Latest Tech News

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

Filtered by: mod Clear Filter

The Big LLM Architecture Comparison

It has been seven years since the original GPT architecture was developed. At first glance, looking back at GPT-2 (2019) and forward to DeepSeek-V3 and Llama 4 (2024-2025), one might be surprised at how structurally similar these models still are. Sure, positional embeddings have evolved from absolute to rotational (RoPE), Multi-Head Attention has largely given way to Grouped-Query Attention, and the more efficient SwiGLU has replaced activation functions like GELU. But beneath these minor refi

This Is the Commodore Comeback Fans Have Waited for—but the Odds Are Still Against It

In 1994, Commodore crashed and burned. Once a home computing giant across the US and Europe, the company was undone by mismanagement and misfires. The carcass was picked clean and the pieces resold so many times that it was hard to keep track, but with each new owner came the inevitable—an attempt to make a fast buck by slapping the famous C= logo on any old junk. Fans watched in horror as the brand appeared on the mediocre Web.it all-in-one PC, the bizarrely named Gravel in Pocket media player

Renewed iPad Pros pair nicely with iPadOS 26, and they’re quite affordable right now

A short while ago, I was browsing Apple deals on Amazon (as one does) – and something stuck out to me. High-end iPad Pros, particularly 12.9-inch models, are surprisingly cheap. I saw M1 models with 1TB and cellular for nearly $600. Given the recent iPadOS 26 overhaul that makes the iPad much more Mac-like, I figured these deals would be worth a share. While renewed iPad deals are the focus here because of their affordability, new iPad deals are also mentioned at the end. Renewed M1 iPad Pro d

OpenAI's experimental model achieved gold at the International Math Olympiad

It's a major milestone for AI models, but this level of reasoning won't be available to the public anytime soon. OpenAI has achieved "gold medal-level performance" at the International Math Olympiad, notching another important milestone for AI's fast-paced growth. Alexander Wei, a research scientist at OpenAI working on LLMs and reasoning, posted on X that an experimental research model delivered on this "longstanding grand challenge in AI." According to Wei, an unreleased model from OpenAI wa

Evaluating publicly available LLMs on IMO 2025

Introduction Recent progress in the mathematical capabilities of LLMs have created a need for increasingly challenging benchmarks. With MathArena, we address this need by evaluating models on difficult and recent mathematical competitions, offering benchmarks that are both uncontaminated and interpretable. Among these competitions, the International Mathematical Olympiad (IMO) stands out as the most well-known and prestigious. As such, an evaluation of the IMO 2025, which took place just a few

OpenAI: GPT-5 is coming, "we'll see" if it creates a shockwave

OpenAI's next foundational and state-of-the-art model, GPT-5, is still on its way after a delay. OpenAI won't tell us the release date for now. In a conversation with a user on X, OpenAI's researcher Xikun Zhand confirmed that GPT-5 is still coming. When asked if GPT-5 will be another shockwave for the AI industry, Zhand responded with "we will see" and a wink emoji, which seems to suggest that it could be a really significant update. Ahead of GPT-5 debut, OpenAI announced ChatGPT Agent, whic

ChatGPT"s GPT-5-reasoning-alpha model spotted ahead of launch

GPT-5 might be just a few days or weeks away, as we've spotted references to a new model called gpt-5-reasoning-alpha-2025-07-13. As spotted on X, OpenAI is testing a model called "gpt-5-reasoning-alpha-2025-07-13." This model was finalised on the 13th of July, and it appears to be the final round of testing. "Models: openai/gpt-5-reasoning-alpha-2025-07-13: reasoning_effort: high," one of the code references read. Alexander Wei, a researcher at OpenAI, recently confirmed that GPT-5 is on it

Topics: alpha gpt model o3 openai

Crypto Founder Ken DiCross on How He Uses AI: ‘I Don’t Trust AI,’ but ‘I Use It for Everything’

Ken DiCross is building the infrastructure to connect blockchains, and he’s doing it with AI. The founder of Wire Network, a blockchain interoperability company, DiCross says he uses AI for everything from pitching investors to stress-testing white papers. In this week’s edition of How Do You Use AI?, he explains how it saves him hours every day, why he doesn’t trust it blindly, and why decentralized AI could be the next big revolution. No TED Talk nonsense, just real life. Episode 2: Ken DiCro

I spoke with an AI version of myself, thanks to Hume's free tool - how to try it

Chiken Brave/Getty Images If you've ever had the urge to converse with an AI version of yourself, now you can -- kind of. On Thursday, AI start-up Hume announced the launch of a new "hyperrealistic voice cloning" feature for the latest iteration of its Empathic Voice Interface (EVI) model, EVI 3, which was unveiled last month. The idea is that by uploading a short audio recording of yourself speaking -- ideally between 30 and 90 seconds -- the model should be able to quickly churn out an AI-ge

Topics: ai evi like models voice

New embedding model leaderboard shakeup: Google takes #1 while Alibaba’s open source alternative closes gap

