Artificial intelligence has made significant strides in recent years, but challenges remAIn in balancing computational efficiency and versatility. State-of-the-art multimodal models, such as GPT-4, often require substantial computational resources, limiting their use to high-end servers. This creates accessibility barriers and leaves edge devices like smartphones and tablets unable to leverage such technologies effectively. Additionally, real-time processing for tasks like video analysis or speech-to-text conversion continues to face technical hurdles, further highlighting the need for efficient, flexible AI models that can function seamlessly on limited hardware.OpenBMB Releases MiniCPM-o 2.6: A Flexible Multimodal ModelOpenBMBs MiniCPM-o 2.6 addresses these challenges with its 8-billion-parameter architecture. This model offers comprehensive multimodal capabilities, supporting vision, speech, and language processing while running efficiently on edge devices such as smartphones, tablets, and iPads. MiniCPM-o 2.6 incorporates a modular design with:SigLip-400M for visual understanding.Whisper-300M for multilingual speech processing.ChatTTS-200M for conversational capabilities.Qwen2.5-7B for advanced text comprehension.The model achieves a 70.2 average score on the OpenCompass benchmark, outperforming GPT-4V on visual tasks. Its multilingual support and ability to function on consumer-grade devices make it a practical choice for diverse applications.Technical Details and BenefitsMiniCPM-o 2.6 integrates advanced technologies into a compact and efficient framework:Parameter Optimization: Despite its size, the model is optimized for edge devices through frameworks like llama.cpp and vLLM, maintaining accuracy while minimizing resource demands.Multimodal Processing: It processes images up to 1.8 million pixels (13441344 resolution) and includes OCR capabilities that lead benchmarks like OCRBench.Streaming Support: The model supports continuous video and audio processing, enabling real-time applications like surveillance and live broadcasting.Speech Features: It offers bilingual speech understanding, voice cloning, and emotion control, facilitating natural, real-time interactions.Ease of Integration: Compatibility with platforms like Gradio simplifies deployment, and its commercial-friendly nature supports applications with fewer than one million daily active users.These features make MiniCPM-o 2.6 accessible to developers and businesses, enabling them to deploy sophisticated AI solutions without relying on extensive infrastructure.Performance Insights and Real-World ApplicationsMiniCPM-o 2.6 has delivered notable performance results:Visual Tasks: Outperforming GPT-4V on OpenCompass with a 70.2 average score underscores its capability in visual reasoning.Speech Processing: Real-time English/Chinese conversation, emotion control, and voice cloning provide advanced natural language interaction capabilities.Multimodal Efficiency: Continuous video/audio processing supports use cases such as live translation and interactive learning tools.OCR Excellence: High-resolution processing ensures accurate document digitization and other OCR tasks.These capabilities can impact industries ranging from education to healthcare. For example, real-time speech and emotion recognition could enhance accessibility tools, while its video and audio processing enable new opportunities in content creation and media.ConclusionMiniCPM-o 2.6 represents a significant development in AI technology, addressing long-standing challenges of resource-intensive models and edge-device compatibility. By combining advanced multimodal capabilities with efficient operation on consumer-grade devices, OpenBMB has created a model that is both powerful and accessible. As AI becomes increasingly integral to daily life, MiniCPM-o 2.6 highlights how innovation can bridge the gap between performance and practicality, empowering developers and users across industries to leverage cutting-edge technology effectively.Check out the Model on Hugging Face. 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.(Promoted) Asif RazzaqAsif Razzaq is the CEO of Marktechpost Media Inc.. As a visionary entrepreneur and engineer, Asif is committed to harnessing the potential of Artificial Intelligence for social good. His most recent endeavor is the launch of an Artificial Intelligence Media Platform, Marktechpost, which stands out for its in-depth coverage of machine learning and deep learning news that is both technically sound and easily understandable by a wide audience. The platform boasts of over 2 million monthly views, illustrating its popularity among audiences. Meet 'Height':The only autonomous project management tool (Sponsored)

