Theres a reason leftover risotto is not really a thingit gets gluey and broken if you try reheating it. Its considered normal practice to make risotto fresh, stirring constantly and ladling small installments of liquid into the pot for 30 minutes until its perfect. I wouldnt blame you if thats exactly why you never make risotto at home; what a chore. But theres actually a less annoying way to make this creamy rice dish. Its overnight risotto, and its my favorite way to prepare it.Perfectly cooked risotto is a thing of beautytender individual grains of rice loosely bound by a creamy, savory gravy of its own making. Theres a short window before this velvety dish becomes thick and claggy, and a limit for reheating it with more liquid. Overdoing it, whether on the stove with too much broth, or microwaving leftovers, overwhelms the starch capsules in the rice and causes them to burst. Youll see this as gritty, broken rice grains (good risotto has visibly whole grains) and spoonfuls of risotto that look closer to stiff mashed potatoes than rice in a flowing sauce.This isnt surprising; many starchy ingredients will act the same way. Mashed potatoes can get sticky if blended with a blade; oats can be overcooked; and mac and cheese will get too soft and break apart if you overcook the pasta. But there is a surprisingly effective trick that I remembered from Ina Gartens cookbook Go-To Dinners, which I reviewed a few months ago to make her Overnight Mac and Cheese. She only partially cooks the pasta so its still hard in the middle. She lets it sit in liquid overnight in the fridge. The next day, simply top it with more cheese and bread crumbs, and bake it off in the oven. The pasta gently absorbs some liquid when its in the fridge; it finishes cooking through in the oven, and the loose starches work together with the shredded cheese and cream to make a decadent sauce without the need for a roux. It's a truly stellar mac and cheese designed to be an easier work load.Overnight risotto uses the same idea: only par-cook the rice, so the starches dont reach capacity, and finish it in a pan in about three minutes the next day. Its absolutely perfect, and so much faster than the traditional way. You might be making risotto a lot more starting today. Heres how to do it.Overnight Risotto RecipeIngredients:1 shallot, minced1 tablespoon olive oilPinch salt cup dry arborio rice cup dry white wine2 to 2 cups broth (divided) cup shredded parmesan cheese cup frozen peas1 tablespoon butter1. In a large frying pan, add the olive oil, minced shallot, and salt. Cook over medium-low heat, stirring occasionally until the shallot is translucent.2. Add the rice and stir for about a minute. Pour in the wine, and stir until its absorbed, about 30 seconds. Add one and a half cups of broth to the pan and let the rice absorb this completely, you can stir once or twice. Once its absorbed, taste a grain. It should have a soft ring on the outside and a stiff-chalky interior. Take the rice off the heat. This process should only take 10 minutes. You can see the outside ring of the par-cooked rice is more translucent than the center. Credit: Allie Chanthorn Reinmann 3. Pour the rice mixture immediately out onto a plate or wide bowl and spread it out to cool for 10 minutes or so. Pour the rice into a storage container and pour a half-cup of the remaining broth over it. (Pouring the liquid in at this step is optional. You can, instead, just add the liquid when youre ready to cook it tomorrow, but it might then take an extra couple minutes.) Cover it with a lid and put it in the fridge overnight. The next day, add any green veggies as you finish off the risotto. Credit: Allie Chanthorn Reinmann 4. When youre ready to have your risotto, pour the cold, par-cooked rice and liquid into a large frying pan. A little broth will have been absorbed but not all of it. Cook over medium low heat, stirring frequently. Once it begins bubbling vigorously, add the cheese, peas (or other delicate veggies), and butter. (Theres a half-cup extra broth built into the ingredient list in case you like a looser sauce.) Stir until the liquid is absorbed but the sauce still pools a bit. This takes about three minutes. Serve immediately.

4 ways design trends are embracing minimalism this year.

Three exotic new species of superconductivity were spotted last year, illustrating the myriad ways electrons can join together to form a frictionless quantum soup.

