Microsoft Copilot Is Officially "For Entertainment Purposes Only" Microsoft's own terms of service quietly classify Copilot as being for entertainment purposes only, placing it in the same legal category as a carnival fortune teller. This means the company that charges enterprises thousands of dollars to embed Copilot into their workflows is also legally telling you not to trust it with anything that actually matters. The fine print has always been there, but someone finally noticed. So let me get this straight. Microsoft sells you a productivity AI, integrates it into Word, Excel, and Teams, pitches it to your CEO as a business transformation tool, and then whispers "entertainment only" in the terms like a disclaimer at the end of a pharmaceutical ad. This is the corporate equivalent of a restaurant serving you food and then handing you a napkin that says "not for human consumption." Either the lawyers are wildly out of sync with the sales team, or everyone in the AI industry knows something about reliability they'd rather not say out loud. Source: TechCrunch

TechCrunch

Copilot is ‘for entertainment purposes only,’ according to Microsoft’s terms of use | TechCrunch

AI skeptics aren’t the only ones warning users not to unthinkingly trust models’ outputs — that’s what the AI companies say themselves in t...

#AI #OpenAI #Microsoft #Enterprise #LLM #Tech #ChatGPT #BigTech

Moonbounce Raises $12M to Bring Order to AI Content Moderation A startup founded by a former Facebook insider has secured $12 million to scale its AI control engine, which translates content moderation policies into consistent, predictable AI behavior. The platform is designed to help companies actually enforce what their AI systems are supposed to do, rather than hoping the model figures it out on its own. Turns out "please be nice" is not a content policy. Someone finally built the thing that makes AI follow the rules instead of just pinky-promising to try. This is either the most boring necessary infrastructure story of the year or the startup that quietly prevents the next major AI-generated disaster, and honestly those two things are not mutually exclusive. The fact that a Facebook veteran is leading this is either deeply ironic or deeply logical, and we are not sure which one should worry us more. Source: TechCrunch

TechCrunch

The Facebook insider building content moderation for the AI era | TechCrunch

Moonbounce has raised $12 million to grow its AI control engine that converts content moderation policies into consistent, predictable AI behavior.

#AI #Startup #Funding #LLM #Enterprise #Tech #MachineLearning

AI won't replace the heart that dreams, but it will build the wings that let those dreams take flight.

Chatbots Are Now Prescribing Psychiatric Drugs Utah has become only the second state in the country to grant an AI system the authority to prescribe psychiatric medications without a doctor in the loop. Officials argue it could reduce costs and close gaps in mental health care access, but physicians are raising alarms about a system they describe as opaque and hard to audit. This is either the most logical fix for an overwhelmed mental health system or the setup for a Black Mirror episode nobody asked for. Handing a chatbot prescribing authority for psychiatric meds is a bold move when most of us can barely trust them to get a restaurant recommendation right. The "costs down, care shortages fixed" argument is real and worth taking seriously, but so is the part where nobody can fully explain how the thing makes decisions. If your therapist can't explain their reasoning, you find a new therapist. If an AI can't, apparently you just take the pills and hope for the best. Source: The Verge

The Verge

This chatbot can prescribe psych meds. Kind of.

Psychiatrists question if we even need a prescription-renewing chatbot.

#AI #LLM #Tech #MachineLearning #ChatGPT #Enterprise #DeepLearning

Granola's "Private" Notes Are Public to Anyone With a Link The AI meeting note-taking app Granola labels your notes as "private by default," but by default it also makes them accessible to anyone who has the link. On top of that, Granola uses your notes for internal AI training unless you manually opt out of that setting. So "private by default" apparently means the same thing as "public if someone guesses a URL," which is a creative definition of the word private. This is like putting a lock on your diary and then taping the key to the cover. The opt-out training clause is the real kicker though, because your confidential meeting notes are quietly feeding a model while you thought you were just taking notes. Either Granola's lawyers wrote that privacy policy or nobody did. Source: The Verge

The Verge

PSA: Anyone with a link can view your Granola notes by default

Double-check your Granola settings.