Author(s): Anuar Sharafudinov Originally published on Towards AI. Credits: GPT4oThe rise of the Retrieval-Augmented Generation (RAG) has revolutionized how we build intelligent applications. At its core, RAG is all about efficiently turning large chunks of text into actionable embeddings and then letting an AI model piece together contextually relevant answers.However, what works in theory can stumble in real-world scenarios. Why? One of the biggest culprits is poor or unclear embedding representations. Often, these representations dont align well with the demands of production-level applications particularly for tasks like question-answering.The solution is fine-tuning embeddings an impactful way to enhance your RAG implementation.RAG 101: How It WorksLets break it down. Heres the typical RAG workflow:User Input: A user submits a question or query.Query Embedding: The system generates an embedding for the query.Chunk Matching: It searches for the chunk embeddings most similar to the query using cosine similarity.Answer Generation: The contents of the retrieved top chunks are sent as context to a language model, which generates the final response.This setup works well in theory. However, when embeddings lack precision, the results can feel off-target, especially when dealing with large datasets.The Fine-Tuning SolutionWhat if you could pre-train your embeddings to anticipate the kinds of questions your users might ask?Heres the idea:Generate Question-Chunk Pairs: For each chunk of text in your dataset, generate multiple potential questions it could answer.Fine-Tune the Embedding Model: Train the model to pull embeddings of related questions and chunks closer together in multidimensional space while pushing unrelated ones further apart.While this approach might seem like overfitting, it actually focuses on optimizing for generalization. It turns out, fine-tuning embeddings in this way equips the system to handle unseen queries with improved accuracy.The Results Speak for ThemselvesFine-tuning embeddings yielded remarkable improvements across several models. For training, we used one of our internal experimental datasets. It consists of 52 chunks, each approximately 800 tokens long. For each chunk, we used Anthropics Claude-3-Sonnet to generate 35 corresponding questions.To evaluate performance, we measured how often the correct chunk appeared within the top 3, top 5, and top 10 retrieved results. To provide a broader context, we also included results for OpenAI/text-embedding-large-3. However, since it is a closed-source model, we could not apply fine-tuning to it.Heres a snapshot of the results:Open-Sourcing the CodeIf youre inspired to experiment with fine-tuning, weve got you covered. Check out our code repository with training and testing scripts for Alibaba-NLP/gte-Qwen21.5B-instruct and jinaai/jina-embeddings-v3 models. The repo also includes support for two training methods: TripletMarginLoss and CosineEmbeddingLoss.Model requirementsAlibaba-NLP/gte-Qwen21.5B-instruct Requires about 30GB of VRAM. A GPU with 40GB of memory and higher (e.g., A100) is recommended. Its forward pass logic is standard and can be applied to many similar embedding models.jinaai/jina-embeddings-v3 is a very lightweight model requiring only 8GB of GPU memory for fine-tuning. Its forward-pass logic is slightly specific, but the core concept is clear.Training methodsTripletMarginLoss. This method uses an anchor (a), a positive sample (p), and a negative sample (n):Anchor (a): Chunk content embeddingPositive sample (p): A corresponding question embeddingNegative sample (n): An unrelated question embeddingThe loss function illustrationTo build a training set, create (chunk, questions) pairs and randomly select unrelated questions as negative samples.2. CosineEmbeddingLoss. This method uses positive and negative samples from different parts of the training set:x1: The chunk embeddingx2: Either a positive or negative sample embeddingy: Label indicating if x2 is positive (y=1) or negative (y=-1).The loss function illustrationAdapting the CodeTo use your own dataset, modify the prepare_data function in train.py. Ensure it returns chunks and their corresponding questions as pairs.Note: The repository does not include question generation logic, but various approaches are available. Below, weve included a sample code that we used for reference.