MacworldYou never realize how much you need a pen until you dont have one. Whether its jotting down a million-dollar idea or catching someones phone number before they lose interest, not having one is a real buzzkill. This keychain pen fixes that problem forever.And get this: it writes with titanium, meaning it writes on any surface and is inkless, smudge-free, and capable of writing underwater because you never know when a mermaid might give you her number. The ForeverPenis only $23.99 here (reg. $40).Its also a great multi-toolWith this 1-inch-long gadget on your keyring, you can do more than write. Its so durable that it can withstand fires, so you can use the keyring multi-tool to open Amazon packages, mark wood surfaces, clean your nails, cut lightweight cords, and maybe even strike up a fire.Grab this small EDC tool to always have a pen and other tools handy while its only $23.99 (reg. $40).ForeverPen Tiny Inkless Pen & Multi-Tool $23.99See DealStackSocial prices subject to change.

Large Language Models (LLMs) have revolutionized generative AI, showing remarkable capabilities in producing human-like responses. However, these models face a critical challenge known as hallucination, the tendency to generate incorrect or irrelevant information. This issue poses significant risks in high-stakes applications such as medical evaluations, insurance claim processing, and autonomous decision-making systems where accuracy is most important. The hallucination problem extends beyond text-based models to vision-language models (VLMs) that process images and text queries. Despite developing robust VLMs such as LLaVA, InstructBLIP, and VILA, these systems struggle with generating accurate responses based on image inputs and user queries.Existing research has introduced several methods to address hallucination in language models. For text-based systems, FactScore improved accuracy by breaking long statements into atomic units for better verification. Lookback Lens developed an attention score analysis approach to detect context hallucination, while MARS implemented a weighted system focusing on crucial statement components. For RAG systems specifically, RAGAS and LlamaIndex emerged as evaluation tools, with RAGAS focusing on response accuracy and relevance using human evaluators, while LlamaIndex employs GPT-4 for faithfulness assessment. However, no existing works provide hallucination scores specifically for multi-modal RAG systems, where the contexts include multiple pieces of multi-modal data.Researchers from the University of Maryland, College Park, MD, and NEC Laboratories America, Princeton, NJ have proposed RAG-check, a comprehensive method to evaluate multi-modal RAG systems. It consists of three key components designed to assess both relevance and accuracy. The first component involves a neural network that evaluates the relevancy of each retrieved piece of data to the user query. The second component implements an algorithm that segments and categorizes the RAG output into scorable (objective) and non-scorable (subjective) spans. The third component utilizes another neural network to evaluate the correctness of objective spans against the raw context, which can include both text and images converted to text-based format through VLMs.The RAG-check architecture uses two primary evaluation metrics: the Relevancy Score (RS) and Correctness Score (CS) to evaluate different aspects of RAG system performance. For evaluating selection mechanisms, the system analyzes the relevancy scores of the top 5 retrieved images across a test set of 1,000 questions, providing insights into the effectiveness of different retrieval methods. In terms of context generation, the architecture allows for flexible integration of various model combinations either separate VLMs (like LLaVA or GPT4) and LLMs (such as LLAMA or GPT-3.5), or unified MLLMs like GPT-4. This flexibility enables a comprehensive evaluation of different model architectures and their impact on response generation quality.The evaluation results demonstrate significant performance variations across different RAG system configurations. When using CLIP models as vision encoders with cosine similarity for image selection, the average relevancy scores ranged from 30% to 41%. However, implementing the RS model for query-image pair evaluation dramatically improves relevancy scores to between 71% and 89.5%, though at the cost of a 35-fold increase in computational requirements when using an A100 GPU. GPT-4o emerges as the superior configuration for context generation and error rates, outperforming other setups by 20%. The remaining RAG configurations show comparable performance, with an accuracy rate between 60% and 68%.In conclusion, researchers RAG-check, a novel evaluation framework for multi-modal RAG systems to address the critical challenge of hallucination detection across multiple images and text inputs. The frameworks three-component architecture, comprising relevancy scoring, span categorization, and correctness assessment shows significant improvements in performance evaluation. The results reveal that while the RS model substantially enhances relevancy scores from 41% to 89.5%, it comes with increased computational costs. Among various configurations tested, GPT-4o emerged as the most effective model for context generation, highlighting the potential of unified multi-modal language models in improving RAG system accuracy and reliability.Check out the Paper. All credit for this research goes to the researchers of this project. Also,dont forget to follow us onTwitter and join ourTelegram Channel andLinkedIn Group. Dont Forget to join our65k+ ML SubReddit. Sajjad Ansari+ postsSajjad Ansari is a final year undergraduate from IIT Kharagpur. As a Tech enthusiast, he delves into the practical applications of AI with a focus on understanding the impact of AI technologies and their real-world implications. He aims to articulate complex AI concepts in a clear and accessible manner. [Recommended Read] Nebius AI Studio expands with vision models, new language models, embeddings and LoRA (Promoted)

