Add source Login
AirPods Max Launched Five Years Ago Today

AirPods Max Launched Five Years Ago Today

Apple's AirPods Max launched five years ago today, marking the company's first push into the high-end over-ear headphones market under its own brand name. Rumors about Apple's work on a pair of high-end headphones, at the time believed to be called the "AirPods Studio, " heated up throughout 2020. They were announced abruptly via a somewhat unexpected press release on December 8, 2020 and went on sale the same day. Orders started arriving to customers one week later on Tuesday, December 15. The ‌AirPods Max‌ offer many popular AirPods features such as the H1 chip, easy pairing, Active Noise Cancellation, Transparency mode, automatic switching, and Spatial Audio with dynamic head tracking, but in a premium over-ear design for the first time. They also offer a headband made of a flexible mesh canopy, replaceable magnetic earcups, a Digital Crown for physical volume controls, a button for switching between ANC and Transparency, and a Smart Case for storage and to put the headphones into a low power state. Demand for the ‌AirPods Max‌ was high immediately after launch, with shipping estimates that stretched out several months . Initial reviews of AirPods Max were favorable , applauding the headphones for being "more than enough to compete with other high-end headphones" in terms of design and sound quality. While the recommended retail price remains at $549, the ‌AirPods Max‌ are often available with discounts of over $100. The ‌AirPods Max‌ have also been subject to criticism since their launch, including for their price relative to rival sets of high-end over-ear headphones, the design of the Smart Case , condensation inside the earcups, poor battery life (something that was later fixed via a software update ), ANC strength seemingly being reduced over time , the over-head canopy's poor durability, and the long period in which the device has been left without meaningful hardware update. Last year, Apple refreshed the AirPods Max 's selection of color options and swapped the Lightning port for USB-C, but there were no other changes. Since the changes were so minor, Apple does not seem to consider the "new" model a second-generation. Related Roundup: AirPods Max Buyer's Guide: AirPods Max (Neutral) Related Forum: AirPods This article, " AirPods Max Launched Five Years Ago Today " first appeared on MacRumors.com Discuss this article in our forums

Filing: Oracle signed ~$150B of data center leases in the three months ending November 30, raising its total data center and cloud capacity commitments to $248B (Martin Peers/The Information)

Filing: Oracle signed ~$150B of data center leases in the three months ending November 30, raising its total data center and cloud capacity commitments to $248B (Martin Peers/The Information)

Martin Peers / The Information : Filing: Oracle signed ~$150B of data center leases in the three months ending November 30, raising its total data center and cloud capacity commitments to $248B —  Oracle struck about $150 billion worth of lease commitments on data centers in the three months ending November …

VolkLocker Ransomware Exposed by Hard-Coded Master Key Allowing Free Decryption

VolkLocker Ransomware Exposed by Hard-Coded Master Key Allowing Free Decryption

The pro-Russian hacktivist group known as CyberVolk (aka GLORIAMIST) has resurfaced with a new ransomware-as-a-service (RaaS) offering called VolkLocker that suffers from implementation lapses in test artifacts, allowing users to decrypt files without paying an extortion fee. According to SentinelOne, VolkLocker (aka CyberVolk 2.x) emerged in August 2025 and is capable of targeting both Windows

