finished risk assessments, documentation, and human oversight by that date. That's the part that's been delayed.

The delay came through something called the "Digital Omnibus on AI," a formal amendment to the AI Act. The European Parliament and EU governments approved it in June 2026, it was published in the EU's Official Journal on July 24, and it became binding law on July 27 — a week before the original deadline. The new dates: December 2, 2027 for those standalone high-risk systems, and August 2, 2028 for AI built into regulated products like medical devices or cars. EU officials said the extra time is needed because the technical standards and testing tools companies need to comply weren't ready.

Read the full article on AI for Regulars

The rule applies to any generative AI system with more than one million monthly users in California — in practice, that covers ChatGPT (OpenAI), Gemini (Google), Meta AI, and other large platforms. Since companies rarely build a separate version of a product just for one state, the changes will likely show up for users everywhere, not only in California.

What's supposed to change

Read the full article on AI for Regulars

And before anything else — this was not your fault. Modern scams are run by organised, well-funded operations that rehearse their scripts, clone real voices with AI, copy real bank websites down to the fonts, and deliberately create the kind of time pressure that switches off careful thinking. They fool lawyers, accountants and IT professionals every week. Shame is the tool that keeps victims quiet and slow, and slow is exactly what the scammer needs from you now.

This is the honest version. Some of these are worth fighting for; some are not coming back, and knowing which is which stops you wasting the hours that matter.

Read the full article on AI for Regulars

The best-known model, Ray-Ban Meta, is a normal-looking pair of glasses with a built-in camera, microphones, small speakers, and Meta's AI assistant. Say "Hey Meta" and you can:

None of this requires touching the glasses or pulling out your phone, which is the whole appeal: it's a camera and an assistant you're already wearing.

Read the full article on AI for Regulars

AI can write a confident-sounding answer that's still wrong. Never let a chatbot's guess about your coverage or payout replace your actual policy document or a conversation with your adjuster.

Storm-damage scams increase after major storms. Unsolicited contractors offering "free inspections" who then pressure you to sign a repair contract on the spot are a well-known pattern after storms. A legitimate contractor will let you get a second opinion.

Read the full article on AI for Regulars

What it actually does. Gemini used to open in a floating window over the page. Now it lives in a side panel that stays open next to whatever tab you're viewing. Click the "Ask Gemini" icon and it can summarize the page you're on, answer questions about it, or — if you choose to share more than one tab — compare information across up to 10 open tabs at once, which is handy for things like comparing prices or reading several reviews side by side. It also connects to other Google apps: it can draft an email and hand it to Gmail as a ready-to-send compose window, pull up relevant results in Maps, check your Calendar, or pull key points from a YouTube video's transcript. People on a paid Google AI Pro or Ultra plan also get "Auto Browse," a feature that lets Gemini carry out a multi-step task on your behalf across tabs, such as researching several options and filling in details for you.

What it can see. When you share a tab with Gemini — the tab you're currently viewing is shared by default once you open the panel — it reads that page's content and web address, and does the same for any extra tabs you add. Google says that data is logged temporarily to your account, but doesn't show up in your regular Gemini Apps Activity, and isn't used to train its models without permission. You can stop sharing a specific tab at any time by clicking the "X" next to its name in the panel, and you're not required to share anything to keep using Chrome normally.

Read the full article on AI for Regulars

What's new. Announced Thursday, July 23, 2026, ChatGPT Health lets you connect your Apple Health data — steps, sleep, workouts — straight into your ChatGPT conversations. If your doctor's office uses certain electronic record systems (Epic and Oracle Health are the two OpenAI named), you can link those too, so ChatGPT can see things like recent lab results or visit summaries. Once connected, you get a dedicated Health tab that keeps its own conversation history, separate from your regular chats, where you can track how a symptom, medication, or sleep pattern changes over time and prepare questions before a doctor's appointment. It works on the web and in the ChatGPT app on iPhone — it hasn't rolled out to Android yet.

Why now. OpenAI says people already send it roughly 300 million health-related questions a week, up from about 230 million back in January. Rather than leaving that happening informally in ordinary chats, the company built a dedicated space for it — with, it says, the same underlying model but a setup designed specifically around tracking health information over time rather than answering a one-off question.

