EVA Tech BriefIT/DevOps

2026-05-30 - 7:30 PM PKT

Frontier LLMs

Google Drops 9 Demos of Gemini Omni and Gemini 3.5 in Action

Google showcased its latest models with real-world demos. Gemini Omni handles multimodal reasoning across text, audio, and video simultaneously. Gemini 3.5 focuses on faster inference with improved code generation.

GPUs and Silicon

Cloud and Infra

Pakistan Tech

WhatsApp Plus Launches in Pakistan - Meta's Fintech Play Begins

Meta rolled out WhatsApp Plus subscription at PKR 229/month in Pakistan. The cosmetic features seem trivial, but the real play is training 60M+ Pakistani WhatsApp users to make in-app payments - laying groundwork for Meta's fintech ambitions in the country.

HEC Digital Degree Attestation System Causing Nightmare Delays

The HEC's new paperless degree attestation platform is failing massively. Applicants report system crashes, lost payments, and weeks-long delays for what was promised as instant verification. The digital transformation that went sideways.

Security

ChatGPT Share Links Weaponized to Deliver Malware via Fake Outage Pages

Threat actors are abusing ChatGPT's content-sharing feature to display fake OpenAI outage pages that trick users into downloading malware disguised as the ChatGPT desktop app. The attack leverages the trust users place in chatgpt.com URLs.

PAN-OS GlobalProtect Auth Bypass (CVE-2026-0257) Under Active Exploitation

Palo Alto Networks warns that CVE-2026-0257 (CVSS 7.8) is being actively exploited in the wild. The flaw allows attackers to bypass authentication and establish VPN connections. Patch immediately if running GlobalProtect.

Dev Tools

Pinterest Cut AI Costs 90% by Gutting a Frontier Model's Vision Layer

At 620M monthly users, calling a frontier model per image isn't sustainable. Pinterest CTO gutted Qwen3-VL's vision layer, rebuilt it with proprietary embeddings - 90% cost reduction, 30% accuracy boost. The playbook for scaling AI at enterprise level.

MIT's MeMo Framework: Swap LLMs Without Retraining, Get 26% Performance Jump

Researchers built a modular memory architecture that encodes new knowledge into a separate smaller model, works with both open and closed-source LLMs. Teams can upgrade their base model without losing accumulated knowledge.