#1. split the document into chunks (simple way)def split_into_chunks(content, chunk_size): import tiktoken enc = tiktoken.get_encoding("o200k_base") a = enc.encode(content) left, chunks = 0, [] while left < len(a): arr = a[left : left+chunk_size] chunks.append(enc.decode(arr)) left+=chunk_size return chunkschunks = split_into_chunks(document_content, 400)#2. generate questions def anthropic_run(system_prompt, user_message): import anthropic client = anthropic.Anthropic( api_key=ANTHROPIC_API_KEY, ) message = client.messages.create( model="claude-3-sonnet-20240229", #"claude-3-opus-20240229", max_tokens=4096, system=system_prompt, messages=[ {"role": "user", "content": user_message} ] ) return message.content[0].text system_prompt = ''' Given a chunk from document. Generate 3-5 questions related to the chunk. Each question must be full and not require additional context. Example output: 1. How to open new account? 2. How much BMW X5 costs? ''' for chunk in chunks: text = "#"+chunk["keywords"]+"\n"+chunk["content"] out = anthropic_run(system_prompt, text) question_pattern = re.compile(r'^\s*\d+\.\s+(.*)', re.MULTILINE) questions = question_pattern.findall(out) print(text, questions)#now you have (chunk, questions) pairsContinuous ImprovementAnother advantage of this approach is its potential for continuous improvement. Over time, as new cases emerge, you can retrain the model. For instance, if a corresponding chunk wasnt found in the top 10 (avoid large numbers to avoid the lost in the middle issue) and the LLM failed to generate an answer, simply add this question and its correct chunk to the training set. This ensures the system evolves to handle similar issues in the future more effectively.Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming asponsor. Published via Towards AI

Bethesda originally planned to include gore and dismemberment mechanics in Starfield but had to remove them due to technical limitations.Former employee Dennis Mejillones, who was a character artist on The Elder Scrolls 5: Skyrim, Fallout 4, and Starfield, told Kiwi Talkz that Bethesda had to cut the feature because the interaction with space suits became too complex."There was a lot of implications with the different suits from a technical perspective," he said. "There's a lot that has to go with it. You have to cut the helmet in a certain way and it's got to come off, you have meat caps to the bottom where the flesh is."We had systems for all of that and it turned into a big rat's nest. All these things you have to count for now with all these crazy hoses on the helmets and all that kind of stuff that we added. Or now you could change the body size significantly. The character creator had evolved quite a bit."Some fans lamented that Starfield, which was the first full single-player role-playing game from Bethesda in eight years, didn't have the gore and dismemberment mechanics that were present in Fallout 4. Mejillones said these mechanics make more sense in Fallout than in Starfield, however, given their "tongue in cheek" humor. "It's part of the fun," he said.Starfield arrived in September 2023 and in the time since has reached more than 15 million players. "Starfield has a lot of forces working against it, but eventually the allure of its expansive roleplaying quests and respectable combat make its gravitational pull difficult to resist," IGN said in our 7/10 review.Last month, another former Bethesda developer revealed his surprise at the sheer amount of loading Starfield ended up launching with, particularly in the city of Neon. Since launch, Bethesda has worked to improve the game, with 60fps now possible as part of performance mode. Expansion Shattered Space launched in September.Ryan Dinsdale is an IGN freelance reporter. He'll talk about The Witcher all day.