This effect is inspired by Zoes ability Sleepy Trouble bubble from League of Legends, reimagined with references in my own style using Unity, focusing only on Shader Graph. Blender for Mesh and Photoshop for Textures (Generated and Hand Painted). Please feel free to give any suggestions or question in the comments, Im a new VFX artist, hoping to improve and make cool looking VFX. Thankyou!My art station post for a Breakdown

Project DigitsNvidiaAt CES 2025, Nvidia announced its first personal AI supercomputer, Project Digits. The rise of generative AI demands access to a new generation of CPUs and GPUs for data scientists and AI engineers working on state-of-the-art models and solutions. Digits is targeted at developers and data scientists looking for an affordable and accessible hardware and software platform to handle the lifecycle of generative AI models. From inference to fine-tuning to developing agents, Digits has everything it takes to build an end-to-end generative AI solution.Here is a deep-dive analysis of Project Digits:With a price tag that starts at $3000, Nvidia Project Digits is a compact device driven by the innovative Nvidia GB10 Grace Blackwell Superchip. It empowers developers to prototype, fine-tune and execute massive AI models locally. This accessibility marks a significant step toward democratizing AI training, making it more accessible for individuals and smaller organizations.Project DigitsNvidiaIndependent software vendors can leverage Digits as an appliance to run their AI-powered software deployed at a customer location. This reduces the reliance on the cloud and delivers unmatched privacy, confidentiality and compliance.Hardware Specs: AI Powerhouse on the DesktopAt the heart of Digits lies the Nvidia GB10 Grace Blackwell Superchip, a marvel of engineering that combines a powerful Blackwell GPU with a 20-core Grace CPU. These two powerhouses are interconnected using NVLink-C2C technology, a high-speed, chip-to-chip interconnect that facilitates rapid data transfer between the GPU and CPU. Think of it as a superhighway connecting two bustling cities, allowing seamless and efficient communication. This tight integration is crucial for Digits impressive performance, enabling it to handle complex AI tasks quickly and efficiently.MORE FOR YOUHeres a breakdown of the key specs:Blackwell GPU: Includes CUDA cores and fifth-generation Tensor Cores for accelerating AI calculations.Nvidia Grace CPU: Features 20 power-efficient Arm cores complementing the GPU for balanced AI workloads.NVLink-C2C interconnect: Provides a high-bandwidth, low-latency connection between the GPU and CPU for efficient data transfer.128GB of unified memory: Shared memory pool for CPU and GPU, eliminating data copying and speeding up processing.High-speed NVMe storage: Ensures rapid access to data for AI model training and execution.1 petaflop of AI performance: Handles complex AI tasks and large AI models.Power-efficient design: Operates using a standard wall outlet.Digits also boasts an impressive 128GB of unified memory. This means the CPU and GPU share the same memory pool, eliminating the need to copy data back and forth and significantly speeding up processing. This is particularly beneficial for AI workloads, which often involve large datasets and complex computations. Digits incorporates high-speed NVMe storage to enhance performance further, ensuring rapid access to the data required for AI model training and execution.Despite its impressive performance capabilities, Digits is designed with power efficiency in mind. Unlike traditional supercomputers that often require specialized power and cooling infrastructure, Digits can operate using a standard wall outlet. This makes it a practical and accessible solution for individuals and smaller teams who may not have access to the resources required to run larger, more power-hungry systems.Proven AI Software Stack Powered by CUDAProject Digits is designed to seamlessly integrate into Nvidias extensive AI ecosystem, providing developers with a cohesive and efficient environment for AI development. Based on the Linux-based Nvidia DGX OS, it ensures a stable and robust platform tailored for high-performance computing tasks. Preloaded with Nvidias comprehensive AI software stack, including the Nvidia AI Enterprise software platform, Project Digits offers immediate access to a wide array of familiar tools and frameworks essential for AI research and development.The systems compatibility with widely used AI frameworks and tools, such as PyTorch, Python and Jupyter Notebooks, empowers developers to use familiar environments for