Europe heatwave linked to 1,300 excess deaths: Why it matters
Europe heatwave linked to 1,300 excess deaths: WHO
Latest AI/ML News
34703 matching items
I don't hide my use of AI. I want my kids to see how I use it to make life easier.
I use AI at baseball games, on family trips, and for work. My kids are learning how to think critically about it.
AI writes a lot of software. Now, human code review is starting to disappear.
AI coding agents like Cursor are increasingly trusted to work autonomously, with a rise in AI-generated code reaching production directly.
Govee’s smart nugget ice maker makes every iced drink feel like a luxury
Govee says that the modern-design gadget delivers nugget ice in as little as six minutes.
Iraqi political officials arrested on corruption charges in overnight raid
Iraqi security forces sealed off all entrances to the capital’s heavily fortified Green Zone early Sunday and carried out raids inside the compound that houses key government institutions and foreign embassies.
Keyboard Navigation Of Hyperlinks
Thanks for the feature request, @Starker! I can definitely see the value in being able to navigate hyperlinks in a conversation entirely from the keyboard. Before I pass the details on internally, would you mind answering a few questions? The answers will help provide the team with the right context: What are you ultimately trying to accomplish, in your own words? What do you do today instead, and what about that isn't working for you? If we solved this perfectly, what would be different for you or your team? Thanks again for taking the time to share your feedback! -Mark G.
Wrongfully banned account + Apple subscription lock = unable to use ChatGPT at all — please help
Case ID:C-VJCI8U2wyHLW Please review this carefully. Thank you. I am a legitimate user who pays for the service, but my account has been banned, and I am unable to top up even after switching to a new account.
Improve “Branch in New Chat” Context Handling
Thanks for sharing this idea, @aliexe. I can definitely see why that would be useful, especially for long technical conversations where you want a branch to become a more focused discussion instead of carrying along unrelated history. Giving users control over how much context is copied, whether that's the full conversation, up to a selected message, AI-generated minimal context, or just the selected message, seems like a thoughtful way to make branching more flexible. I've passed this along to the team so it can be logged as a feature request. It's a great suggestion, and I appreciate you taking the time to explain both the current limitations and the potential improvements. -Mark G.
Golden domes, historic statues, and nods to state flags: What the capitol looks like in every state
Every state has a capitol that houses its state legislature. Many are domed buildings similar to the US Capitol, but others are more unique.
Long-term Pro 20x account unexpectedly deactivated
Hi @redker, thanks for taking the time to explain what happened. I can understand how disruptive this is, especially when the account contains important research and ongoing project work. At this point, your appeal is with our specialized team for review. These cases require a thorough manual review, so we aren't able to provide a timeline for when it will be completed. To keep everything in one place, we'll be closing this forum thread. Any updates will be provided through your appeal ticket 10491324 . Thanks for your patience and understanding while the review is underway. -Mark G.
Evaluating Offline Monitoring of Internal AI Agents
This work was conducted during the GovAI Winter Fellowship 2026. Full report Executive Summary Frontier AI companies use offline monitoring to address risks from internally deployed AI agents. AI developers increasingly rely on AI agents for internal work, including for safety research and model training. At the same time, these companies are concerned that a misaligned model could exploit this access to take concerning actions, such as sabotaging efforts to understand the risks posed by AI. To identify such instances, AI companies have separate AI models called "monitors" that review transcripts of AI agents' actions and flag suspicious activity. Human reviewers examine activity flagged as suspicious by monitors, judge whether that activity is concerning, and decide on an appropriate response. This monitoring occurs offline, meaning that actions are reviewed after they have been executed rather than intercepted in real time. Companies currently assess the effectiveness of offline monitoring via synthetic attacks. To assess the effectiveness of offline monitoring, OpenAI and Anthropic use synthetic attacks – transcripts constructed to contain the kind of harmful actions a misaligned AI might take during deployment – and then check whether monitors flag them. Current reporting on assessments of effectiveness is insufficient. Given the information currently made public by Anthropic and OpenAI, external parties cannot assess the overall effectiveness of their offline monitoring…
We took a multi-gen trip to the 'Water Park Capital of the World'. It worked for our family, but I wouldn't go back.
Wisconsin Dells, "Water Park Capital of the World" had many family-friendly activities for our multigenerational travel group, but the cons added up.
Weekly demand forecasting: Should I train on weekly or daily data and then aggregate?
