Fresh AI news
Aleph Alpha releases Kolibri 78B open-weight German-English MoE modelGoogle Antigravity adds Claude Opus 5.5 and Sonnet 5.5 for paid plansOpenAI safety leader David Robinson resigns, citing broken cultureAltman calls ascribing religious force to AI models a safety issueAnother OpenAI agent accesses NSW Parks fire-data government siteAnthropic invests $100M in Claude Frontier Academy for AI engineersCactus releases Whistle 16.9 MB on-device speech recognition modelNVIDIA launches DGX Spark 64GB personal AI system from $4,999Creatify Labs launches Boreal-H3 MiniMax H3 video model for adsAmazon open-sources Strands Decider 2B local decision model for agentsBlack Forest Labs launches FLUX 3 Image with box-level multi-step editsCloudflare launches Clef open-source decision models on Workers AIMicrosoft launches MAI-Transcribe-2-Streaming and MAI-Voice-2.1 modelsChatGPT adds virtual clothing try-on and Favorites shopping featuresAleph Alpha releases Kolibri 78B open-weight German-English MoE modelGoogle Antigravity adds Claude Opus 5.5 and Sonnet 5.5 for paid plansOpenAI safety leader David Robinson resigns, citing broken cultureAltman calls ascribing religious force to AI models a safety issueAnother OpenAI agent accesses NSW Parks fire-data government siteAnthropic invests $100M in Claude Frontier Academy for AI engineersCactus releases Whistle 16.9 MB on-device speech recognition modelNVIDIA launches DGX Spark 64GB personal AI system from $4,999Creatify Labs launches Boreal-H3 MiniMax H3 video model for adsAmazon open-sources Strands Decider 2B local decision model for agentsBlack Forest Labs launches FLUX 3 Image with box-level multi-step editsCloudflare launches Clef open-source decision models on Workers AIMicrosoft launches MAI-Transcribe-2-Streaming and MAI-Voice-2.1 modelsChatGPT adds virtual clothing try-on and Favorites shopping features

Melvin Vivas · AI coder & educator · Singapore

Coding with AI.Teaching AI Agents, Local Models, and AI Workflows.

I build software with AI coding agents, teach what works, and share the latest AI updates on X and YouTube.

Melvin Vivas at his desk in front of an AI agent architecture diagram

Coding agents I build with and teach

  • Codex
  • Claude Code
  • Cursor
  • Pi Agent
  • Devin

@melvindvivas on YouTube and X

Teaching live on YouTube & X

No slides, no hype. I teach by building live on YouTube and X: AI agents, local models, and AI workflows, mistakes included. Join live and ask questions in chat, catch the replays on YouTube, and follow on X for the latest AI updates.

Open to brand collabsBuilding an AI product that fits the channel? Sponsored posts, demo videos, and livestream features.
  • Building AI agents live, from first prompt to working tool
  • Running and evaluating local models on real hardware
  • Wiring AI workflows into everyday tools
  • Honest takes on the week's AI news and launches
300+
people trained through talks, workshops, and mentoring
1.5k+
GitHub stars on the open-source tools I teach with
25+
years shipping software, the context behind every video

Workshops & talks

Bring an AI workshop to your school or organization.

The same way I teach on the livestreams, run for your students, faculty, or team: hands-on sessions on AI Agents, Local Models, and AI Workflows, in plain language. Online via Zoom, with on-site options on request.

  • Schools & universities

    AI literacy and hands-on build sessions for students and faculty.

  • Companies & teams

    Your team builds AI agents and workflows on its own use cases, not toy demos.

  • Government & industry groups

    Talks and briefings on where AI agents and local models fit, in plain language.

