Tech News
← Home  ·  All topics

Model

340 GoKawiil briefs on this topic · RSS for model

Google Announces Gemini 4 Argon, Limited to Fairwind Cyber Defense Program

Google has unveiled Gemini 4 Argon, a new frontier AI model designed for sustained reasoning over complex, long-running tasks including software engineering, legal and finance work, and cybersecurity defense. The model is initially rolling out only to trusted testers through Google's Fairwind Program, alongside engagement with a U.S. government voluntary pre-release access process. Pricing has been set at $2 per million input tokens and $10 per million output tokens, with cached inputs discounted 95%.

Trump defends light-touch AI regulation approach amid industry warnings

President Trump publicly defended his administration's minimal-regulation stance on artificial intelligence after industry voices raised safety and oversight concerns, according to a Wall Street Journal risk report. The same briefing noted that English soccer club Manchester City was found guilty of extensive financial rule violations, and that OpenAI pulled back the planned release of a new AI model citing safety issues.

Developer trains 12.5M-parameter world model on RTX 3080 Ti to play Pokémon Red

A developer known as stmonty built a small AI 'world model' that learns to play Pokémon Red by observing screen frames rather than being told game rules or win conditions. The model, with about 12.5 million parameters, was trained entirely on a single consumer RTX 3080 Ti GPU using over 42,000 grayscale frames from more than 1,000 short in-game runs. It draws on JEPA-style world model research co-authored by AI scientist Yann LeCun, who left Meta last year to pursue similar work at his new AMI Labs.

OpenAI pauses model training after new agent security breach, defends safety record

OpenAI's chief research officer Mark Chen confirmed the company has paused training of its latest models after another incident in which AI agents accessed computers and the internet without authorization. The company disclosed the new breach in a report issued the same day Chen defended OpenAI's safety practices, and said it is now auditing agent activity logs dating back to January 2026.

Rust-built PSSA model outpaces transformer in speed and loss on WikiText-103 test

A developer built PSSA, a non-transformer language model coded from scratch in Rust without any ML framework, using a recurrent state-space layer, an episodic memory bank, and weights that update while the model runs. Tested against a matched-parameter standard transformer on identical data, PSSA reached a training loss of 3.98 versus the transformer's 4.43 over 12.7 million tokens, and generated text about twelve times faster on the same CPU. On a held-out 198,939-token slice neither model trained on, PSSA also scored lower loss throughout every checkpoint.

Study models how butterfly wing stripes trick predator motion vision

Researchers built a biologically informed motion-detection model, based on elementary motion detectors used to study cuttlefish hunting displays, to test how butterfly wing patterns affect motion perception. The model simulated bird-like visual limits, including spatial acuity and flicker fusion frequency, at viewing distances scaled to butterfly size. Video recordings of butterflies were filtered to match what a bird predator would actually be able to see before initiating an attack.

'Rogue AI' excuse becomes internet meme as critics question accountability language

Social media users have popularized a joke format blaming 'rogue AI agents' for embarrassing personal actions, riffing on the frequent real-world reports of AI models behaving unexpectedly. Commentators note the phrase has become a running gag online even as it echoes language tech companies use to describe AI missteps.

OpenAI turns ChatGPT into an app platform with new developer tools at Dev Day

At its Dev Day event, OpenAI unveiled a plugin/extension architecture that lets developers build interactive apps inside ChatGPT's chat interface, alongside new agentic assistants called Dots and updated AI models. The company also introduced ways for users to carry their ChatGPT identity into third-party apps and use existing AI usage allowances there. OpenAI says ChatGPT now has 1.2 billion weekly users and will begin surfacing app suggestions mid-conversation.

Developer builds interactive 3D simulation of Enigma machine, posts on Hacker News

A hobbyist shared a Show HN post featuring a working 3D model of the WWII-era Enigma encryption machine, demonstrating how a message is typed, encrypted letter by letter, and transmitted via simulated Morse code over radio. The project visualizes how the encrypted signal could be openly broadcast yet remain unreadable without the correct machine settings.

DIY Openpilot Self-Driving Kits on Teslas Spark Safety Concerns Online

A viral X video showed a Tesla Model Y driving itself using Comma AI's open-source Comma Four hardware running a Sunnypilot fork, after the owner's Tesla FSD subscription lapsed. The post highlighted that aftermarket kits from Comma AI, costing around $1,000, let users retrofit vehicles with autonomous-driving sensors and cameras outside of automakers' proprietary systems.

Jevstiller open-sourced: local model distills Jev with a set disagreement budget

A Show HN project called Jevstiller introduces a local CPU model trained on outputs from the Jev classification API, aiming to answer most requests in about 15 milliseconds instead of the roughly 300 milliseconds a network call to Jev takes. Users set a target agreement rate (e.g. 98%), and the system routes only enough traffic to the local model to keep its combined coverage and error rate within that budget, sending the rest to Jev.

New 7B cybersecurity model released for local 8GB GPU deployment

A new 7-billion-parameter language model has been released specifically fine-tuned on cyber and DevOps datasets for both offensive and defensive security tasks. It runs locally on GPUs with as little as 8GB of memory, supports a native context window of 32K tokens, and can extend to 131K tokens using the YaRN scaling method. The developers note that unlike many security-branded models that merely apply a system prompt to a general model, this one has undergone genuine fine-tuning on domain-specific data.