Read the full article on AI for Regulars

The good news: the underlying tricks haven't changed, even if the writing has gotten better. These six checks catch most AI-polished scam texts in under a minute.

AI has also learned to imitate people you know. Scammers increasingly use AI to write a text or voice message that sounds like your adult child, grandchild, or a coworker asking for urgent help — often money sent quickly. If a text from a "known" number feels off, or asks for money or gift cards, call that person back on a number you already have saved, not one provided in the message.

Read the full article on AI for Regulars

What Meta's own watchdog found. On June 5, 2026, the Oversight Board — the independent body Meta itself set up to review its content moderation — published its first ruling on an account disablement case. It upheld Meta's decision to permanently disable the specific Instagram account in question, which had posted threats and slurs. But the board used the case to call out much bigger problems: it found Meta's system for sorting violations into "severe" and "egregious" categories, which determines whether an account gets a warning or an instant permanent ban, to be "unclear and often contradictory." The board's most pointed recommendation was that people should be told whether and how AI was involved in reviewing their account and deciding any penalty against it — something Meta currently doesn't disclose.

Thousands say they were caught up in it. A petition from the Canadian nonprofit People Over Platforms, titled "Hold Meta Accountable for Wrongful Account Disabling," has drawn tens of thousands of signatures over 2026 from people who say Facebook, Instagram, or Threads accounts — including long-running small business pages — were disabled without warning, often citing "severe" categories like child exploitation, with no human reachable to explain why. Local news outlets including CBS affiliates in Baltimore and Philadelphia have covered similar cases. Meta disputes that AI is the problem: a company spokesperson told Boston 25 News that its current moderation tools make 13% fewer mistakes than human reviewers did and catch 10% more genuine violations — but didn't dispute that appeals can be slow or unclear.

Read the full article on AI for Regulars

"AI info" is a label Meta applies automatically across Facebook, Instagram, and Threads, most often when it detects that a photo or video carries an industry-standard technical signal called Content Credentials, built on a standard known as C2PA. In that case, Meta doesn't need the person posting to tell it anything — the system reads hidden data embedded in the file itself, invisible to anyone just looking at the image. Meta also runs its own AI-detection systems as a backup for content with no embedded signal, but the metadata path is what causes most of the confusing, real-photo cases below.

That data can say a few different things: that the whole image was generated by an AI tool like DALL-E or Midjourney, or simply that some part of it — even one small edit — was made using an AI-powered feature in an editing app. Meta treats both cases the same way: it shows the label.

Read the full article on AI for Regulars

What Google is accused of. The case, In re Google Assistant Privacy Litigation (Case No. 5:19-cv-04286, U.S. District Court for the Northern District of California), alleges that Google Assistant sometimes activated and recorded audio without anyone actually saying "Hey Google" or "OK Google" — a mistake the industry calls a "false accept" — and that some of those recordings were reviewed by outside contractors to help improve the software. Google denies any wrongdoing and settled without admitting fault.

Who can file a claim. Two groups qualify, and you don't need to belong to both. You're likely eligible if, between May 18, 2016, and March 19, 2026, you or your minor child either bought a Google-made device — a Google Home, Home Mini, Home Max, Nest Hub, Nest Hub Max, or Pixel phone — or used any Google Assistant-enabled device (including third-party smart speakers with Assistant built in) and believe a conversation was picked up without the wake word being said. Both groups must also confirm they live in the US or a US territory.

Read the full article on AI for Regulars

This guide covers both sides.

In 2026, Visa announced an AI Financial Assistant that lets cardholders ask plain-English questions about their own transactions, lock their card, and set spending alerts — all inside their existing bank app. (Source: Visa, 2026.) Major banks have been adding similar features of their own, and what all of these have in common is the same basic idea: instead of hunting through menus, you just ask.

Read the full article on AI for Regulars

You have more options than that letter suggests.

Not every rejection is automated, but the clues are often there:

Read the full article on AI for Regulars

Data privacy is the top concern many Europeans have about AI tools. That concern is well founded — and this guide will help you understand what is actually going on, in plain English.