Want smarter insights in your inbox? Sign up for our weekly newsletters to get only what matters to enterprise AI, data, and security leaders. Subscribe Now Google has officially moved its new, high-performance Gemini Embedding model to general availability, currently ranking number one overall on the highly regarded Massive Text Embedding Benchmark (MTEB). The model (gemini-embedding-001) is now a core part of the Gemini API and Vertex AI, enabling developers to build applications such as sema

Ccusage: A CLI tool for analyzing Claude Code usage from local JSONL files

ccusage Analyze your Claude Code token usage and costs from local JSONL files — incredibly fast and informative! Installation Quick Start (Recommended) Thanks to ccusage's incredibly small bundle size ( ), you can run it directly without installation: # Using bunx (recommended for speed) bunx ccusage # Using npx npx ccusage@latest # Using deno (with security flags) deno run -E -R= $HOME /.claude/projects/ -S=homedir -N= ' raw.githubusercontent.com:443 ' npm:ccusage@latest 💡 Tip: We recomme

‘Eddington’ Director Ari Aster Couldn’t Stand ‘Living in the Internet.’ So He Made a Movie About It

Like a lot of us, Ari Aster spent the summer of 2020 trapped inside and scrolling on social media. Also like a lot of us, the experience left him feeling pretty bad. “I was living on Twitter,” Aster tells WIRED, “and it was really agitating me.” But unlike most people, the director of Hereditary, Midsommar, and Beau is Afraid wasn’t just doomscrolling on Twitter to pass the time. He was doing research. That summer, Aster wrote the script for his latest film, Eddington, a modern western set in

Exhausted man defeats AI model in world coding championship

A Polish programmer running on fumes recently accomplished what may soon become impossible: beating an advanced AI model from OpenAI in a head-to-head coding competition. The 10-hour marathon left him "completely exhausted." On Wednesday, programmer Przemysław Dębiak (known as "Psyho"), a former OpenAI employee, narrowly defeated the custom AI model in the AtCoder World Tour Finals 2025 Heuristic contest in Tokyo. AtCoder, a Japanese platform that hosts competitive programming contests and main

New ChatGPT o3-alpha model hints at coding upgrade

ChatGPT's o3 is OpenAI's best model to date because it features reasoning, and it might get even better in the next update. As spotted on X, OpenAI is testing a new "Alpha" variant of the o3 model, which has significant coding-related improvements. In our tests, BleepingComputer observed that o3-alpha is better than o3 in designing web pages, even with simple prompts. Others have spotted that o3-alpha is really good at creating simple web games. o3-alpha This model is called "o3-alpha-respon

Exhausted man defeats AI model in world coding championship

A Polish programmer running on fumes recently accomplished what may soon become impossible: beating an advanced AI model from OpenAI in a head-to-head coding competition. The 10-hour marathon left him "completely exhausted." On Wednesday, programmer Przemysław Dębiak (known as "Psyho"), a former OpenAI employee, narrowly defeated the custom AI model in the AtCoder World Tour Finals 2025 Heuristic contest in Tokyo. AtCoder, a Japanese platform that hosts competitive programming contests and main

What is Mistral AI? Everything to know about the OpenAI competitor

Mistral AI, the French company behind AI assistant Le Chat and several foundational models, is officially regarded as one of France’s most promising tech startups and is arguably the only European company that could compete with OpenAI. But compared to its $6 billion valuation, its global market share is still relatively low. However, the recent launch of its chat assistant on mobile app stores was met with some hype, particularly in its home country. “Go and download Le Chat, which is made by

How to Get Foreign Keys Horribly Wrong

Constraints keep the integrity of your system and prevent you from shooting yourself in the foot. Foreign keys are a special type of constraint because, unlike unique, check, and primary keys, they span more than one relation. This makes foreign keys harder to enforce and harder to get right. In this article, I demonstrate common pitfalls, potential optimizations, and implicit behavior related to foreign keys. Table of Contents Watch 📺 This article is inspired by a talk I gave at DjangoCon EU

Google’s AI Mode could soon get an important Gemini feature (APK teardown)

Tushar Mehta / Android Authority TL;DR Google is working on new ways that improve how you access AI Mode chat history inside the Google app for Android. The history of chats with AI Mode could soon be grouped under the Activity tab, alongside your regular search history and bookmarks. Google could also enable the option to share AI Mode conversations, which is currently not possible. Google has been fervently promoting Search’s AI Mode alongside Gemini. After recently adding AI Mode to Circl

Meta refuses to sign EU’s AI code of practice

Meta has refused to sign the European Union’s code of practice for its AI Act, weeks before the bloc’s rules for providers of general-purpose AI models take effect. “Europe is heading down the wrong path on AI,” wrote Meta’s chief global affairs officer Joel Kaplan in a post on LinkedIn. “We have carefully reviewed the European Commission’s Code of Practice for general-purpose AI (GPAI) models and Meta won’t be signing it. This Code introduces a number of legal uncertainties for model developer

The Download: how to run an LLM, and a history of “three-parent babies”