Why agentic AI needs a new category of customer data

Why agentic AI needs a new category of customer data

Presented by Twilio The customer data infrastructure powering most enterprises was architected for a world that no longer exists: one where marketing interactions could be captured and processed in batches, where campaign timing was measured in days (not milliseconds), and where "personalization" meant inserting a first name into an email template. Conversational AI has shattered those assumptions. AI agents need to know what a customer just said, the tone they used, their emotional state, and their complete history with a brand instantly to provide relevant guidance and effective resolution. This fast-moving stream of conversational signals (tone, urgency, intent, sentiment) represents a fundamentally different category of customer data. Yet the systems most enterprises rely on today were never designed to capture or deliver it at the speed modern customer experiences demand. The conversational AI context gap The consequences of this architectural mismatch are already visible in customer satisfaction data. Twilio’s Inside the Conversational AI Revolution report reveals that more than half (54%) of consumers report AI rarely has context from their past interactions, and only 15% feel that human agents receive the full story after an AI handoff. The result: customer experiences defined by repetition, friction, and disjointed handoffs. The problem isn't a lack of customer data. Enterprises are drowning in it. The problem is that conversational AI requires real-time, portable memory of customer interactions, and few organizations have infrastructure capable of delivering it. Traditional CRMs and CDPs excel at capturing static attributes but weren't architected to handle the dynamic exchange of a conversation unfolding second by second. Solving this requires building conversational memory inside communications infrastructure itself, rather than attempting to bolt it onto legacy data systems through integrations. The agentic AI adoption wave and its limits This infrastructure gap is becoming critical as agentic AI moves from pilot to production. Nearly two-thirds of companies (63%) are already in late-stage development or fully deployed with conversational AI across sales and support functions. The reality check: While 90% of organizations believe customers are satisfied with their AI experiences, only 59% of consumers agree. The disconnect isn't about conversational fluency or response speed. It's about whether AI can demonstrate true understanding, respond with appropriate context, and actually solve problems rather than forcing escalation to human agents. Consider the gap: A customer calls about a delayed order. With proper conversational memory infrastructure, an AI agent could instantly recognize the customer, reference their previous order, details about a delay, proactively suggest solutions, and offer appropriate compensation, all without asking them to repeat information. Most enterprises can't deliver this because the required data lives in separate systems that can't be accessed quickly enough. Where enterprise data architecture breaks down Enterprise data systems built for marketing and support were optimized for structured data and batch processing, not the dynamic memory required for natural conversation. Three fundamental limitations prevent these systems from supporting conversational AI: Latency breaks the conversational contract. When customer data lives in one system and conversations happen in another, every interaction requires API calls that introduce 200-500 millisecond delays, transforming natural dialogue into robotic exchanges. Conversational nuance gets lost. The signals that make conversations meaningful (tone, urgency, emotional state, commitments made mid-conversation) rarely make it into traditional CRMs, which were designed to capture structured data, not the unstructured richness AI needs. Data fragmentation creates experience fragmentation. AI agents operate in one system, human agents in another, marketing automation in a third, and customer data in a fourth, creating fractured experiences where context evaporates at every handoff. Conversational memory requires infrastructure where conversations and customer data are unified by design. What unified conversational memory enables Organizations treating conversational memory as core infrastructure are seeing clear competitive advantages: Seamless handoffs: When conversational memory is unified, human agents inherit complete context instantly, eliminating the "let me pull up your account" dead time that signals wasted interactions. Personalization at scale: While 88% of consumers expect personalized experiences, over half of businesses cite this as a top challenge. When conversational memory is native to communications infrastructure, agents can personalize based on what customers are trying to accomplish right now. Operational intelligence: Unified conversational memory provides real-time visibility into conversation quality and key performance indicators, with insights feeding back into AI models to improve quality continuously. Agentic automation: Perhaps most significantly, conversational memory transforms AI from a transactional tool to a genuinely agentic system capable of nuanced decisions, like rebooking a frustrated customer's flight while offering compensation calibrated to their loyalty tier. The infrastructure imperative The agentic AI wave is forcing a fundamental re-architecture of how enterprises think about customer data. The solution isn't iterating on existing CDP or CRM architecture. It's recognizing that conversational memory represents a distinct category requiring real-time capture, millisecond-level access, and preservation of conversational nuance that can only be met when data capabilities are embedded directly into communications infrastructure. Organizations approaching this as a systems integration challenge will find themselves at a disadvantage against competitors who treat conversational memory as foundational infrastructure. When memory is native to the platform powering every customer touchpoint, context travels with customers across channels, latency disappears, and continuous journeys become operationally feasible. The enterprises setting the pace aren't those with the most sophisticated AI models. They're the ones that solved the infrastructure problem first, recognizing that agentic AI can't deliver on its promise without a new category of customer data purpose-built for the speed, nuance, and continuity that conversational experiences demand. Robin Grochol is SVP of Product, Data, Identity & Security at Twilio. Sponsored articles are content produced by a company that is either paying for the post or has a business relationship with VentureBeat, and they’re always clearly marked. For more information, contact sales@venturebeat.com .

