Latest Tech News

Stay updated with the latest in technology, AI, cybersecurity, and more

Filtered by: model Clear Filter

New TP-Link zero-day surfaces as CISA warns other flaws are exploited

TP-Link has confirmed the existence of an unpatched zero-day vulnerability impacting multiple router models, as CISA warns that other router flaws have been exploited in attacks. The zero-day vulnerability was discovered by independent threat researcher Mehrun (ByteRay), who noted that he first reported it to TP-Link on May 11, 2024. The Chinese networking equipment giant confirmed to BleepingComputer that it is currently investigating the exploitability and exposure of the flaw. Though a pat

Xgimi’s incredibly bright 4K projector targets daytime gamers

is a deputy editor and Verge co-founder with a passion for human-centric cities, e-bikes, and life as a digital nomad. He’s been a tech journalist for 20 years. Posts from this author will be added to your daily email digest and your homepage feed. The new Horizon 20 lineup of 4K Google TV projectors from Xgimi targets gamers who want to play day and night on the biggest screen possible. In addition to promising an incredibly bright image from the flagship model, Xgimi is at IFA 2025 saying th

DuckDuckGo adds access to advanced AI models to its subscription plan

Privacy-focused consumer tech company DuckDuckGo launched a subscription plan last year that bundled a VPN service, personal information removal, and identity theft restoration. The company said Thursday that the subscription now gives users access to the latest AI models through Duck.ai without paying extra. The Duck.ai chatbot is free to use, and users get access to models like Anthropic’s Claude 3.5 Haiku, Meta’s Llama 4 Scout, Mistral AI’s Mistral Small 3 24B, and OpenAI’s GPT-4o mini. Wit

I pay for Google Gemini, but GPT-5 has convinced me to switch to ChatGPT and here’s why

Calvin Wankhede / Android Authority ChatGPT has become the Xerox of the AI world, but OpenAI no longer has the stage to itself. Google has been pushing Gemini deeper into its ecosystem, and every few weeks it feels like another rival pops up with promises of being faster, smarter, or more useful. So when the latest GPT-5 model launched last month, it felt like a release OpenAI couldn’t afford to get wrong. While OpenAI’s benchmarks claim that GPT-5 is better than the competition, the numbers o

DuckDuckGo's subscription now offers some of the latest chatbots from OpenAI and Anthropic

Since the start of last summer, DuckDuckGo has offered a handful of AI chatbots from OpenAI, Anthropic and others directly through its browser. And while it's mostly low-cost models like GPT-4o mini and Claude 3.5 Haiku on offer, the feature, Duck.ai, allows you to use those tools without sacrificing your privacy. As a service, that appears to have resonated with DuckDuckGo users, and now the company is reorienting its Privacy Pro subscription around access to more advanced models from those pro

DuckDuckGo adds access to advanced models to it subscription plan

Privacy-focused consumer tech company DuckDuckGo launched a subscription plan last year that bundled a VPN service, personal information removal, and identity theft restoration. The company said Thursday that the subscription now gives users access to the latest AI models through Duck.ai without paying extra. The Duck.ai chatbot is free to use, and users get access to models like Anthropic’s Claude 3.5 Haiku, Meta’s Llama 4 Scout, Mistral AI’s Mistral Small 3 24B, and OpenAI’s GPT-4o mini. Wit

Google closer to powering new Siri and Apple’s AI search tool as Perplexity fades

While they’re not exactly tying the knot yet, Siri and Gemini are making plans to spend a lot more time together. That’s according to a new report that details the current state of Apple’s delayed effort to upgrade Siri for the modern era. Apple has considered Google’s Gemini AI as a potential solution to fuel the AI-powered version of Siri that’s currently in development. No decision has been made, however, as Apple also tests its own AI models as well as OpenAI and Anthropic models. The new

OpenAI is hiring 'AI-pilled' academics to build a scientific discovery accelerator

Andriy Onufriyenko/Moment via Getty Follow ZDNET: Add us as a preferred source on Google. ZDNET's key takeaways OpenAI for Science was announced in a Tuesday X post. Its goal is to accelerate scientific discovery through AI. The post suggests GPT-5 will play a key role in the effort. Artificial intelligence researchers have long dreamed of automating the process of scientific discovery. Now OpenAI is setting out to turn that vision into reality. The company is launching an initiative call

Switzerland releases its own AI model trained on public data

is a NYC-based AI reporter and is currently supported by the Tarbell Center for AI Journalism. She covers AI companies, policies, and products. Posts from this author will be added to your daily email digest and your homepage feed. Switzerland launched an open-source model called Apertus on Monday as an alternative to proprietary models like OpenAI’s ChatGPT or Anthropic’s Claude, reports SWI as spotted by Engadget. The model’s source code, training data, model weights, and detailed developmen