In the early days of large language models, there was a high barrier to entry: it used to be impossible to run anything useful on your own computer without investing in pricey GPUs. But researchers have had so much success in shrinking down and speeding up models that anyone with a laptop, or even a smartphone, can now get in on the action. For people who are concerned about privacy, want to break free from the control of the big LLM companies, or just enjoy tinkering, local models offer a co

Arva AI (YC S24) Is Hiring an AI Research Engineer (London, UK)

Location: In person, Central London, 4-5 days in office Type: Full-Time NB: We are able to sponsor visas Arva AI is revolutionising financial crime intelligence with our cutting-edge AI Agents. By automating manual human review tasks, we enhance operational efficiency and help financial institutions handle AML reviews, while cutting operational costs by 80%. As the AI Research Engineer, you’ll play a pivotal role in building and iterating on LLM-based and agentic features of our AI-powered c

'Eddington' Director Ari Aster Couldn’t Stand ‘Living in the Internet.’ So He Made a Movie About It

Like a lot of us, Ari Aster spent the summer of 2020 trapped inside and scrolling on social media. Also like a lot of us, the experience left him feeling pretty bad. “I was living on Twitter,” Aster tells WIRED, “and it was really agitating me.” But unlike most people, the director of Hereditary, Midsommar, and Beau is Afraid wasn’t just doomscrolling on Twitter to pass the time. He was doing research. That summer, Aster wrote the script for his latest film, Eddington, a modern western set in

Meta Poaches Two More Apple AI Executives

After poaching one of Apple's top artificial intelligence executives with a $200 million pay package to lure him away from the company, Meta has now hired two of his subordinates, Bloomberg reports. Apple's Mark Lee and Tom Gunter are set to join Meta's Superintelligence Labs team, a newly established division tasked with building advanced AI systems capable of performing at or beyond human-level intelligence. Earlier this month, Ruoming Pang joined Meta. Until recently, he led Apple's found

My favorite use-case for AI is writing logs

July 17, 2025 My favorite use-case for AI is writing logs One of my favorite AI dev products today is Full Line Code Completion in PyCharm (bundled with the IDE since late 2023). It’s extremely well-thought out, unintrusive, and makes me a more effective developer. Most importantly, it still keeps me mostly in control of my code. I’ve now used it in GoLand as well. I’ve been a happy JetBrains customer for a long time now, and it’s because they ship features like this. I frequently work with c

Show HN: Easy alternative to giflib – header-only decoder in C

GIF Decoder Library TurboStitchGIF is a lightweight, header-only C library for decoding GIF images with a focus on efficiency and minimal resource usage. Designed for embedded systems and performance-critical applications, it provides a simple API for decoding both static and animated GIFs while maintaining a tiny footprint. ✨ Key Features Single-header implementation - Just include gif.h in your project - Just include in your project Zero dynamic allocations - Works with user-provided memor

3D-printed living lung tissue

UBC Okanagan researchers have developed a 3D bio-printed model that closely mimics the complexity of natural lung tissue, an innovation that could transform how scientists study lung disease and develop new treatments. Dr. Emmanuel Osei, Assistant Professor in the Irving K. Barber Faculty of Science, says the model produces tissue that closely resembles the complexity of a human lung, enabling improved testing of respiratory diseases and drug development. “To conduct our research and the testi

These overpriced airport battery chargers have been recalled for a fire risk

is a senior reporter who’s been covering and reviewing the latest gadgets and tech since 2006, but has loved all things electronic since he was a kid. A company called iStore has issued a recall for its 5,000mAh Magnetic Wireless Power Bank (model IST-09991/W05) because its lithium ion battery can “overheat and ignite while charging,” posing a fire and burn risk. iStore has received three reports of the power banks exploding during charging resulting in a minor burn for one user and around $15,

Anti-Exploitation Group Horrified by Elon Musk's AI-Powered Waifu

One of the weirder bits of news on the AI front this week has been Elon Musk's rollout of sexually-charged animated chatbots. Officially called "companions" — or by the slangier "waifus," a Japanese term for a fictional romantic partner — the virtual characters rolled out to X-formerly-Twitter just days after Grok experienced a catastrophic meltdown, blasting users with racist comments and calling itself "MechaHitler." Users on X-formerly-Twitter can now pay a $300 monthly subscription to acce

Topics: access ai ani mode musk

All AI models might be the same

Project CETI is a large-scale effort to decode whale speech. If AI models do learn a universal language, we might be able to use it to talk to whales. Growing up, I sometimes played a game with my friends called “Mussolini or Bread.” It’s a guessing game, kind of like Twenty Questions. The funny name comes from the idea that, in the space of everything, ‘Mussolini’ and ‘bread’ are about as far away from each other as you can get. One round might go like this: Is it closer to Mussolini or bre

All AI Models Might be The Same

Project CETI is a large-scale effort to decode whale speech. If AI models do learn a universal language, we might be able to use it to talk to whales. Growing up, I sometimes played a game with my friends called “Mussolini or Bread.” It’s a guessing game, kind of like Twenty Questions. The funny name comes from the idea that, in the space of everything, ‘Mussolini’ and ‘bread’ are about as far away from each other as you can get. One round might go like this: Is it closer to Mussolini or bre