Bolmo’s architecture unlocks efficient byte‑level LM training without sacrificing quality

Bolmo’s architecture unlocks efficient byte‑level LM training without sacrificing quality

Enterprises that want tokenizer-free multilingual models are increasingly turning to byte-level language models to reduce brittleness in noisy or low-resource text. To tap into that niche — and make it practical at scale — the Allen Institute of AI (Ai2) introduced Bolmo , a new family of models that leverage its Olmo 3 models by “bytefiying” them and reusing their backbone and capabilities. The company launched two versions, Bolmo 7B and Bolmo 1B, which are “the first fully open byte-level language model,” according to Ai2 . The company said the two models performed competitively with — and in some cases surpassed — other byte-level and character-based models. Byte-level language models operate directly on raw UTF-8 bytes, eliminating the need for a predefined vocabulary or tokenizer. This allows them to handle misspellings, rare languages, and unconventional text more reliably — key requirements for moderation, edge deployments, and multilingual applications. For enterprises deploying AI across multiple languages, noisy user inputs, or constrained environments, tokenizer-free models offer a way to reduce operational complexity. Ai2’s Bolmo is an attempt to make that approach practical at scale — without retraining from scratch. How Bolmo works and how it was built Ai2 said it trained the Bolmo models using its Dolma 3 data mix, which helped train its Olmo flagship models , and some open code datasets and character-level data. The company said its goal “is to provide a reproducible, inspectable blueprint for byteifying strong subword language models in a way the community can adopt and extend.” To meet this goal, Ai2 will release its checkpoints, code, and a full paper to help other organizations build byte-level models on top of its Olmo ecosystem. Since training a byte-level model completely from scratch can get expensive, Ai2 researchers instead chose an existing Olmo 3 7B checkpoint to byteify in two stages. In the first stage, Ai2 froze the Olmo 3 transformer so that they only train certain parts, such as the local encoder and decoder, the boundary predictor, and the language modeling head. This was designed to be “cheap and fast” and requires just 9.8 billion tokens. The next stage unfreezes the model and trains it with additional tokens. Ai2 said the byte-level approach allows Bolmo to avoid the vocabulary bottlenecks that limit traditional subword models. Strong performance among its peers Byte-level language models are not as mainstream as small language models or LLMs, but this is a growing field in research. Meta released its BLT architecture research last year, aiming to offer a model that is robust, processes raw data, and doesn’t rely on fixed vocabularies. Other research models in this space include ByT5 , Stanford’s MrT5 , and Canine . Ai2 evaluated Bolmo using its evaluation suite, covering math, STEM reasoning, question answering, general knowledge, and code. Bolmo 7B showed strong performance, outperforming character-focused benchmarks like CUTE and EXECUTE, and also improving accuracy over the base LLM Olmo 3. Bolmo 7B outperformed models of comparable size in coding, math, multiple-choice QA, and character-level understanding. Why enterprises may choose byte-level models Enterprises find value in a hybrid model structure, using a mix of models and model sizes. Ai2 makes the case that organizations should also consider byte-level models not only for robustness and multilingual understanding, but because it “naturally plugs into an existing model ecosystem.” “A key advantage of the dynamic hierarchical setup is that compression becomes a toggleable knob,” the company said. For enterprises already running heterogeneous model stacks, Bolmo suggests that byte-level models may no longer be purely academic. By retrofitting a strong subword model rather than training from scratch, Ai2 is signaling a lower-risk path for organizations that want robustness without abandoning existing infrastructure.