Show HN: Entropy-Guided Loop – How to make small models reason

Logprobs Reasoning Loop with Weights & Biases Weave, an observability tool Uncertainty-Aware Generation with OpenAI's Responses API This project demonstrates a novel approach to improving AI model reasoning by leveraging token-level uncertainty metrics (logprobs) to create self-correcting generation loops. We compare this uncertainty-aware approach against traditional reasoning models to test whether explicit uncertainty handling can match or exceed the performance of dedicated reasoning archi

This Robot Only Needs a Single AI Model to Master Humanlike Movements

Atlas, the humanoid robot famous for its parkour and dance routines, has recently begun demonstrating something altogether more subtle but also a lot more significant: It has learned to both walk and grab things using a single artificial intelligence model. What is more, the robot’s single learning model is showing some tantalizingly “emergent” skills, like the ability to instinctively recover when it drops an item without having been trained to do so. Boston Dynamics, the company that makes A

VibeVoice: A Frontier Open-Source Text-to-Speech Model

VibeVoice: A Frontier Open-Source Text-to-Speech Model VibeVoice is a novel framework designed for generating expressive, long-form, multi-speaker conversational audio, such as podcasts, from text. It addresses significant challenges in traditional Text-to-Speech (TTS) systems, particularly in scalability, speaker consistency, and natural turn-taking. A core innovation of VibeVoice is its use of continuous speech tokenizers (Acoustic and Semantic) operating at an ultra-low frame rate of 7.5 Hz.

Understanding Transformers Using a Minimal Example

The internal mechanisms of Transformer Large Language models (LLMs), particularly the flow of information through the layers and the operation of the attention mechanism, can be challenging to follow due to the vast amount of numbers involved. We humans can hardly form a mental model. This article aims to make these workings tangible by providing visualizations of a Transformer's internal state. Utilizing a minimal dataset and a deliberately simplified model, it is possible to follow the model's

How to watch Apple announce the iPhone 17

Apple is hosting its “Awe dropping” hardware event on Tuesday at 10 a.m. PT. The company is expected to announce its iPhone 17 lineup, which could include a slimmer version, as well as updates for the Apple Watch and AirPods. The keynote will take place at Apple Park in Cupertino. Viewers can tune in to the livestream here or watch it on Apple’s website. As usual, TechCrunch will be covering the event, so stay tuned for news. There have been numerous rumors surrounding the upcoming iPhone line

Turns out it’s possible to get a Pixel 10 with a physical SIM slot in the US

Rita El Khoury / Android Authority TL;DR Google is reportedly sending out global Pixel 10 models with physical SIM slots as replacement units to users. These global models lack support for mmWave 5G. While some people are viewing this as an upgrade, others are concerned about connectivity issues. If you bought a Pixel 10, 10 Pro, or 10 Pro XL in the US, your phone came without a physical SIM slot. That’s the standard for US models of Google’s latest flagships. Instead of having a traditional

Finding thousands of exposed Ollama instances using Shodan

The rapid deployment of large language models (LLMs) has introduced significant security vulnerabilities due to misconfigurations and inadequate access controls. This paper presents a systematic approach to identifying publicly exposed LLM servers, focusing on instances running the Ollama framework. Utilizing Shodan, a search engine for internet-connected devices, we developed a Python-based tool to detect unsecured LLM endpoints. Our study uncovered over 1,100 exposed Ollama servers, with appro

Microsoft VibeVoice: A Frontier Open-Source Text-to-Speech Model

VibeVoice: A Frontier Open-Source Text-to-Speech Model VibeVoice is a novel framework designed for generating expressive, long-form, multi-speaker conversational audio, such as podcasts, from text. It addresses significant challenges in traditional Text-to-Speech (TTS) systems, particularly in scalability, speaker consistency, and natural turn-taking. A core innovation of VibeVoice is its use of continuous speech tokenizers (Acoustic and Semantic) operating at an ultra-low frame rate of 7.5 Hz.

Medicare Will Start Paying AI Companies a Share of Any Claims They Automatically Reject

It's long been the practice of private health insurers to require "prior authorization" before you can get the treatment you need. Often, they'll try to deny as many of these claims as possible — including with the use of AI models. Government-backed plans like Medicare, however, have tended to cover what private insurers don't, and without the laborious application process. But that could be poised to change. The Centers for Medicare and Medicaid Services said it'll experiment with its own ve

The iPhone’s ‘boring’ era officially ends next week

Apple launches new flagship iPhones every year with a variety of upgrades. But after years of relatively stagnant innovation on the design front, the iPhone 17 launch will kick off several years of exciting new innovations that put an end to the recent ‘boring’ iPhone era. iPhone design innovation has long been stagnant, but that era is over If you’ve followed Apple long enough, you likely remember the days when new iPhone announcements came with a lot more optimism and excitement. But more r