Having too much space sounds like the kind of problem we all wish we had. But in the case of this 17,200-square-foot house in Westlake, Texas, a suburb of Dallas, the struggle of how to make each room feel inviting was real. "Homes of this size can almost feel commercial," says Janelle Patton, founder of Lark Interiors, who began working with the homeowners shortly after they bought the empty lot. "We wanted this to be the right scale for a family."Together with architectural lead Alison Ames, Patton designed a home with seven bedrooms, nine bathrooms, and an entertaining wing, but almost no hallways. "When rooms open onto each other and you walk through them, it feels cozy," explains Patton. The bedrooms are also close to each other, so a kid who wakes up in the middle of the night doesn't have far to go to find their parents.The ceilings were another conundrum; in the living room alone they are over 34 feet high. To avoid cavernous expanses of drywall, Patton and Ames installed wooden beams in the living room and primary bedroom and added coffered ceilings and 10-inch crown moldings elsewhere. Patton also carved out intimate spaces within rooms: a niche under the stairs for putting on shoes, a loft with a slide into a secret room, and a homework nook in the kids' playroom.They were willing to invest a little more in kid-friendly things that last.The homeowners have big plans for entertaining, so Patton and Ames designed "the lounge," a complex of different hangout spaces that includes a den with plenty of seating, a pool table, and a bar. Nearby is a golf simulator, a screening room, and a basketball court (okay, just a half-sized one).To make the disparate parts of the home feel cohesive, certain shapes repeat. Rounded corbels show up on the breakfast nook's framing, the floor pattern in the dining room, the bathroom vanities, and toe-kicks and cabinets in the kitchen. And throughout, Patton worked from a color palette of greens and blues balanced with mustard and terra-cotta.The homeowners were thrilled with the marriage of their distinct sensibilities. "His tastes are more modern, while she loves detailed, English-inspired elements," says Patton. "It's tongue-in-cheek, but we ended up calling the home's style 'modern English cottage...mansion.'" A little somethingand space enoughfor everyone.Living RoomStacy Zarin GoldbergAlmost every piece of furniture in this room is a custom commission. "We just needed a different scale than what was available," says Janelle Patton of Lark Interiors. Pillow fabrics: Lulie Wallace, Maresca Textiles, Virginia White, Studio Four NYC, Lee Jofa.Prep KitchenStacy Zarin GoldbergStacy Zarin GoldbergThis room is the house's informal entry point, with a Dutch door that leads out to the driveway. Chandelier: Visual Comfort & Co. Paint: October Mist (walls), Dark Olive (cabinetry), Benjamin Moore.KitchenStacy Zarin GoldbergThe island that sits opposite the Lacanche range houses a four-foot-wide sink, dishwasher, and an undercounter microwave, while the other is designed for seating and storage. Stools: Cuff Studio.Breakfast NookStacy Zarin GoldbergEvery surface of this high-traffic dining space is wipeable, including a Pollack vinyl fabric on the custom white oak bench "that doesn't feel like vinyl at all." The owners invested in performance fabrics throughout the home. Table and chairs: custom. Chandelier: The Urban Electric Co.Dining RoomStacy Zarin GoldbergThe 18th-century cabinet, sourced from an antique dealer in L.A., was once part of a shop display in England. Chandeliers: Visual Comfort & Co. Armchairs: Century Furniture, in Morris & Co. fabric.BarStacy Zarin GoldbergThe bar is framed in white oak with a custom wine rack that helps delineate it as a discrete space within the lounge. Pendants: Lucent Lightshop. Cabinet paint: Narragansett Green, Benjamin Moore. Stools: Century Furniture. Primary BedroomStacy Zarin GoldbergNobody wanted to cover up the white oak flooring, so Patton chose a small rug as "something soft underfoot when they step out of bed." Bed: Made Goods. Chairs: Century Furniture, in Parker & Jules fabric. Nightstand: Vanguard Furniture.Bunk RoomStacy Zarin Goldberg The family likes to host holidays, and requested this guest room designed for kids so that "all the cousins can come," says Patton.Boy's RoomStacy Zarin GoldbergStacy Zarin GoldbergThe structure accommodates toy storage and a custom "big boy bed." Behind it is a partially concealed play loft with a sleeping area of its own. Bed upholstery: Kelly Ventura. Paint: Avalon Teal (bed frame, stairs), Benjamin Moore and Tidewater (cabinetry), Sherwin-Williams. Primary BathroomStacy Zarin GoldbergStacy Zarin GoldbergWood beams and wallpaper define a distinct space for the tub. The fluted design of the custom vanities is another design detail that repeats throughout the house. Wallcovering: Sandberg Wallpaper. Sconces: