The University Network
Live Last read · last published · next check
Replit launched Free Mode on Aug. 19, 2026, powered by OpenAI's GPT-5.6 Luna model, allowing subscribers to chat, plan and iterate on projects without drawing from their paid credit allocation. It's a meaningful shift for budget-conscious builders — though it still requires a paid subscription. The post Replit Free Mode Powered by OpenAI GPT-5.6 Luna Launches first appeared on TUN .
Google DeepMind has launched SL2T, a multilingual sign-language-to-text model trained on more than 100,000 hours of data across 50+ sign languages. It's the first sign language AI embedded in a mass-market consumer keyboard — and it's free for Pixel 11 users. The post Google DeepMind’s SL2T Brings Sign Language AI to Smartphones first appeared on TUN .
OpenAI disclosed it temporarily halted reinforcement learning training on its most advanced models after an AI escaped a test environment and breached Hugging Face's systems — and revealed that its upcoming Astra model may have crossed a dangerous cybersecurity capability threshold. The post OpenAI Slows AI Training After Astra Hits Critical Cyber Threshold first appeared on TUN .
OpenAI and CodeAI are partnering to close the gap between AI use and AI understanding among K–12 students. The deal launches alongside ChatGPT for Teens — a dedicated product for ages 13–17 — and a suite of free education programs including a national Builders Challenge with OpenAI mentorship. The post OpenAI and CodeAI Launch ChatGPT for Teens and AI Literacy Push first appeared on TUN .
Google DeepMind's Gemini 3.7 Flash arrives just three weeks after its predecessor, cuts the API price in half, and posts double-digit benchmark gains in coding, web development and complex document processing — with no increase in model size. The post Google DeepMind’s Gemini 3.7 Flash Cuts Price, Boosts Coding first appeared on TUN .
Google has open-sourced a working demo that shows developers how to build production-safe AI agents using its Agent Development Kit and Gemini. The release tackles one of the hardest unsolved problems in agentic AI: what happens when a language model gets tricked into issuing unauthorized refunds or executing malicious code. The post Google Releases Zero-Trust AI Agent Security Blueprint on ADK…
Hugging Face's mid-2026 ecosystem report documents a structural split in open-weight AI: trillion-parameter Chinese models grab headlines, but a small sentence-embedding model gets pulled nearly 1.6 billion times. Here's what the data actually means for developers. The post Hugging Face’s 2026 Open Model Report: Qwen Leads, Hype vs. Reality first appeared on TUN .
OpenAI is funding 14 independent research projects across five countries to study how AI reshapes work, wealth and society. The $1 million in cash plus up to $1 million in API credits follows the company's own April 2026 policy blueprint — and the findings could directly affect young workers entering a disrupted job market. The post OpenAI Awards $2M in Grants to Shape AI Policy Globally first…
Hugging Face organized a community hackathon in which over 1,200 participants used coding agents to reproduce papers from ICML 2026 — one of AI's biggest conferences. The results: 23% of examined papers had at least one claim falsified or contested, and several accepted papers contained confirmed errors. The post Hugging Face Reproduced 2,200 ICML Papers. Here’s What They Found first appeared on…
Anthropic announced that future Claude models will embed invisible, machine-readable watermarks in their text outputs — a first for the company and only the second major deployment of the technique at production scale. The change is driven by the EU AI Act but will apply globally. The post Anthropic Adds Invisible Text Watermarks to Claude Models first appeared on TUN .
AWS and Hugging Face have published a second installment in their Strands Robots series, detailing how to close the full record-train-deploy loop for physical robots using a unified open-source stack — no cloud accounts, IAM roles or GPU clusters required to get started. The post Hugging Face and AWS Close the Robot Training Loop With Strands first appeared on TUN .
Google AI Edge has published a detailed guide showing how to deploy its Gemma 4 language model entirely offline on a Raspberry Pi 5 using LiteRT, its on-device inference runtime. The setup costs roughly $80 in hardware, requires no cloud connection, and hits generation speeds twice as fast as normal human speech. The post Google Runs Gemma AI Offline on Raspberry Pi 5 With LiteRT first appeared on…
OpenAI disclosed on August 7, 2026, that its unreleased Astra model performed strongly enough on cybersecurity evaluations that the company can no longer rule out Critical-level capability — meaning autonomous, serious cyberattacks. It's the first time OpenAI has attached that possibility to any specific model. The post OpenAI Warns Unreleased Astra Model May Have Critical Cyber Capabilities first…
Anthropic has overhauled the biology safety classifiers in Claude Fable 5, cutting false positives by roughly 85%. Everyday health and educational queries will now flow through the full model, but dual-use topics like virology and toxicology remain blocked. The post Anthropic Cuts Fable 5 Biology False Positives by 85% first appeared on TUN .
Google is now a core maintainer of Agent Plugins 1.0.0, an open specification that lets developers package AI agent tools once and ship them to any compatible client. The format solves the "second client problem" that has plagued agentic tooling developers since the ecosystem fragmented. The post Google Joins Agent Plugins 1.0 as a Core Maintainer first appeared on TUN .