'World Models,' an old idea in AI, mount a comeback

The latest ambition of artificial intelligence research — particularly within the labs seeking “artificial general intelligence,” or AGI — is something called a world model: a representation of the environment that an AI carries around inside itself like a computational snow globe. The AI system can use this simplified representation to evaluate predictions and decisions before applying them to its real-world tasks. The deep learning luminaries Yann LeCun (of Meta), Demis Hassabis (of Google Dee

'World Models,' an Old Idea in AI, Mount a Comeback

The latest ambition of artificial intelligence research — particularly within the labs seeking “artificial general intelligence,” or AGI — is something called a world model: a representation of the environment that an AI carries around inside itself like a computational snow globe. The AI system can use this simplified representation to evaluate predictions and decisions before applying them to its real-world tasks. The deep learning luminaries Yann LeCun (of Meta), Demis Hassabis (of Google Dee

OpenAI to route sensitive conversations to GPT-5, introduce parental controls

OpenAI said Tuesday it plans to route sensitive conversations to reasoning models like GPT-5 and roll out parental controls within the next month — part of an ongoing response to recent safety incidents involving ChatGPT failing to detect mental distress. The new guardrails come in the aftermath of the suicide of teenager Adam Raine, who discussed self-harm and plans to end his life with ChatGPT, which even supplied him with information about specific suicide methods. Raine’s parents have filed

Switzerland launches its own open-source AI model

There's a new player in the AI race, and it's a whole country. Switzerland has just released Apertus, its open-source national Large Language Model (LLM) that it hopes would be an alternative to models offered by companies like OpenAI. Apertus, Latin for the world "open," was developed by the Swiss Federal Technology Institute of Lausanne (EPFL), ETH Zurich and the Swiss National Supercomputing Centre (CSCS), all of which are public institutions. "Currently, Apertus is the leading public AI mod

You can try Apple’s lightning-fast video captioning model right from your browser

A few months ago, Apple released FastVLM, a Visual Language Model (VLM) that offered near-instant high-resolution image processing. Now, you can take it for a spin, provided you have an Apple Silicon-powered Mac. Here’s how. When we first covered FastVLM, we explained that it leveraged MLX, Apple’s own open ML framework specifically designed for Apple Silicon, to deliver up to 85 times faster video captioning, while being more than 3 times smaller than similar models. Since then, Apple has wor

Lessons from building an AI data analyst

AI/ML Data Analytics Malloy Malloy TL;DR Text-to-SQL is not enough. Answering real user questions requires going the extra mile like multi-step plans, external tools (coding) and external context. Answering real user questions requires going the extra mile like multi-step plans, external tools (coding) and external context. Context is the product. A semantic layer (we use Malloy ⎋) encodes business meaning and sharply reduces SQL complexity. A semantic layer (we use Malloy ⎋) encodes busines

Why Runway is eyeing the robotics industry for future revenue growth

Runway has spent the past seven years building visual-generating tools for the creative industry. Now, it sees a new opportunity for its technology: robotics. New York-based Runway is known for its video and photo generation AI world models, or large language models that create a simulated version of the real world. Most recently, the company released Gen-4, its video-generating model, in March and Runway Aleph, its video editing model, in July. As Runway’s world models started to improve — an

Tesla Hit With Another Major Recall

Tesla is recalling 7,301 Model Y SUVs produced in 2025, following identification of a software defect in the driver’s side window’s automatic protection system. The recall, issued by Australia’s Department of Infrastructure, Transport, Regional Development, Communications, Sport and the Arts, warns that the window may close with excessive force if it fails to detect obstructions, posing a risk of injury. The latest recall is a perfect example of the dual-edge of automotive digitization, becaus

Latam-GPT: The Free, Open Source, and Collaborative AI of Latin America

The supercomputing infrastructure at the University of Tarapacá (UTA) in Arica, Chile, is a fundamental pillar for Latam-GPT. With a projected investment of $10 million, the new center has a cluster of 12 nodes, each equipped with eight state-of-the-art NVIDIA H200 GPUs. This capacity, unprecedented in Chile and the region more broadly, not only enables large-scale model training in the country for the first time, it also encourages decentralization and energy efficiency. The first version of L

Ask HN: Best foundation model for CLM fine-tuning?

Hi, I have a largish (2 GB) corpus of curated, high-quality text in some low-resource language, and I want to build a model that would provide an advanced "auto complete" service for writers. I'm thinking of taking a decoder-only model such as Llama, Mistral or Gemma, slice off the embedding layers (which are based on unneeded languages), create new ones (perhaps initialized based on a FastText model trained on the corpus), paired with a tokenizer newly created from my corpus, then train the m

Ford and the Birth of the Model T

This is an excerpt from my forthcoming book, The Origins of Efficiency, out September 23rd. Ford’s status as a large-volume car producer began with the predecessor to the Model T: the Model N, a four-cylinder, two-seater car initially priced at $500. At the time, the average car in the US cost more than $2,000, and it seemed nearly unimaginable that a car with the capabilities of the Model N could cost so little. In 1906, the year the Model N was introduced, Ford sold 8,500 of them, making the