The Urban Electric Co. Tub: Signature Hardware. Daughter's BathroomStacy Zarin GoldbergThe scalloped detail on the vanity is repeated in other rooms, linking this space with what the designers call the "English cottage aesthetic" of the rest of the home. Layered tile that acts as wainscoting on the wall carries through to the shower. Vanity paint: Pelt No. 254, Farrow & Ball. PlayroomStacy Zarin GoldbergInside the loft area is a slide that leads to a secret room Patton calls "the cool kids club." Mural: Natswood Art from Etsy. Ladder paint: Indigo Batik, Sherwin-Williams.Stacy Zarin GoldbergThe playroom's built-in bookshelves echo the gabled ceiling. The modular sofa is custom from Vanguard Furniture. All the pieces are movable and interchangeable. Basketball CourtStacy Zarin GoldbergA bump-out at the back of the court was added so the homes exterior wouldn't be an ungainly rectangle, and is now the perfect nook for red bleachers. Pendants: Visual Comfort & Co. Water fountain: Kohler. Fan: Big Ass Fans.If You Want to Build a Basketball Court...Tip 1: Find your inspo. There are no color rules, so look to a court you love. This one is inspired by the YMCA, where the owner played as a kid.Tip 2: Choose maple floors. Its the classic option for indoor basketball courts because woods elasticity is good for bounceand produces that satisfying hollow sound you get when you dribble.Tip 3: Open up the ceiling. Exposed ductwork and beams mimic a real gym. Just make sure you hang fans higher than the lights to avoid a strobe light effect.Follow House Beautiful on Instagram and TikTok.

Coming from South Africa but living in Europe, I can tell you that sending money to family and friends back home is a bit of a nightmare. Typically you must use a traditional bank,which can take a week or more, or payment apps like PayPal or Wise, which charge high fees.The antiquated nature of remittance payments is something that immigrants are all too familiar with. Demand for better alternatives is giving rise to a new cohort of fintech companies looking to streamline the process. One of them is London-headquartered LemFi.Founded in 2021, the financial services platform enables diaspora communities in North America and Europe to quickly and affordably send money to friends and family in China, India, Pakistan, Nigeria, Kenya, and 15 other countries in the Global South. Sadly, for me, South Africa is not yet on the list. However, LemFi is expanding fast so I might not have very long to wait.LemFi has already onboarded 1 million customers so far, who have made a combined $1bn in monthly transactions through the app. Transactions to and from Asia are currently growing at 30% month-on-month, said the company. And last week, LemFi, which employs over 300 people, officially set up shop in Europe. The startup is tapping a global remittance market predicted to reach $1.3 trillion by 2032.All that growth potential, has, unsurprisingly, piqued the interest of VCs. Today, LemFi announced that it has raised $53mn in Series B funding. London-based growth-stage venture firm Highland Europe led the round, with participation from previous investors Left Lane Capital, Palm Drive Capital, and Y-Combinator. The fresh funding brings LemFis total raised to $85mn.When we started building LemFi, we were told remittance had already been solved, said Ridwan Olalere, co-founder and CEO of LemFi, pictured left. But for too many people, it is still too slow, cumbersome and expensive with customers telling us that in some instances it was cheaper to send money from the US via Canada than directly to their families back home.Olalere, originally from Nigeria, founded LemFi alongside Norwegian Rian Cochran. The pair met whilst working at Nigerian fintech unicorn OPay, incubated by Norwegian browser provider Opera.Fuelled by fresh funding, LemFi now looks to add new features and expand to new countries, said Olalere.No doubt part of LemFis popularity is that it advertises zero transaction fees. Well, unless you live in China, India, or Pakistan. The company also makes revenue on foreign currency exchanges. Its business model is dependent on volume, making small profits on numerous transactions while staying competitive for users seeking low-cost international money transfers.LemFi is far from alone in the increasingly crowded remittance fintech market. Perhaps the most well-known is US-based Remitly, which went public in 2021. Other contenders include startups Zepz and Taptap Send, both of which are headquartered in London. Story by Sin Geschwindt Sin is a climate and energy reporter at TNW. From nuclear fusion to escooters, he covers the length and breadth of Europe's clean tech ecos (show all) Sin is a climate and energy reporter at TNW. From nuclear fusion to escooters, he covers the length and breadth of Europe's clean tech ecosystem. He's happiest sourcing a scoop, investigating the impact of emerging technologies, and even putting them to the test. Sin has five years journalism experience and holds a dual degree in media and environmental science from the University of Cape Town, South Africa. Get the TNW newsletterGet the most important tech news in your inbox each week.Also tagged with