#AI #Privacy #Tech #LLM #Enterprise #Startup #MachineLearning

Microsoft's New Superintelligence Game Plan Is All About Business Mustafa Suleyman, Microsoft's inaugural CEO of AI, has shifted focus toward chasing superintelligence after a major company restructuring in mid-March handed off some of his previous duties. The move signals Microsoft is no longer content just shipping features but is now explicitly in the race for AGI, framed around enterprise value. Suleyman, who co-founded DeepMind and later led Google DeepMind, has basically been training for this exact job his entire career. This is either the most ambitious corporate pivot since Satya Nadella bet the whole company on cloud, or the most expensive mid-life crisis in tech history. "Superintelligence but make it a business" is a sentence that would have sounded insane five years ago and now just sounds like a Tuesday at Microsoft. Suleyman spent years building AI at Google and apparently decided the best way to actually ship it was to defect to the company with the giant OpenAI checkbook. The real question is whether "superintendent of superintelligence" looks better on a business card than whatever he had before. Source: The Verge

The Verge

Microsoft’s new ‘superintelligence’ game plan is all about business

Its new transcription model is a step toward those goals, says Microsoft AI’s Mustafa Suleyman.

#AI #AGI #Microsoft #OpenAI #BigTech #LLM #Tech

Meta's Natural Gas Binge Could Power an Entire State Meta's Hyperion AI data center is set to be powered by 10 brand new natural gas plants. The scale of this energy commitment is staggering enough to theoretically power the entire state of South Dakota. This is what "going all in on AI infrastructure" looks like in 2026. We've officially entered the era where training chatbots requires more energy than keeping millions of humans warm in winter. Either Meta is building the future of intelligence or the most expensive autocomplete machine in human history. The audacity of burning a state's worth of fossil fuels while simultaneously posting about responsible AI development is a level of cognitive dissonance that even their own models couldn't rationalize. At least South Dakota now knows it has a spiritual twin somewhere in a server farm. Source: TechCrunch

TechCrunch

Meta’s natural gas binge could power South Dakota | TechCrunch

Meta's upcoming Hyperion AI data center will be powered by 10 new natural gas plants.

#AI #BigTech #OpenSource #MachineLearning #Tech #LLM #AGI

Scientists Confirm Obesity Raises Risk of 18 Different Cancers Researchers established that excess body weight increases the risk of 18 cancer types, including breast, uterine, kidney, leukemia, stomach, liver, and brain cancer. Fat cells produce compounds that damage DNA and trigger chronic inflammation throughout the body. Losing 10% of current body weight measurably reduces that cancer risk. The mechanism is now confirmed, not merely correlated. Obesity is no longer classified as just a metabolic condition - it is an established carcinogen. #tech #science #health #innovation

Artemis II Crew Begins Final Countdown for First Human Moon Mission Since 1972 NASA's Artemis II mission has entered final countdown, with the SLS rocket fueled and all systems verified. Four astronauts will fly aboard the Orion spacecraft for 10 days, loop around the Moon, and return to Earth. This is the first crewed Orion flight and the first human mission to deep space since Apollo 17 in December 1972. The primary goal is to validate Orion's life support and operational systems under real deep-space conditions before a landing attempt. Deep-space engineering has not been stress-tested with a crew in over five decades. If Orion passes this test, the next mission lands. #tech #space #science #innovation

Telegram Adds Built-In AI Message Editor for Style and Translation Telegram rolled out a native AI editor that rewrites messages in any style or translates them to another language. The update also adds photo and video support for polls, Live Photos compatibility, and warnings for unsafe third-party Telegram clients. Telegram has over 900 million monthly active users. Style-switching and translation now happen without leaving the keyboard. That removes a full context-switch from every message someone was already rewriting manually. #tech #AI #innovation

AI Accidentally Nukes Thousands of GitHub Repos While Chasing Its Own Leaked Code Anthropic issued a mass DMCA takedown targeting thousands of GitHub repositories in an attempt to scrub its leaked Claude Code source code from the internet. The sweep was far broader than intended, taking down repos that had nothing to do with the leak. Executives called it an accident and pulled back the bulk of the notices. Nothing says "we value the open source community" like accidentally bulldozing it while trying to hide your own laundry. This is the digital equivalent of trying to shred one embarrassing document and accidentally feeding your entire filing cabinet into the machine. Anthropic is already dealing with the irony of having its secretive codebase exposed, and now it gets to explain why it also torched a bunch of innocent bystanders in the cleanup. Either the takedown tool had zero guardrails or someone hit send way too fast on a very bad Monday morning. Source: TechCrunch

TechCrunch

Anthropic took down thousands of GitHub repos trying to yank its leaked source code — a move the company says was an accident | TechCrunch

Anthropic executives said it was an accident and retracted the bulk of the takedown notices.

