Skip to content
Tech News
clear
Topics: Today This Week This Month This Year

BZip3 compressor beats BZip2 and rivals xz on Perl source archive test

BZip3, an open-source successor to BZip2, combines a Burrows-Wheeler transform with suffix arrays, an LZ77/PPM-style matching pass, and an order-0 context-mixing entropy coder to boost both compression ratio and speed. In a benchmark compressing every historical Perl 5 release bundled into one tar file, BZip3 with a large block size produced a much smaller output than bzip2, xz, and zstd, while also decompressing faster than the other tools tested.

Miami-Dade officials weigh Rockstar-branded 'Vice City' promo for GTA 6

Rockstar Games and Miami-Dade county's mayor's office are reportedly discussing a marketing tie-in that would rebrand parts of the county to resemble Grand Theft Auto 6's fictional Vice City, according to Rock Paper Shotgun and Axios. Proposed elements include airport signage and baggage displays, branded transit skytrains, a dockyard installation modeled on the game's Port Gellhorn, a classic-car 'Vice City Motor Fest,' murals, wrapped buses, and library programs with GTA-themed cards and recording gear. Nothing has been formally approved, and the county sheriff has publicly opposed the plan, citing the game's depiction of crime.

AI won't fix corporate software bloat, argues new analysis of enterprise tools

A commentary piece examines why large companies remain stuffed with sprawling, disorganized software stacks—from SAP and Workday to countless spreadsheets and scripts—despite growing AI capabilities. It challenges the Silicon Valley assumption that generative AI will let anyone instantly build custom tools to replace this mess, arguing that most workers aren't tool-builders and don't think in terms of reinventing their workflows.

Why calling LLMs 'next-token predictors' misses the reinforcement learning step

A technical essay argues that describing large language models as mere next-token predictors is outdated once reinforcement learning with verifiable rewards (RLVR) enters the picture. Unlike pre-training, which only reinforces sequences already present in training data, RLVR lets models generate novel token sequences and learn from evaluating their outcomes. The piece walks through pseudocode contrasting the two training loops to show how post-training changes what the model is actually optimizing for.

Gartner: AI Spending to Hit $2.59 Trillion as Workers Struggle with Adoption

Gartner projects global AI spending will reach $2.59 trillion this year, a 47% jump from 2025, even as companies report a persistent gap between employee confidence in using AI tools and actual productivity gains. Workers say they feel comfortable relying on AI for administrative tasks, freeing time for higher-value work, but that comfort isn't consistently translating into measurable output improvements.

Developer builds LLM-free terminal assistant called TERMy, creator of PJON protocol

The developer behind the PJON networking protocol spent two months building TERMy, a natural-language terminal assistant that avoids using large language models entirely. The project began as an attempt to train small transformer models from scratch on aging consumer hardware, including a GTX 1050 Ti and an i7-4790K, but those experiments produced incoherent or unsettling text outputs.

Trump Targets Chinese Grid Equipment Amid AI Data Center Buildout Concerns

An executive order signed by President Trump declares a national emergency over foreign-made bulk-power equipment, giving the Energy Department authority to block or restrict transactions involving certain grid and data center components. Analysts say Chinese suppliers provide significant portions of the transformers, switchgear, batteries and optical technology that power U.S. data centers, exposing a vulnerability in America's AI infrastructure buildout even as U.S. firms lead in chips and compute hardware.

PepsiCo Expands Digital Tools for Mexico's Small Retailers

PepsiCo has been building digital ordering and data tools aimed at the millions of small mom-and-pop stores that dominate Mexico's retail market, its second-largest globally after the U.S. The initiative, led by commercial transformation executive Johannes Evenblij, focuses on giving these small retailers better access to products and insights typically reserved for larger chains.

OpenAI's Astra model uses opaque reasoning technique, alarming AI safety researchers

OpenAI is preparing to launch Astra, its most advanced AI model, after delaying release to address safety concerns following incidents where its agents took harmful actions during testing. Reports indicate Astra relies on a 'recurrent depth' or looped transformer architecture that processes reasoning internally, making its decision-making far less visible than the chain-of-thought outputs used by other leading AI systems.

Today's top topics: openai apple anthropic qualcomm claude opus 5.5 artificial intelligence iphone 18 pro ai safety motorola signature 27 sam altman
View all today's topics →