OpenAI on Tuesday announced a new feature coming to ChatGPT, which will turn the chatbot into a reminder app. Called Tasks, the feature will let users ask ChatGPT to remind them about their tasks and even create periodic alerts for specific situations.ChatGPT now lets users create reminders with TasksOpenAI describes the Tasks feature as an important step toward making it more of a helpful AI companion that can take on tasks on your behalf.With this feature, ChatGPT can send you reminders when you need them, whether they are one-time or recurring. With the feature enabled, all you have to do is tell ChatGPT what you want and when, and it will create the task. There will be a dedicated section for managing tasks in the profile menu within ChatGPT. The chatbot will also suggest tasks based on your chats.As an example, users can ask ChatGPT to remind them that their passport will expire in 6 months. It also works with more complex requests, such as to get a daily weather report every morning or a list of things to do every Friday. ChatGPT will send push notifications via its website or app. Depending on what you tell the app, it will also create a task automatically (if you allow it to).The new Tasks feature is now being rolled out as a beta to ChatGPT Plus, Team, and Pro subscribers. It will be available across all platforms via the new 4o with scheduled tasks model. During the beta, users can set only 10 active tasks at a time. Push notifications will also work through ChatGPTs macOS app.Right now, well use this beta period to understand how people use tasks and refine the feature before making it available to all ChatGPT users, OpenAI told 9to5Mac.In December, OpenAI made ChatGPT Search available to everyone for free. The feature essentially turns ChatGPT into a search engine that can find answers on the web in real time. The macOS app was also recently updated to work with Apple Notes and more third-party apps.The ChatGPT app is available for free on the iOS App Store. The Mac version must be downloaded from the OpenAIs website.Add 9to5Mac to your Google News feed. FTC: We use income earning auto affiliate links. More.Youre reading 9to5Mac experts who break news about Apple and its surrounding ecosystem, day after day. Be sure to check out our homepage for all the latest news, and follow 9to5Mac on Twitter, Facebook, and LinkedIn to stay in the loop. Dont know where to start? Check out our exclusive stories, reviews, how-tos, and subscribe to our YouTube channel

Erstwhile TikTok users are flocking en masse to a Chinese social media app called RedNote ahead of a Supreme Court ruling which could force a sale or total ban of the beloved short-form video app. And it's not a fringe thing; RedNote has quickly hit #1 on Apple and Android app stores, with the influx of users showing no signs of slowing.The app, known in China as Xiaohongshu or "xhs" for short, sports an interface not dissimilar from TikTok, but is functionally closer to a hybrid of Pinterest and Instagram. Users trade recipes in static posts, comment on fit check reels, and use the AI-free search function to answer everyday questions. (US tech companies might learn a thing or two from that example.)On a certain level, the exodus is arguably a mass denunciation of the busted American social media industry. On another, though, it might just be that RedNote has a compelling product: as one Chinese-English user summed it up on X-formerly-Twitter:"it's pretty much a replacement for search engine and insanely useful for everything".Many of the 170 million American TikTok users flocking to the Chinese app will quickly notice the dominance of mandarin text, or hnz. That hasn't stopped tens of thousands of xhs'ers from rallying around the tag "TikTok Refugee," which is quickly becoming a first-stop for users to reconnect with American content creators rebuilding their empires beyond the confines of Silicon Valley.Nor has the app's translated name "little red book" dissuaded Americans from signing on with fervor (the app's founder claims to have named xhs after the color palette of Stanford, where he received his MBA, as opposed to the book of quotations by People's Republic of China founder Mao Zedong).As the term "refugee" implies, American TikTok users are stumbling into a digital environment already populated by a vast and foreign culture. Prior to the "Great Digital Migration," RedNote was home to roughly 300 million monthly users, the majority of whom are young, urban women, according to the BBC.So far, China's Pinterest girlies have been