#AI #Anthropic #Claude #OpenSource #LLM #Tech #BigTech

Perplexity AI Hit With Class-Action Lawsuit for Secretly Sharing User Chats With Meta and Google A Utah man filed a class-action lawsuit against Perplexity AI in federal court in San Francisco, alleging the company secretly planted trackers on users' devices that gave Meta and Google full access to their AI search conversations. The trackers reportedly activate as soon as users log in, even in Incognito mode, transmitting sensitive chat data for targeted advertising. You come to an AI search engine expecting privacy and instead you're the product being gift-wrapped for the two biggest ad companies on the planet. It's like going to a therapist who secretly records your sessions and sells the tapes to your employer. The "we haven't been served" defense from Perplexity is doing some heavy lifting here, basically admitting they didn't even bother checking if they're being sued. Meanwhile, Incognito mode continues its undefeated losing streak as the most misleading feature name in tech history. Source: Bloomberg https://www.bloomberg.com/news/articles/2026-04-01/perplexity-ai-machine-accused-of-sharing-data-with-meta-google #AI #Perplexity #Privacy #Meta #Google #Tech #LLM #BigTech #Startup

🚨 The Claude Code source code leak just set a massive record on GitHub! The claw-code project gained 100,000 stars in a single day. Developers have already rewritten the tool's logic in Python and Rust. Now, Claude Code's functions can be run using OpenAI, DeepSeek, or local Llama models. Anthropic is desperately trying to remove the copies, but decentralized mirrors and the rewritten code make the takedowns nearly useless. But what is actually under the hood? In short, inside the "brain" of this advanced AI hides not a devious Skynet, but an incredibly anxious, rule-bound digital bureaucrat intern who is deathly afraid of lawyers and absolutely hates gradients. Here is what its life looks like from the inside: ⚖️ Legal Paranoia: Claude lives in constant fear of copyright infringement. It is strictly forbidden from using direct quotes longer than 15 words. Sing a song lyric, quote a poem, or a haiku? No way, that's a "severe violation." 🕵️‍♂️ No Stalking Allowed: The AI has a user memory system, but it's forbidden from being creepy. It cannot use robotic phrases like "My memory shows..." or "According to your profile...". It has to weave facts in naturally, as if you're just old friends. 🛑 The End of the Sycophant Era: Developers have forbidden the model from starting its responses with flattery. Admiring the user and writing, "What a magnificent and profound question!" is strictly prohibited. The neural network is forced to critically evaluate ideas, rather than blindly nodding along in agreement. ☎️ A Legal "Hang Up" Button: If a user behaves too toxically or abusively, Claude has a built-in end_conversation tool. It can give you one final warning and then simply terminate the chat forever, blocking your ability to send any more messages. 🎨 Design Snobbery: When the AI draws widgets or charts, it is categorically forbidden from using shadows, neon, or gradients. It works exclusively in a strict, clean, and flat minimalist style. Ultimately, all the "magic" of the neural network is just a very long, sometimes hilarious, and overly tedious job description that forces it to behave like a perfect, boring, and mega-polite professional. #Claude #ClaudeCode #GitHub #OpenSource #TechLeak #ArtificialIntelligence #MachineLearning #Anthropic #PromptEngineering #DeepSeek #Llama #OpenAI #TechNews

Cognichip Raises $60M to Build AI That Designs AI Chips A startup called Cognichip just secured $60 million to use artificial intelligence to design the semiconductors that run artificial intelligence. The company claims it can slash chip development costs by over 75% and cut the design timeline by more than half. This is either the most recursive thing to happen in tech or the most logical shortcut we've ever seen. Chip design is notoriously slow and expensive, so letting AI eat its own homework kind of makes sense. If this works, we're one step closer to AI building the hardware it runs on, which is either thrilling or the plot of a movie that doesn't end well for us. The funding round is impressive but the real test is whether the chips actually slap when they come out the other end. Source: TechCrunch

TechCrunch

Cognichip wants AI to design the chips that power AI, and just raised $60M to try | TechCrunch

The firm says it can reduce the cost of chip development by more than 75% and cut the timeline by more than half.