model development and experimentation. Additionally, it supports the Nvidia NeMo framework, which enables the fine-tuning of large language models, and the RAPIDS libraries, which accelerate data science workflows.In terms of connectivity and scalability, Project Digits employs Nvidia ConnectX networking, enabling high-speed data transfer and efficient communication between systems. This feature allows two Project Digits units to be interconnected, effectively doubling the capacity to handle models with up to 405 billion parameters. This scalability ensures that as complex AI models grow, Project Digits can adapt to meet the increasing computational demands.Moreover, Project Digits is engineered to seamlessly integrate cloud and data center infrastructures. Thanks to the consistent architecture and software platforms across Nvidias ecosystem, developers can prototype and fine-tune AI models locally on the device and then deploy them to larger-scale environments without compatibility issues. This flexibility streamlines the transition from development to production, enhancing efficiency and reducing time to deployment.Nvidia Project Digits comes with the below software stack:Linux-based Nvidia DGX OS: Runs on a robust, Linux-based operating system optimized for AI workloads, ensuring stability and performance.Pre-installed Nvidia AI Software Stack: Provides immediate access to Nvidias extensive AI tools and frameworks, streamlining the development process.Nvidia AI Enterprise: Delivers a suite of AI and data analytics software, ensuring enterprise-grade support and security for AI workflows.Nvidia NGC Catalog: Offers a rich repository of software development kits, frameworks, and pre-trained models, facilitating efficient AI model development and deployment.Nvidia NeMo Framework: Enables fine-tuning and deploying large language models, supporting advanced natural language processing tasks.Nvidia RAPIDS Libraries: Accelerates data science workflows by leveraging GPU-optimized libraries for data processing and machine learning.Support for Popular AI Frameworks: Compatible with widely used tools such as PyTorch, Python and Jupyter Notebooks, allowing developers to work within familiar environments.With its petaflop of AI performance and integrated software stack, Project Digits can handle massive AI models 200 billion parameters. This capability was previously limited to large-scale supercomputers, but Digits brings this power to the desktop, enabling developers to experiment with and deploy cutting-edge AI models locally.Ecosystem Support for Project DigitsNvidias CEO Jensen Huang shed light on the development of Digits and a key partnership that played a crucial role in its creation. He highlighted the collaboration with MediaTek, a leading fabless semiconductor company, in designing an energy-efficient CPU specifically for Digits. This partnership allowed Nvidia to leverage MediaTeks expertise in low-power CPU design, contributing to Digits impressive power efficiency.Huang also emphasized how Digits bridges the gap between Linux and Windows environments. While Digits itself runs on a Linux-based operating system, it is designed to seamlessly integrate with Windows PCs through the Windows Subsystem for Linux technology. This allows developers primarily working in Windows environments to easily utilize the power of Digits for their AI projects.Target Audience and Use CasesNvidia Digits is specifically designed for AI researchers, data scientists, students and developers working with large AI models. It empowers these users to prototype, fine-tune and run AI models locally, in the cloud, or within a data center, providing flexibility and control over their AI development workflows.By bringing the power of an AI supercomputer to the desktop, Digits addresses the growing need for greater AI performance in a compact and accessible form factor. This allows individuals and smaller teams to tackle complex AI challenges without relying on expensive cloud computing resources or large-scale supercomputing infrastructure.Potential use cases for Digits are vast and varied. Developers can use it to prototype new AI applications, fine-tune LLMs for specific tasks, generate AI-powered content, and research new AI algorithms and architectures. The ability to run large AI models locally opens up new possibilities for AI development, enabling faster iteration and experimentation.Nvidia Digits is expected to be available in May from Nvidia and its partners, with a starting price of $3,000. This competitive pricing makes it a viable option for a broader range of users, further contributing to the democratization of AI development.ConclusionNvidia Digits represents a significant leap forward in AI technology. By combining powerful hardware, a comprehensive software stack and a compact, power-efficient design, Digits brings the capabilities of an AI supercomputer to the desktop. This can potentially democratize AI development, making it more accessible to individuals, researchers and smaller organizations. The ability to run large AI models locally and the flexibility to deploy in the cloud or data center provides developers unprecedented control and power over their AI workflows. As Digits becomes available, witnessing the innovative applications and advancements that emerge from this powerful new platform will be exciting.