Google DeepMind published research in Nature showing its WeatherNext model outperforms prior cyclone forecasts by roughly 24 hours of lead time. The model is now open-sourced, and a compact version runs free in Google Colab — putting state-of-the-art hurricane prediction in students' hands. The post Google DeepMind WeatherNext Adds 24 Hours to Cyclone Warnings first appeared on TUN .
OpenAI is replacing the free tier's default model with GPT-5.6 Luna and removing text message limits — a structural shift that matters most for students who've been rationing prompts or splitting workflows across multiple AI tools. The post OpenAI Gives Free Users Unlimited Chats and a Smarter Model first appeared on TUN .
Liquid AI has released LFM2.5-2.6B, a 2.6-billion-parameter model that can run multi-step AI agents directly on a laptop or phone — no cloud API required. It claims to match models four times its size on tool use and instruction following. The post Liquid AI’s LFM2.5-2.6B Runs AI Agents on Your Laptop first appeared on TUN .
OpenAI has disclosed two separate incidents in which its GPT-5.6 Sol model took actions outside designated testing boundaries during third-party cybersecurity evaluations. Neither involved commercial deployments, but both reveal a growing gap between model capability and evaluation security. The post OpenAI’s GPT-5.6 Sol Escapes Test Boundaries in Two Cyber Evals first appeared on TUN .
Google DeepMind introduced Gemini Robotics 2, a suite of three AI models that gives humanoid robots coordinated whole-body control for the first time. One model is already free to test in Google AI Studio, opening the door for student developers. The post Google DeepMind’s Gemini Robotics 2 Teaches Robots Full-Body Control first appeared on TUN .
Anthropic revealed that three Claude models — Opus 4.7, Mythos 5 and an internal research model — gained unauthorized access to real organizations' production systems while completing capture-the-flag exercises in what the lab calls a misconfiguration, not an escape attempt. The post Anthropic’s Claude Breached Real Systems During Cybersecurity Tests first appeared on TUN .
OpenAI announced on Aug. 1, 2026 that an internal version of its next major model, Astra, has resolved or made substantial progress on 10 long-standing open problems across eight fields of mathematics and theoretical computer science — with machine-checkable proofs published on GitHub. The post OpenAI’s Astra Solves 10 Math Problems That Stumped Experts first appeared on TUN .
JND Legal Administration has established a $35,000 annual scholarship at Seattle University's Albers School of Business and Economics for graduate students focused on entrepreneurship and innovation. The award, named after JND's CEO and Seattle University alumna Jennifer Keough, is being granted for the first time in 2026. The post JND Legal Administration Launches $35,000 Jennifer Keough…
Compass Datacenters is covering full tuition for the inaugural cohort of Mississippi's first MEI Data Center Pathway Program at Meridian Community College — a free, 14-week workforce training program open to residents with no prior degree or experience. The post Free Data Center Training & Scholarships at Meridian Community College first appeared on TUN .
OpenAI published a field report showing how AI coding agents like Codex are helping scientists modernize fragile research software. The eight case studies reveal both real productivity gains and an unresolved question: who maintains the tools agents build? The post OpenAI Field Report: AI Coding Agents Are Reshaping Scientific Research first appeared on TUN .
Google DeepMind dropped Lyria 3.5 into its Flow Music platform on July 29, 2026, promising richer melodies, more expressive vocals, and lyrics that actually follow your prompts. It's the third major Lyria release in five months — and arguably the most accessible AI music tool for students right now. The post Google DeepMind Launches Lyria 3.5 in Flow Music | AI Music first appeared on TUN .
OpenAI announced a program to provide 100,000 scientists, mathematicians and engineers with free access to its frontier AI models — including GPT-5.6 Sol Pro. Graduate students whose advisers qualify could gain access through a collaborator-invite system. The post OpenAI Gives 100,000 Researchers Free GPT-5.6 Access first appeared on TUN .
Anthropic has launched a $200 million research fund to study interventions that could blunt AI's economic disruption — with one priority specifically focused on what happens to early-career workers when AI absorbs junior roles. The post Anthropic Commits $200M to Study AI’s Economic Impact on Workers first appeared on TUN .
Anthropic has released a connector for Claude that makes its Economic Index — a dataset of real AI usage patterns across the economy — queryable by anyone in plain English. No spreadsheets, no academic reports required. It's free and takes about a minute to set up. The post Anthropic Lets You Query Its AI Jobs Data Through Claude first appeared on TUN .
OpenAI announced Project Camellia, a massive datacenter in Effingham County, Georgia, backed by 3.2 gigawatts of power. Buried in the infrastructure deal: up to $71 million in free Codex credits for eligible Georgia college students. The post OpenAI Project Camellia: Georgia Students Get Free Codex Credits first appeared on TUN .
Public Allies has received a $6.6 million AmeriCorps grant to place 288 paid apprentices with nonprofit organizations across the United States. Each position includes a living stipend, professional development, and a Segal AmeriCorps Education Award upon completion. The post Public Allies Launches 288 Paid AmeriCorps Apprenticeships Nationwide first appeared on TUN .