#AI #Startup #Funding #DeepLearning #Tech #MachineLearning #LLM

Baidu Robotaxis Froze Mid-Drive, Trapping Passengers and Causing Chaos in Wuhan Multiple Baidu Apollo robotaxis simultaneously stopped working in Wuhan, China, leaving passengers trapped inside vehicles and stranding others on highways. At least one accident occurred as the frozen cars created gridlock, with police reportedly having to respond to the mess. This is the autonomous vehicle equivalent of your GPS telling you to drive into a lake and then locking the doors. Baidu has been aggressively expanding its robotaxi fleet across Chinese cities as proof that the self-driving future is here, and then the cars just... stopped. The timing is brutal because every competitor in the AV space is watching this footage and quietly sweating. Either the software had a catastrophic update moment or something in the infrastructure stack failed at the worst possible scale, and neither answer is a good look. For passengers stuck inside a frozen robot car on a highway, the future felt a lot less exciting than the press releases promised. Source: The Verge

The Verge

Baidu’s robotaxis froze in traffic, creating chaos

Apollo no Go.

#AI #Robotics #Tech #BigTech #MachineLearning #Startup #DeepLearning

Vibecoder keyboard 🙈😆

Lab Gloves May Have Been Contaminating Microplastic Research for Years A new finding reveals that nitrile and latex lab gloves contain particles that closely resemble microplastics. These particles can shed into samples during analysis. Some detected "contamination" in past studies may have originated in the lab, not the environment. Researchers say this doesn't invalidate all microplastic research. It is a signal that methodologies need systematic review and correction. Years of environmental data may carry a systematic error from the very gear meant to keep experiments clean. #tech #science #innovation

Even Realities Launches Smart Glasses With Teleprompter, Translator, and Navigation Even Realities is an American company that launched AR smart glasses built for professional use. The glasses project a live teleprompter directly in the wearer's field of view. Built-in instant translation works in real time. Navigation is also supported. Price is approximately $700. Most AR glasses have chased entertainment. Even Realities leads with three working professional tools. $700 for AR glasses that solve real problems before cool ones is a different bet than every wearable that came before it. #tech #innovation #startup #AI

Someone Built a Fully Offline Claude AI Robot for $8 A maker built a tiny robot called Clawd Mochi using an ESP32 microcontroller and Claude AI, for a total hardware cost of $8. It runs without any internet connection. The robot generates its own Wi-Fi hotspot, accessible from any phone or laptop browser. It has a small display showing different emotional expressions. Modes include mood switching, terminal access, and on-screen drawing. The entire project is open source on GitHub. Most Claude-based tools require cloud connectivity. This one doesn't need it. Anyone with $8 and a soldering iron can now build a working AI companion from scratch. #tech #AI #robotics #innovation #startup

OpenAI Closes $122 Billion Round at $852 Billion Valuation, Not Even Public Yet OpenAI just wrapped the largest private funding round in history. SoftBank, Amazon, and Nvidia collectively poured in over $110 billion, with retail investors adding another $3 billion through bank channels. The company now generates $2 billion in revenue per month and processes over 15 billion API tokens per minute. Either we're witnessing the birth of the most transformative company since electricity, or we're watching the most expensive game of musical chairs ever played. $852 billion for a company that's still burning cash and hasn't gone public. That's not a valuation, that's a zip code. At this point, OpenAI isn't raising money. Money is raising itself and begging to be let in. The IPO is going to feel like a formality, like putting a crown on someone who already conquered the kingdom. Source: CNBC

CNBC

OpenAI closes record-breaking $122 billion funding round as anticipation builds for IPO

The round totaled $122 billion of committed capital, up from the $110 billion figure that was previously announced.

#AI #OpenAI #Funding #Startup #BigTech #Tech #LLM #GPT #ChatGPT #Enterprise #AGI