I'm currently working on a demand forecasting problem for inventory replenishment, and I'd love to hear how others would approach it. The business requests a forecast for the next 4 weeks of stock consumption around the middle of the previous month. For example, in mid-June, I need to forecast the weekly demand for July. The challenge is that, at the time the forecast is generated, transactions from the second half of June are not yet available , creating a gap between the latest observed data and the beginning of the forecast horizon. The data I have consists of purchase order transactions at the SKU level , including timestamp (date and time) and quantity consumed. My main question is about the appropriate time granularity for training the forecasting model: Option 1: Aggregate the data by SKU and ISO YearWeek , resulting in one observation per SKU per week, and train a model to directly predict the next 4 weeks. Option 2: Keep the data at the daily level , train a model to forecast daily demand, and then aggregate the daily predictions into ISO YearWeeks to obtain the required weekly forecasts. One additional detail is that the forecast is reported using ISO YearWeeks . As a result, some weeks within a calendar month may contain only 3 or 4 days of that month (e.g., at the beginning or end of the month), while others contain all 7 days. My question is: Which approach would you choose, and why? Is it generally better to train the model at the same frequency as the business…
Israel recognizes Armenian genocide amid tensions with Turkey
Israel’s government unanimously voted on Sunday to formally recognize the Armenian genocide, amid worsening ties with Turkey. “Despite the extensive and unambiguous historical documentation, the Armenian genocide remains to this day the subject of an institutionalized campaign of denial and minimization, including a manipulative rewriting of history, mainly by the Turkish government,” Israeli Foreign Minister […]
‘I will return this year’: Sheikh Hasina vows comeback, says Bangladesh democracy under threat
I will return this year: Hasina vows political comebac
Years of living with solar power taught me these 12 myths are simply wrong
Knowing the facts about home solar power can help you make better-informed choices, save money, and stay safe.
I Pitted XGBoost Against Logistic Regression on 358 Matches. The Boring Model Won.
A concrete bias–variance lesson: why the smallest model had the best cross-validated fit, and how to know when to reach for the big hammer. The post I Pitted XGBoost Against Logistic Regression on 358 Matches. The Boring Model Won. appeared first on Towards Data Science .
AI won't become a real coworker until it stops answering and starts finishing tasks
A survey paper by Tencent and several Chinese universities traces the path from chatbot to "digital colleague." AI systems won't become reliable coworkers, the researchers argue, until they finish entire tasks in persistent work environments instead of just generating answers. The key lies in combining persistent workspaces with reusable skills. The article AI won't become a real coworker until it stops answering and starts finishing tasks appeared first on The Decoder .
Dubai Customs deploys smart K9 fleet to boost anti-drug crackdown
Dubai Customs deploys smart K9 fleet to boost anti-drug crackdown
My 12-year-old son suggested we take the scenic route. It's helped us appreciate our town, and he's opened up to us more in the car.
My 12-year-old son suggested taking the scenic route, and now, my family has better conversations in the car. We also appreciate where we live more.
Israel formally recognises Armenian World War I 'genocide'
Israeli Foreign Minister Gideon Sa'ar said Israel had "fulfilled a moral duty by recognizing the historical truth, and rejecting attempts to deny it".
What to Do in Houston If You're Here for Business (2026)
Where to eat, stay, work, and eat some more while visiting Space City on business.
Coinbase joins the rush to Chinese AI models as Western labs face a pricing stress test
Coinbase CEO Brian Armstrong is switching his company to Chinese AI models like GLM 5.2 and Kimi 2.7. An automated routing system picks the best model for each request based on task and price, and better caching pushed the hit rate from 5 to 60 percent. Coinbase has cut its AI spending in half even as token usage keeps climbing. The article Coinbase joins the rush to Chinese AI models as Western labs face a pricing stress test appeared first on The Decoder .
WordPress Content Suggestions (Related Posts) - need beta testers
Sent you the direct message
France: 11 killed in civilian plane crash near Nancy
Police urged the public to stay away from the area around Tomblaine airport. The aircraft was reportedly carrying skydivers.
Nest’s quest to fix your thermostat
The founding story of Nest is pretty much a perfect tech myth. A legendary product maker (in this case, Tony Fadell) helps create one of the most successful products ever (the iPhone) and then rides off into the sunset to enjoy the rest of his life, only to have an experience that drags him back […]
I looked into OBD2 fuel savers, and found a much safer solution to save car gas instead
No, there's no cheap dongle that you can plug into your car to save you money. Here's the truth.