When you type something into a consumer AI chatbot and press send, your message goes to the company's servers. The AI generates a reply, which comes back to you. So far, so normal — this is how almost every web service works.

Read the full article on AI for Regulars

A NerdWallet study found that roughly half of prospective U.S. home buyers (48%) plan to use AI tools during their search, and about one in four (27%) to help estimate housing costs — though, as we will see, that last use case needs some care (Source: NerdWallet). Here is how to use AI well, and where to stop before it leads you astray.

Walking into a meeting with a mortgage broker without knowing the basic terms puts you at an instant disadvantage. AI is excellent at closing that gap before you sit down.

Read the full article on AI for Regulars

This guide walks you through the main categories, explains what each type actually does, and helps you decide which one fits your situation.

A smart doorbell replaces or sits alongside your regular doorbell. When someone walks up to your front door, the built-in camera sends a live picture to your phone or tablet. You can see who is there — and even talk to them through a speaker — without opening the door.

Read the full article on AI for Regulars

The reassuring part: you do not need to be technical to protect yourself. The four steps below work whether you have an iPhone, an Android phone, a Windows PC, or a Mac — and the most important first step takes about five minutes.

When you hear that AI can "crack" passwords, here is what is actually happening. Criminals obtain lists of usernames and passwords from old data breaches — leaks from large websites that happened years ago. Those lists are sold online. Software, increasingly powered by AI, then tries those username-and-password combinations on hundreds of other websites automatically. This is called credential stuffing.

Read the full article on AI for Regulars

Here is what it actually says, in plain English.

The law sorts AI systems into four categories based on how much harm they could cause. Think of it like food-safety ratings — a hospital kitchen and a home kitchen face very different rules.

Read the full article on AI for Regulars

The good news is that European AI tools do exist — and some of them are genuinely excellent for specific tasks. The honest answer, though, is that they are generally smaller and less capable than ChatGPT or Gemini for open-ended questions. This guide walks you through the main options, what each one is actually good at, and how to choose.

Here are the three main tools worth knowing about as of mid-2026:

Read the full article on AI for Regulars

How it works. A dedicated AI classifier scans chats with Meta AI for signs of self-harm or suicidal distress. Flagged conversations get a human review before any alert goes out, and Meta says that when the signal isn't clear-cut, it errs on the side of notifying a parent anyway.

What you'll see — and what you won't. If your teen's account is flagged, you'll get a notification letting you know there's cause for concern, along with crisis resources and tips for starting a conversation. You won't see the exact wording of what your teen wrote — Meta keeps that private.

Read the full article on AI for Regulars

That gap matters less because parents are being kept in the dark on purpose, and more because most teen AI use is genuinely unremarkable: asking for homework help, looking something up, or having a bit of fun. The real question isn't whether your teen has talked to a chatbot — most have — it's whether that use has drifted from casual into something that's replacing real relationships.

Not all AI chats are the same. General-purpose assistants like ChatGPT, Gemini, and Copilot are built to answer questions, help with writing, and handle tasks. Companion apps, most notably Character.AI (and others like Replika), are built differently — around ongoing, personality-driven conversations designed to feel like an evolving relationship. Some teens interviewed by researchers described these bots using language like "soulmate," and use them for emotional support or even romantic role-play, not homework.

Read the full article on AI for Regulars

What it actually does. Normally, you chat with ChatGPT back and forth: you ask something, it answers, you ask a follow-up. ChatGPT Work is different — you describe the outcome you want, and it works through the steps on its own, sometimes for a long stretch of time, and comes back with a finished result. That could mean turning a rough list of numbers into a formatted spreadsheet, building a slide deck from a set of notes, drafting a document, or pulling together a short webpage. It can keep working after you close the app, and it will check in with you before doing anything it considers a "consequential" action, like sending an email or making a purchase — you can also pause, redirect, or stop it at any point.