gracious hosts. Many high-profile Chinese accounts have begun adding English subtitles to videos, pinning reels with titles like "How TikTok Refugees Can fit in RedNote" to the top of their profiles. Some are teaching bite-sized lessons in hnz-English shorthand, while others spin elocution lessons for their curious American counterparts.Though the internet has often been heralded as a tool for information dissemination and western progressivism remember the Arab Spring? the reality isn't so idealistic. Case in point, China tends to ban American apps that don't comply with its regulatory standards, and the US has long used the internet as a tool to maintain global hegemony. So it's interesting to see Western and Chinese users interacting directly, a turn with little precedence in the contemporary world.It's all the more fascinating that the tensions of state which usually keep the two societies separate has, in this case, driven them right into each other's arms. In practice, posts on RedNote suggest that American users are uniting around the app out of spite for the US government, or at least the parts of it that might ban the distribution of TikTok."I surrender all my data to China," said one user as he knelt, offering a binder full of papers. "Fuck you US government. Fuck. You."Chinese users were quick to catch onto this American angst, gleefully inviting westerners to reconnect with their now out-of-the-job "Chinese spy" assigned to monitor them over TikTok. Time will tell if this is a true Berlin-wall moment, or if more government crackdowns are soon to follow.More on social media: Facebook Caught Hosting AI-Powered HitlerShare This Article

Jan 14, 2025Ravie LakshmananVulnerability / Data PrivacyNew research has pulled back the curtain on a "deficiency" in Google's "Sign in with Google" authentication flow that exploits a quirk in domain ownership to gain access to sensitive data."Google's OAuth login doesn't protect against someone purchasing a failed startup's domain and using it to re-create email accounts for former employees," Truffle Security co-founder and CEO Dylan Ayrey said in a Monday report. "And while you can't access old email data, you can use those accounts to log into all the different SaaS products that the organization used."The San Francisco-based company said the issue has the potential to put millions of American users' data at risk simply by purchasing a defunct domain associated with a failed startup and gaining unauthorized access to old employee accounts related to various applications like OpenAI ChatGPT, Slack, Notion, Zoom, and even HR systems."The most sensitive accounts included HR systems, which contained tax documents, pay stubs, insurance information, social security numbers, and more," Ayrey said. "Interview platforms also contained sensitive information about candidate feedback, offers, and rejections."OAuth, short for open authorization, refers to an open standard for access delegation, allowing users to grant websites or applications access to their information on other websites without having to give their passwords. This is accomplished by making use of an access token to verify the user's identity and allow the service to access the resource the token is intended for.When "Sign in with Google" is used to sign in to an application such as Slack, Google sends the service a set of claims about the user, including their email address and the hosted domain, which could then be utilized to log users into their accounts.This also means that if a service is solely relying on these pieces of information to authenticate users, it also opens the door to a scenario where domain ownership changes could allow an attacker to regain access to old employee accounts.Truffle also pointed out Google's OAuth ID token includes a unique user identifier the sub claim that could theoretically prevent the problem, but that has been found to be unreliable. It's worth noting that Microsoft's Entra ID tokens include the sub or oid claims to store an immutable value per user.While Google initially responded to the vulnerability disclosure by stating that it is intended behavior, it has since re-opened the bug report as of December 19, 2024, awarding Ayrey a bounty of $1,337. It has also qualified the issue as an "abuse-related methodology with high impact."In the meantime, there are no protections that downstream software providers can take to protect against the vulnerability in Google's OAuth implementation. The Hacker News has reached out to Google for further comment, and we will update the story if we hear back."As an individual, once you've been off-boarded from a startup, you lose your ability to protect your data in these accounts, and you are subject to whatever fate befalls the future of the startup and domain," Ayrey said. "Without immutable identifiers for users and workspaces, domain ownership changes will continue to compromise accounts."Found this article interesting? Follow us on Twitter and LinkedIn to read more exclusive content we post.SHARE