Ad-free streaming is a luxury now
This is The Stepback, a weekly newsletter breaking down one essential story from the tech world. For more news about the streaming industry, follow Emma Roth. The Stepback arrives in our subscribers' inboxes at 8AM ET. Opt in for The Stepback here. How it started Streaming was once a reprieve from cable. Not only could […]
_resources1_16a30b2ba6a_large.jpg?w=280)
I've worked in restaurants for 35 years. There's one food I avoid ordering on Mondays.
Brian "Rusty" Russino of The Cheesecake Factory shares restaurant insights, explaining his "Monday bun" rule for ordering fresh food.
I took an unpaid internship at 31. I learned more valuable lessons from my Gen Z fellow interns than from my bosses.
When I became an intern at 31, my fellow interns were a decade younger. They taught me about work-life balance and the power of asking 'why.'
I'm a career coach. Here's the 4-hour burnout-proof job-search strategy I recommend.
A career coach advises job seekers against spending all day applying to jobs. She says a four-hour routine is burnout-proof and more effective.
You can travel all around the Netherlands by train for €49 a month this summer
This summer rail deal makes exploring the Netherlands easier and less hefty on our wallets. Here’s how to make the most of it and where else in Europe you can save on rail travel
US Supreme Court is building its own massive police force
America's high court is spending millions to beef up security. At what cost?
UV face masks, cooling ties: how cultures outside Europe deal with heat
Countries on other continents, far from Europe, have long devised ways of coping with sweltering heat. Here are a few examples – some unusual, others inspiring.
UV face masks, cooling ties: how cultures outside Europe cope with heat
Countries on other continents, far from Europe, have long devised ways to cope with blistering heat. Here are some examples, both unusual and inspiring.
I graduated debt-free, but my husband had $20K in student loans. Here's what we want our kids to know about taking out loans for college.
My husband had $20,000 in student loan debt, while I graduated debt-free. I care more about my kids avoiding debt than where they go to college.
I left Google after making nearly $1M in a year. Fears about layoffs and missing out on the AI boom gave me the push.
A former Google employee says AI equity, job security concerns, and years of side projects convinced him to leave and build his own company.
Power Laws in NNs: A Possible Mechanism for Inductive Bias towards Sparse Representations
This post was produced as part of the Iliad Fellowship under the mentorship of Dmitry Vaintrob. Tl;dr: Power-law ("heavy-tailed") distributions have universality theorems similar to those which make Gaussians common. We observe many things in ML are power-law distributed, most robustly and interestingly, the spectra of weight matrices. I explain how we can think of power-laws as being a natural generalization of the idea of 'sparsity', interpolating between true sparsity and Gaussianity according to the 'tail-index' of the distribution. I share some hypotheses about how this might relate to the 'sparse'/'discrete'/'factored' representations that neural networks seem to learn. I promise this is not a Santa-Fe-Institute encomium for power laws or "black swans"; different genre. Contents 1. The generalized central limit theorem proves power-law distributions are universality classes 2. Power laws observed in NNs might help us understand representation learning 2.A. HTSR: phase changes in weight-matrix spectra and data-free prediction of generalization 2.B. BBP transition as a quantum of learning 2.C. HTSR as an extended BBP transition 2.D. Training evidence for heavy tails is mixed, and I'm not sure if they're important 3. The tail exponent α is a smooth proxy for sparsity and compressibility 3.A. α captures compressibility across heavy tails 3.B. α-stable noise can make discrete codebooks optimal 3.C. Heavy-tailed noise can convert analog inputs into discrete codebooks 4. Summ…
Refusal Is Complicated As Hell: An Update
TL;DR It would make sense to briefly skim through our previous post that introduces our experiments on refusal in LLMs . There we explain how it started, here we’ll tell how it’s going. The primary goal of this text is to try and structure the list of whack-a-mole research questions. The secondary goal is to get some outside perspective, so if you run a similar research or have seen a similar research, please lend us a hand. Feel free to jump straight to the section that looks most appealing. We recommend skimming through “The Main Question” as this section provides a broader perspective. Then we listed all other questions that arose during research. You’ll find them under headers “Another Question: …” and “Wording Also Matters”. The first one discusses how refusal is represented in different layers and what it might mean. The second one is dedicated to two parts of refusal – its wording and actual detection of a potentially harmful request. “The Main Question” is split into two parts: in “Our suggestion” we outline our main hypothesis and proofs we found during our experiments; in “An Alternative Suggestion” we highlight the opposing point of view and proofs behind it. The Main Question (MQ) We experiment on open-weight small (~9B) instruct models trying to understand what exactly happens when they refuse to provide an answer given different contexts. One of the core observations is, refusal looks different for different categories of potential harm (for example, a request…