LAS VEGAS, NV - MARCH 8: Dana White attends the Power Slap 1 pre-event press conference on March 8, ... [+] 2023, at the UFC APEX in Las Vegas, NV. (Photo by Amy Kaplan/Icon Sportswire via Getty Images)Icon Sportswire via Getty ImagesThe UFC 311 card has taken a hit, and so has Johnny Walkers stock as a light heavyweight contender.The light heavyweight scrap between Walker and Bogdan Guskov is off. Walker suffered what he described as a bad injury that is forcing him out of the fight.Thankfully for Guskov, the UFC has moved quickly to get the hard-charging light heavyweight another opponent. Walker will be replaced by Billy Elekana, a 29-year-old with a 7-1 record.SAN ANTONIO, TX - AUGUST 04: Billy Elekana looks on before fighting Chuck Campbell during PFL 2023 ... [+] week 7 at Boeing Center at Tech Port on August 4, 2023 in San Antonio, Texas. (Photo by Cooper Neill/Getty Images)Getty ImagesElekana is the rare fighter making his UFC debut who didnt earn a spot on the roster through the Dana White Contender Series. If Elekana performs well, he should get an opportunity to join the promotion officially.Guskov is ranked No. 13, so while a tall task, its a great opportunity for Elekana. If anyone is getting a bit of a bad hand, its Guskov. Fighting Walker was an opportunity to move up in the rankings for Guskov. Walker is currently No. 10, but his stock seems to be falling.Injuries can happen to anyone, but the current situation with Walker is the latest setback since he erupted on the UFC scene in August 2018.The 6-foot-6 uber-athletic Brazilian defeated Luis Henrique da Silva on DWCS and then rolled off three straight first-round stoppage wins over Khalil Rountree Jr., Justin Ledet, and Misha Cirkunov.Read More: UFC Vegas 101 Results, Bonus Winners, Highlights And ReactionWalker was being mentioned as the guy who could ultimately dethrone Jon Jones. Yes, it got that serious.After the win over Cirkunov at UFC 235, Walker dislocated his shoulder doing the Worm as a wild celebration. Quite honestly, he hasnt been the same since the mishap.LAS VEGAS, NV - MARCH 02: Johnny Walker of Brazil celebrates his victory over Misha Cirkunov of ... [+] Latvia in their light heavyweight bout during the UFC 235 event at T-Mobile Arena on March 2, 2019 in Las Vegas, Nevada. (Photo by Jeff Bottari/Zuffa LLC/Zuffa LLC)Zuffa LLCSince beating Cirkunov and injuring his shoulder, Walker is 3-6-1. He was promptly stopped by Corey Anderson in his next fight and then lost a unanimous decision to Nikita Krylov.Walker got back on the winning track by defeating Ryan Spann in September 2020, but he went on another two-fight losing streak when he was defeated by Thiago Santos and Jamahal Hill in 2021 and early 2022.It seemed like Walker might be back on track as he reeled off wins over Ion Cutelaba, Paul Craig, and Anthony Smith, but he hit his ceiling in 2023 when he fought Magomed Ankalaev.In a fight Ankalaev looked set to win, Walker said he was unable to continue after being hit with an accidental illegal knee.In the rematch, Ankalaev made quick work of Walker with a second-round KO. Walker met the same fate against Volkan Oezdemir in June 2024.Gone are the thoughts that Walker can be a serious championship contender, and thats crazy considering hes only 32 years old. The recent injury only makes the fading shot at contention all the more unlikely.Is Walker a bust?CHARLOTTE, NORTH CAROLINA - MAY 13: Johnny Walker of Brazil has his hands wrapped backstage during ... [+] the UFC Fight Night event at Spectrum Center on May 13, 2023 in Charlotte, North Carolina. (Photo by Mike Roach/Zuffa LLC via Getty Images)Zuffa LLC via Getty ImagesIf that term can be applied in mixed martial arts, he probably fits the definition as much as anyone in recent memory.Hopefully, Walker can rebound and make a run in his mid-30s. Perhaps he can make some noise at heavyweight. At 66, its not out of the question.