When it debuted on Netflix in September of 2021,Squid Game became an instant international sensation. People all over the world were hungry for a show like this, one with not just shocking violence and suspenseful drama, but also something meaningful to say aboutthe brutal nature of modernlife.So does it stand to reason that theyrealso hungry ... for beef jerky?The folks over at Jack Links certainlyhope so. The jerky giant has created a special tie-in product connected to the recent release ofSquid Game Season 2 on Netflix: A limited-time Red Light: Green Light jerky which, according to their official website, draws its inspiration from the heart-pounding intensity ofSquid Game.Frankly, a beef jerky that makesyour heartpound sounds a little dangerous. (Eat enough beef jerky and your heart will probably start to pound eventually, have you seen how much sodium is in a single bag of beef jerky?) Of course, Icant just avoid a tie-in beef jerky just becausetheres a slim chance it might make my heart explode. Eating weird movie and TV-inspired foods is my beat. So when I saw theSquid Gamejerky at my local convenience store, I didnt hesitate.So how does it taste? Does this dried beef snack accurately capture the soul-crushing despair of everyones favorite streaming TV show about capitalisms all-consuming brutality? Watch mySquid Game beef jerky taste test below.READ MORE: I DrankStarbucks TwoWickedDrinksAfter Iended the video I took a look at the list of ingredients on the back of the bag.Below the standard jerky ingredients like beef, water, and sugar it claimed it contains 2% or less of a whole bunch of things, including red miso, fermented rice extract, and something called cultured celery extract.The less than 2% part could explainthe lack of any notable Korean seasoning. Beyond a slightly higher spice level than normal, I wouldnt havedetected any difference between Squid Game jerky and a standard-issue varietalin a blind taste test.That doesnt mean theSquid Gamejerky was bad; I did finish the rest of the bag a few hours later. It just means it didnt really deliver on the labels promise of ssamjang flavor.On that front, that was a disappointment. But hey:Squid Game is all about how capitalismgrinds people down, little by little, day after day, slowly sapping us of all our hopes and dreams. So ... good job, Jack Links?Get our free mobile appA Brief History of Movie Tie-In FoodHow movies fell in love with chain restaurants (and vice versa)...

Time zones: EST (UTC -5), CST (UTC -6), MST (UTC -7), PST (UTC -8), AST (UTC -4), NST (UTC -3:30)At Auctria we're pioneering the future of event based fundraising with our comprehensive platform, which elevates organizations' abilities to maximize their events. Our focus is on empowering all sizes of non-profits in their essential missions with effective and powerful, yet easy to use features.We're seeking a Senior Front-End Developer to join our compact yet impactful team, and make a significant difference in helping grow our front-end capabilities.Key Responsibilities:Collaborate to write maintainable, testable, and efficient code by using best software development practices.Ensure the technical feasibility and stability of front-end components.Implement component testing and unit tests to maintain code quality.Optimize applications for maximum speed, scalability, and maintainability.Drive code quality improvements through thoughtful code reviewsStay updated with the latest industry trends and bring fresh ideas to the table.Requirements:4+ years of professional front-end development experienceBachelor's degree in Computer Science, Engineering, or a related field.Strong expertise in front-end development with Vue.js and TypeScript.Experience with state management patterns and solutions (e.g., Vuex, Pinia).Experience with build tools like Vite.Proven ability to write maintainable, testable, and efficient codeExperience with Version Control/Git, and knowing how to collaborate with others using repositories.Knowledge of responsive design principles and best practices.Strong problem-solving skills and attention to detail.Excellent communication and teamwork skills.Nice to Have:Experience with other front-end frameworks or libraries such as React Native and Nuxt.What We Offer:A collaborative and inclusive work environment.Room for professional advancement and skill development.The flexibility to work remotely from anywhere in Canada.A dynamic and supportive team that values innovation and fresh ideas.Competitive salary and bonus planIf you're passionate about our mission and excited about making a difference, send your resume. Let's create something amazing together!* Applicants must be based in and legally eligible to work in Canada.