It works by connecting to your other apps. To do this, ChatGPT Work needs access to things like your email, calendar, or file storage (for example, Gmail, Google Drive, or Slack), depending on which connections you set up. That is the trade-off worth understanding before you turn it on: the more accounts you connect, the more the agent can see and touch on your behalf, so it's worth connecting only the accounts you actually want it working in, and reviewing what permissions you're granting.

Read the full article on AI for Regulars

How to use it. Look for a magnifying glass icon near the top of the sidebar, or press Ctrl+K on Windows or Cmd+K on a Mac, and start typing. The search looks through everything at once — your past chats, any Projects you've set up, and images or documents you've uploaded — and lets you narrow the results down by type instead of scrolling through one long mixed list.

It even finds archived chats. If you've ever archived a conversation to tidy up your sidebar, it normally disappears from view entirely. The new search still finds it — archiving hides a chat from the list, it doesn't remove it from search.

Read the full article on AI for Regulars

How to get it. Sign up for free through the Apple Beta Software Program, then install iOS 27 Public Beta from Settings > General > Software Update on your iPhone. No developer account and no fee required — just some time and a supported device.

Back up your phone first. A public beta is, by definition, not finished software. Apps can crash, battery life can suffer, and features can behave unpredictably. Before you install it, back up your iPhone to iCloud or your computer so you can restore it if something goes wrong.

Read the full article on AI for Regulars

Older hearing aids amplified sound using fixed rules — turn up all sound above a certain pitch, for instance. That helps with volume but doesn't do much to separate a voice from restaurant clatter or wind noise.

The AI in newer hearing aids is typically a small neural network — a chip trained on millions of recorded sound scenes — running directly on the device. Its job is to constantly classify what you're hearing: is this quiet, one person talking, a group conversation, or a noisy room? Based on that read, it adjusts amplification and noise reduction dozens of times per second, aiming to keep speech clear while turning down everything around it.

Read the full article on AI for Regulars

How it works. Avatar creation starts inside the Gemini app: tap "Add Files" in the message box and choose Avatar, then scan the QR code that appears using your phone. From there, you'll be asked to look into the camera, turn your head a few different ways, and read some numbers aloud so the system can map your face and voice. The whole thing takes a few minutes. Afterward, you can bring your avatar into a conversation by typing "@me" or your name, and Gemini's video-generation model — called Omni — can produce short videos where you appear to be speaking, using your actual face and voice.

Who can use it. Avatar requires a paid Google AI plan (Plus, Pro, or Ultra) — it isn't available on the free tier. You also need to be 18 or older, and you have to be the one physically present during setup; you can't create an avatar of someone else from a photo.

Read the full article on AI for Regulars

This is part of the EU AI Act's transparency obligations, which are now coming into force for consumers. Some business-side requirements under the Act have been delayed, but the rules that directly affect what you see and hear as a regular user are moving ahead on schedule, according to coverage from InsidePrivacy (May 2026).

What you'll notice

Read the full article on AI for Regulars

What's actually new. When you turn on Incognito Chat before talking to Meta AI, those messages are processed in a Trusted Execution Environment (TEE) — a secure, isolated part of the server that Meta says even its own engineers can't access. The AI chats also disappear by default when the conversation ends, leaving no history stored on Meta's servers.

Why this matters. Normally, when you chat with Meta AI in WhatsApp, those messages don't have the same protections as your regular WhatsApp conversations. Your regular messages to friends and family have always been end-to-end encrypted — Incognito Chat doesn't change that. What it does is bring your Meta AI chats closer to that standard for the first time.

Read the full article on AI for Regulars

Sol is the flagship. It's built for tasks that take serious thinking: complex research, writing long reports, coding projects, and multi-step problems where the AI has to plan ahead. If you've heard the phrase "deep reasoning," this is the model that does that. There's even a stronger version called Sol Ultra for the most demanding work. Most everyday users won't need this one for routine tasks.

Terra is the middle option. OpenAI describes it as balanced — roughly as capable as what ChatGPT used for last year, but at half the running cost. The most important thing for regular users: Terra is now the default model for Free and Go accounts. That means if you use ChatGPT without paying, you're getting a better AI starting today, automatically, with no action required on your part.

Read the full article on AI for Regulars