Two years ago, I got a new phone number. In the eyes of my social media and ride-hailing apps, Amazon, my bank, and the state of Pennsylvania, that effectively meant I lost my identity. Only recently have I emerged from this technology-induced quagmire.Sitting at my desk with a new smartphone in hand, I slowly rotated my head up and down while staring into the front-facing camera. Then, without breaking eye contact with the lens, I carefully turned my head from one ear to the other. The short clip was supposed to prove to Instagram's security system what I had been insisting for weeks: that I was not an imposter that I was, in fact, me."Thank you for your selfie video," the automated email from Instagram said. "We received this information and it's pending review."I not only had forgotten my Instagram password but also no longer had access to my old phone number: Since I'd last logged in, I'd switched to a money-saving family plan with a new mobile provider. To get past Instagram's two-factor authentication security where the app verifies your identity by texting you a secret code and regain access to 12 years of cat photos and filtered city skylines, I had to upload a video of my face.As I awaited Instagram's response, I imagined somebody from the company's security team watching my selfie video, eyeballing the pictures I'd posted to my account over the years, and confirming my identity once and for all. A few minutes later far too quickly for a real person to be behind it an email arrived. "Your Information Couldn't Be Confirmed," the subject line said. My information? You mean, my face?"We weren't able to confirm your identity from the video you submitted," the message said. "You can submit a new video and we'll review it again."As fun as this game sounded, I had other things to take care of.I figured that suddenly switching phone numbers would result in some headaches, but I didn't foresee how complicated it would actually become. From Lyft and Cash App to Instagram and Amazon, I suddenly had to jump through logistical hoops that varied from one platform to the next to verify my flesh-and-blood identity and regain access to my digital life. Confirming my identity became a part-time job. And as I learned the hard way, fellow human beings who could help me sort things out were hard to come by.While the origin of text-message-based two-factor authentication goes back as far as the 1990s, it wasn't until the early 2010s that it really started to proliferate. As more people bought smartphones, it seemed convenient to use a person's phone number as a way to confirm their identity. At the same time, the growing frequency and sophistication of data breaches and cyberattacks rendered traditional passwords nearly useless, prompting then-President Barack Obama to write an op-ed in The Wall Street Journal in 2016 imploring citizens to "move beyond passwords" and embrace additional layers of security to protect their data. Even in 2024, hackable passwords like "1234" and "password" remained alarmingly common.By requiring people to take an extra step (or two) to verify themselves, multifactor authentication offers a massive security improvement over the traditional password. Thanks to remote work, MFA adoption is on the rise: 64% of people using Okta use some form of multifactor authentication, a 2023 report by the company found. Before the pandemic, it was 35%.Despite its benefits, the system developed a serious flaw somewhere along the way: SMS-based two-factor authentication, which relies on calling or texting someone's phone to verify their identity. Unlike MFA methods that rely on an authenticator app, text-based authentication is arguably the least secure way to verify someone's identity. Unfortunately, it's also one of the most common."We often see that less mature organizations have standardized on using that SMS-based code," said Cristian Rodriguez, the field chief technical officer for the Americas at the cybersecurity firm CrowdStrike. Apple, Google, Zoom, Slack, Dropbox, PayPal, and most major US banks and universities are on the long list of sites that still use it.Unlike our fingerprints or faces, our phone number is not a permanent feature of our identity."It's also easy to intercept," Rodriguez said of the method. "SIM swapping is a really easy way to circumvent that as an attacker."In a SIM-swapping attack, a hacker can gain control of someone's phone number and wreak havoc on their lives. From their bank and social media accounts to credit cards stored in digital wallets like Apple Pay, the amount of access a cybercriminal can gain is staggering. It shows just how thoroughly our lives are tied to our phone numbers.Recently, hackers aligned with the Chinese government managed to gain access to US phones via their telecommunication networks in a massive hack dubbed Salt Typhoon, which the federal government says is among the worst in the nation's history. While the exact origins and overall impact of Salt Typhoon are still being investigated, experts say that the two-year infiltration could have affected millions of Americans. In the wake of the attack, which was first reported in October, the feds are advising people to stop using SMS-based authentication.Unlike our fingerprints or faces, our phone number is not a permanent feature of our identity. It's just a sequence of digits randomly assigned to us by a cell provider when we sign up. If we stop paying, switch mobile providers, or move to a new country, the number is no longer ours. Even an email address in my case, a 20-year-old Gmail account that I'm certain will continue receiving marketing promotions long after I'm dead would make a more reliable long-term indicator of who I am than my phone number. After all, my email address will never be reassigned. It's mine. My phone number along with the roughly 35 million numbers that get reassigned each year, according