Bolmo’s architecture unlocks efficient byte‑level LM training without sacrificing quality

Bolmo’s architecture unlocks efficient byte‑level LM training without sacrificing quality

Enterprises that want tokenizer-free multilingual models are increasingly turning to byte-level language models to reduce brittleness in noisy or low-resource text. To tap into that niche — and make it practical at scale — the Allen Institute of AI (Ai2) introduced Bolmo , a new family of models that leverage its Olmo 3 models by “bytefiying” them and reusing their backbone and capabilities. The company launched two versions, Bolmo 7B and Bolmo 1B, which are “the first fully open byte-level language model,” according to Ai2 . The company said the two models performed competitively with — and in some cases surpassed — other byte-level and character-based models. Byte-level language models operate directly on raw UTF-8 bytes, eliminating the need for a predefined vocabulary or tokenizer. This allows them to handle misspellings, rare languages, and unconventional text more reliably — key requirements for moderation, edge deployments, and multilingual applications. For enterprises deploying AI across multiple languages, noisy user inputs, or constrained environments, tokenizer-free models offer a way to reduce operational complexity. Ai2’s Bolmo is an attempt to make that approach practical at scale — without retraining from scratch. How Bolmo works and how it was built Ai2 said it trained the Bolmo models using its Dolma 3 data mix, which helped train its Olmo flagship models , and some open code datasets and character-level data. The company said its goal “is to provide a reproducible, inspectable blueprint for byteifying strong subword language models in a way the community can adopt and extend.” To meet this goal, Ai2 will release its checkpoints, code, and a full paper to help other organizations build byte-level models on top of its Olmo ecosystem. Since training a byte-level model completely from scratch can get expensive, Ai2 researchers instead chose an existing Olmo 3 7B checkpoint to byteify in two stages. In the first stage, Ai2 froze the Olmo 3 transformer so that they only train certain parts, such as the local encoder and decoder, the boundary predictor, and the language modeling head. This was designed to be “cheap and fast” and requires just 9.8 billion tokens. The next stage unfreezes the model and trains it with additional tokens. Ai2 said the byte-level approach allows Bolmo to avoid the vocabulary bottlenecks that limit traditional subword models. Strong performance among its peers Byte-level language models are not as mainstream as small language models or LLMs, but this is a growing field in research. Meta released its BLT architecture research last year, aiming to offer a model that is robust, processes raw data, and doesn’t rely on fixed vocabularies. Other research models in this space include ByT5 , Stanford’s MrT5 , and Canine . Ai2 evaluated Bolmo using its evaluation suite, covering math, STEM reasoning, question answering, general knowledge, and code. Bolmo 7B showed strong performance, outperforming character-focused benchmarks like CUTE and EXECUTE, and also improving accuracy over the base LLM Olmo 3. Bolmo 7B outperformed models of comparable size in coding, math, multiple-choice QA, and character-level understanding. Why enterprises may choose byte-level models Enterprises find value in a hybrid model structure, using a mix of models and model sizes. Ai2 makes the case that organizations should also consider byte-level models not only for robustness and multilingual understanding, but because it “naturally plugs into an existing model ecosystem.” “A key advantage of the dynamic hierarchical setup is that compression becomes a toggleable knob,” the company said. For enterprises already running heterogeneous model stacks, Bolmo suggests that byte-level models may no longer be purely academic. By retrofitting a strong subword model rather than training from scratch, Ai2 is signaling a lower-risk path for organizations that want robustness without abandoning existing infrastructure.