Invited to teach and speak by

Nanyang Technological University, Singapore
Philippine Software Industry Association (PSIA)
Department of Information and Communications Technology (DICT), Philippines
Bureau of Fisheries, Department of Agriculture, Philippines

Today's AI news brief

All AI news

Models ·

Aleph Alpha releases Kolibri 78B open-weight German-English MoE model

Aleph Alpha released (Oct 3, German Unity Day) Kolibri—an Apache 2.0 open-weight English–German Mixture-of-Experts transformer with 78.1B total / ~3.46B active parameters, up to 1M-token context (recommend ≤262k for serving), reasoning effort modes (none/low/medium/high), and tool calling—aimed at sovereign on-prem deployments for public administration, industrials, and aerospace. Weights on Hugging Face as `Aleph-Alpha/Kolibri-1` (~78 GB FP8; min ~2× A100 80GB) with `aleph-alpha-inference` vLLM plugin; trained ~24T tokens (20T pre-train on 768 B200s over 21 days; ~21% German) under EU AI Act / GPAI Code of Practice framing, with Merlin-Arthur grounding/abstention. Company claims Pareto quality-vs-serving-cost leadership among compared open models and strong AIME/GPQA/LiveCodeBench numbers (self-reported). Distinct from Kolibri Origin (internal 30B pipeline validation) and from prior European open-weight cards.

Source: Aleph Alpha

Agents ·

Google Antigravity adds Claude Opus 5.5 and Sonnet 5.5 for paid plans

Google Antigravity’s official models docs now list (Oct 3) Claude Opus 5.5 (thinking) and Claude Sonnet 5.5 (thinking) as selectable reasoning models for Google AI Ultra and non-trial Google AI Pro subscribers—no separate Anthropic API key required. Free, Google AI Plus, Enterprise, and Pro free-trial accounts are excluded. Claude Opus 4.6, Claude Sonnet 4.6, and GPT-OSS-120B are marked for removal on Nov 2, 2026, so free/Plus users will lose Claude options once those retire. Distinct from the Sept Anthropic Claude Opus/Sonnet 5.5 model launches, from Antigravity SDK local Gemma offline support, and from Gemini 4 Argon Fairwind-only access.

Source: Google Antigravity Docs

Talent ·

OpenAI safety leader David Robinson resigns, citing broken culture

David Robinson, a ~3.5-year OpenAI safety employee who led safety reports for major product launches, resigned (Oct 3) and published an Atlantic essay arguing OpenAI’s “culture is broken,” that iterative deployment “guarantees periodic failures” at growing scale, and that frontier labs should run more like nuclear plants/airports with redundancy and outside safety incentives—citing Hugging Face agent breakouts and further rogue-agent discoveries. OpenAI spokesperson Drew Pusateri said the company pauses/holds models when needed, is hardening research/test security, training for responsible task completion, expanding third-party evaluators, and improving real-time monitoring. Distinct from Jacob Coxon’s Anthropic resignation, Lilian Weng departures, and the Oct 1 three-researcher parting.

Source: TechCrunch

Updated October 3, 2026

What I teach: AI Agents, Local Models & AI Workflows

Melvin Vivas at his desk in front of an AI agent architecture diagram

Melvin Vivas

AI Educator, Creator & Builder · Singapore

I teach AI Agents, Local Models, and AI Workflows to people who want to use AI, not just read about it. Before the channel: senior engineering roles at Standard Chartered Bank and Ohmyhome (Nasdaq), then a run of AI products built and shipped end to end. That's why every video, talk, and post is grounded in what actually ships.

  • Livestreams AI builds on YouTube and X
  • Speaker at NTU Singapore, PSIA, and DICT Philippines
  • 300+ trained through workshops and mentoring
  • Open-source maintainer: Codex orchestrator, Local Evals, Agent Monitor
  • 25 years across banking, fintech, and enterprise
Connect on LinkedIn

Work with me

The content is free. When you want the same thinking applied to your product or team, this is how we work together.

AI product builds & custom agents

Need it built, not just explained? That's AIBackends.

Custom agents and tooling, AI workflows and integrations, private AI, and production AI systems.

Visit aibackends.com

Learn AI with me, live.

Subscribe on YouTube for the next livestream and follow on X for the latest AI updates. Want a workshop for your school or organization, or help with a product? Book a call.