to the Federal Communications Commission is another story.Few companies seem to recognize the problem: In early 2023, X decided to end support for SMS-based two-factor authentication for nonverified users, citing its weakness. Following major breaches, tech giants like Google and Microsoft have also begun to make moves away from SMS authentication. Still, X stands alone among major social media platforms in abandoning it altogether a fact that I had to learn the hard way.In most cases, regaining access to my accounts was simple. My bank, for instance, just required a quick phone call to a customer service agent to confirm my identity, bypass MFA, update my phone number, and reset my password.Instagram, an operation notably larger and more resourced than the small credit union I bank with, did not offer a customer service hotline. Instead, after about half a dozen clicks from Instagram's login screen, I found my way into the depths of a customer service FAQ page that suggested I send a video selfie. I had to send several videos into the automated void before the app caved and let me back into my account.My Amazon account, which includes services like Audible, Alexa, and Whole Foods shopping, proved a surprisingly impenetrable fortress. After clicking my way through a labyrinth of links, I eventually reached a prompt that asked me to upload a photo of my passport to verify my identity. After not hearing anything for a week, I tried again. A few months later, Amazon let me back in. (Instagram and Amazon did not respond to multiple requests for comment.)Thanks to the ongoing AI-ification of customer service and the growing use of chatbots, the privilege of speaking with a real person is increasingly rare. To this day, I'm unable to verify my LinkedIn profile. Even after asking for my current phone number, the identity-verification platform used by LinkedIn, Clear, continues to text a six-digit code to my old phone number, which it somehow summons from the cloud. Of course, there's no one to speak with to resolve the issue. Oh well.If our tech is this advanced, why is it still so hard to convince the machines it's actually us?Getting locked out of social networks and online shops is extremely annoying. But it's nothing compared to the nightmare I suffered through when I tried to sign up for unemployment benefits after getting laid off from my journalism job.The state government of Pennsylvania uses the aptly named ID.me service to handle its website login and identity verification for services like unemployment. On its website, ID.me boasts integrations with 19 federal agencies, 35 healthcare-related organizations, and more than 600 online stores in its mission to empower consumers with "a single log-in that lets you easily prove you're you." There's just one problem: ID.me ties your digital identity to your phone number.When I tried logging in to the Pennsylvania unemployment site to claim my benefits, I discovered I already had an ID.me account presumably from some other government site I had previously accessed. That account, of course, was linked to my old phone number.Without access to that number and having no clue what my password might have been my only option was to submit a request to bypass MFA and regain access to my ID.me account through the company's customer service help desk.After I got an automated confirmation email from Roy, the self-described "virtual agent" at ID.me, my request went ignored for 48 hours. After multiple follow-ups and a great deal of patience, I was finally able to set up a phone screening with a real human. Ten days after my initial request, I was on the phone with a customer service agent who emailed me some instructions: Fill out a few forms and send them back along with digital scans of my passport and Social Security card. Normally, sending such sensitive information over email would give me pause, but in this case, the ID.me customer service agent was holding my identity and my financial security hostage, so I was inclined to do whatever he asked. I could only hope that ID.me's cybersecurity practices were more robust than its customer service operation.After 30 days of submitting and resubmitting various identifying documents, I was finally able to log in to my account. By then, I had taken care of my unemployment needs the old-fashioned way: filing my application manually over the phone a process that took nearly three hours over the course of two excruciatingly tedious phone calls. (ID.me did not respond to a request for comment.)Technology that used to promise to simplify our lives now seems to make everything more complicated. Before automated self-checkouts entered the grocery store, the checkout process was never interrupted by a confused machine thinking you forgot to scan an item, forcing you to wait for a human to help. For all the innovative "smart" technology that has wormed its way into modern cars, it often feels like we're one software glitch away from getting locked out of them, too.It's 2025 we have artificial intelligence that's capable of forming romantic relationships. I should be able to unlock my laptop with my thumbprint or scan my face to quickly access my online shopping accounts. If our tech is this advanced, why is it still so hard to convince the machines it's actually us?You don't have to take my word for it. You can ask Keith, the person who had my new phone number before me. I have no idea who Keith is, but I know he, too, is having trouble getting back into his websites and apps because I keep getting texts containing six-digit verification codes that I never asked for. While I eventually got back into most of my accounts, it seems that Keith is still struggling.Keith, if you're out there: Your prescription is ready to pick up at Rite Aid.John Paul Titlow is a freelance journalist who writes about technology, digital culture, travel, and mental health.