Nvidia debuts Nemotron 3 with hybrid MoE and Mamba-Transformer to drive efficient agentic AI

Nvidia debuts Nemotron 3 with hybrid MoE and Mamba-Transformer to drive efficient agentic AI

Nvidia launched the new version of its frontier models, Nemotron 3, by leaning in on a model architecture that the world’s most valuable company said offers more accuracy and reliability for agents. Nemotron 3 will be available in three sizes: Nemotron 3 Nano with 30B parameters, mainly for targeted, highly efficient tasks; Nemotron 3 Super, which is a 100B parameter model for multi-agent applications and with high-accuracy reasoning and Nemotron 3 Ultra, with its large reasoning engine and around 500B parameters for more complex applications. To build the Nemotron 3 models, Nvidia said it leaned into a hybrid mixture-of-experts (MoE) architecture to improve scalability and efficiency. By using this architecture, Nvidia said in a press release that its new models also offer enterprises more openness and performance when building multi-agent autonomous systems. Kari Briski, Nvidia vice president for generative AI software, told reporters in a briefing that the company wanted to demonstrate its commitment to learn and improving from previous iterations of its models. “We believe that we are uniquely positioned to serve a wide range of developers who want full flexibility to customize models for building specialized AI by combining that new hybrid mixture of our mixture of experts architecture with a 1 million token context length,” Briski said. Nvidia said early adopters of the Nemotron 3 models include Accenture, CrowdStrike, Cursor, Deloitte, EY, Oracle Cloud Infrastructure, Palantir, Perplexity, ServiceNow, Siemens and Zoom. Breakthrough architectures Nvidia has been using the hybrid Mamba-Transformer mixture-of-experts architecture for many of its models, including Nemotron-Nano-9B-v2 . The architecture is based on research from Carnegie Mellon University and Princeton, which weaves in selective state-space models to handle long pieces of information while maintaining states. It can reduce compute costs even through long contexts. Nvidia noted its design “achieves up to 4x higher token throughput” compared to Nemotron 2 Nano and can significantly lower inference costs by reducing reasoning token generation by up 60%. “We really need to be able to bring that efficiency up and the cost per token down. And you can do it through a number of ways, but we're really doing it through the innovations of that model architecture,” Briski said. “The hybrid Mamba transformer architecture runs several times faster with less memory, because it avoids these huge attention maps and key value caches for every single token.” Nvidia also introduced an additional innovation for the Nemotron 3 Super and Ultra models. For these, Briski said Nvidia deployed “a breakthrough called latent MoE.” “That’s all these experts that are in your model share a common core and keep only a small part private. It’s kind of like chefs sharing one big kitchen, but they need to get their own spice rack,” Briski added. Nvidia is not the only company that employs this kind of architecture to build models. AI21 Labs uses it for its Jamba models, most recently in its Jamba Reasoning 3B model . The Nemotron 3 models benefited from extended reinforcement learning. The larger models, Super and Ultra, used the company’s 4-bit NVFP4 training format, which allows them to train on existing infrastructure without compromising accuracy. Benchmark testing from Artificial Analysis placed the Nemotron models highly among models of similar size. New environments for models to ‘work out’ As part of the Nemotron 3 launch, Nvidia will also give users access to its research by releasing its papers and sample prompts, offering open datasets where people can use and look at pre-training tokens and post-training samples, and most importantly, a new NeMo Gym where customers can let their models and agents “workout.” The NeMo Gym is a reinforcement learning lab where users can let their models run in simulated environments to test their post-training performance. AWS announced a similar tool through its Nova Forge platform , targeted for enterprises that want to test out their newly created distilled or smaller models. Briski said the samples of post-training data Nvidia plans to release “are orders of magnitude larger than any available post-training data set and are also very permissive and open.” Nvidia pointed to developers seeking highly intelligent and performant open models, so they can better understand how to guide them if needed, as the basis for releasing more information about how it trains its models. “Model developers today hit this tough trifecta. They need to find models that are ultra open, that are extremely intelligent and are highly efficient,” she said. “Most open models force developers into painful trade-offs between efficiencies like token costs, latency, and throughput.” She said developers want to know how a model was trained, where the training data came from and how they can evaluate it.