Yep… this cat video is AI (unfortunately 😿). But it’s getting impossible to tell what’s REAL online anymore. Time to do something about it, maybe?
🤯 Featured: 16 Days: Needle In The Hashtag - MLAI Hackathon
📅 Saturday 29th November, Stone & Chalk, 121 King St Melbourne CBD
Join MLAI, University of Melbourne and the Australian Government’s eSafety Commissioner in the 16 Days hackathon to hunt hoaxes, hazards and hate in a synthetic social network. Build detection, safety UX and trust tools with real mentors, curated datasets and serious prizes. Ship a prototype that makes feeds safer for everyone, especially kids and at-risk users.
🎟 Grab your seat on Humanitix. Come for the coding, chaos, and the occasional “wait, did we just make social media safer?” moment.
Welcome to MLAI’s newsletter, where we spot the game-changers before they change the game.
From Claude Sonnet 4.5 rewriting the rules of coding, to ChatGPT stepping into personal shopping, to founders turning AI into your new productivity sidekick, this week is all about AI that actually gets stuff done.
🌏 Ecosystem Pulse
Claude Sonnet 4.5 just dropped…
For months, everyone’s been sprinting to GPT-5. Claude? Felt like a relic. But Anthropic just came back swinging with a claim:
SONNET 4.5 IS THE BEST CODING MODEL IN THE WORLD
So what’s actually going on? Let’s break it down.
⚰️ RIP Mega Models?
Here’s the plot twist: Sonnet 4.5 basically made Opus irrelevant.
Why? Because
Opus costs 10x more for a 2% edge.
Sonnet 4.5 gives you Opus-level power at the old Sonnet price.
Translation: Forget giant trillion-parameter flexes. We’re entering the era of “lean, mean, coding machines.”
💡 Where It Shines
Coding beast: State-of-the-art on Software Engineering Bench.
Focus king: Can work on a task for 30+ hours straight.
Speedy: Starts coding instantly (GPT-5 is more like that kid double-checking their homework before handing it in).
But….
UI work? Still mid. GPT-5 wins at building shiny front-ends.
The Real Game Changer ⚡
The model’s sick, but the product updates? Even juicier:
Checkpoints: Save progress & roll back anytime.
Native VS Code extension: No more jank.
Context editing + memory tools: Finally run long tasks without stuffing the whole prompt.
And here’s a key part 👇
Anthropic pulled a rebrand: Claude Agent SDK.
Quick explainer: an SDK (Software Development Kit) is basically a toolbox for developers, it gives them pre-built code, libraries, and docs so they can easily plug Claude into their own apps and workflows.
Cool move, but still closed-source though, so don’t get too comfy 🙃).
The Takeaway 🎯
Claude Sonnet 4.5 isn’t just “keeping up” anymore. It’s making the AI race interesting again.
Faster. Cheaper. More focused.
Maybe the start of the “post-super-model” era.
In short: GPT-5 just got company. 🥊
Talking ChatGPT… and it just went shopping 🛒
OpenAI turned ChatGPT into your AI personal shopper. U.S. users can now buy straight from Etsy sellers + Shopify stores - all without leaving the chat.
🤔 Why Should You Care?
1 in 10 users are already thinking about buying through ChatGPT.
Now you don’t need to jump between tabs, ChatGPT handles browse + checkout in one flow.
Merchants still get their data (thanks to Stripe), and only pay a small cut.
Translation: ChatGPT just slid into Amazon/Google’s DMs and said: “I can do your job, but chiller.”
🎯 The Takeaway
ChatGPT is no longer just a chatbot. It’s officially:
Your shopping buddy. Faster. Easier. Smarter.
Possibly the beginning of “RIP shopping carts as we know them.” 🪦
🎓 Learn in Less Than 1 Minute: Overfitting vs. Underfitting
Two classic traps: overfitting and underfitting. Know them, or your model will be… bad at life.
Think of teaching a kid to recognise dogs:
Show only golden retrievers → they think all dogs are golden retrievers → overfitting.
Barely teach them anything → they can’t recognise dogs at all → underfitting.
Overfitting = memorises too much, can’t generalise.
Underfitting = knows too little, can’t even get the basics right.
🎯 Takeaway: Find the sweet spot, teach your model enough to generalise, not memorise.
🛠️ New AI tools you should try
Exa-code - Tool that helps AI coding assistants find hyper-relevant web context to reduce hallucinations
Wan Video – Turn a still pic + your voice into a talking-head video. Perfect for whipping up quick video content without hiring a film crew.
Exa-code = AI fact-checker. Wan Video = AI mini-Hollywood.
🎧 Podcast Picks
🎙️ The future of software creation with Replit CEO Amjad Masad
Amjad Masad, co-founder and CEO of Replit, a platform now valued at $3B, has spent nearly a decade making programming accessible to everyone. In this talk, he explores the evolution of computing, the rise of AI agents that can write software on demand, and what this shift means for the future of work and the value of traditional software
📅 Upcoming Events
📅 StartSpace x Build Club & MLAI Monthly Saturday Co-working Day
Saturday 4 October, 10:00 am – 1:00 pm, State Library Victoria, Melbourne
Build Club and MLAI, are joining forces for a Saturday of collaboration, demos, and co-working at the State Library’s StartSpace.
Whether you’re here to build, demo, or just hang out with the community, you’ll walk away with fresh ideas, new connections, and plenty of motivation.
👉 Request to Join here – Registration subject to approval
📅 AI Leap - Make it happen with Google AI
October 15, 2025, One Darling Island Rd, Pyrmont NSW 2009, Australia
Google is partnering with SXSW and MLAI to bring you AI Leap - an exclusive event hosted at Google Sydney for founders, innovators and investors, showcasing AI’s vast potential.
🚨Lottery Alert! 🚨
If you’ll be in Sydney on 15 October, enter the lottery for a free ticket to the event. Fill in this form and maybe luck’s on your side!\
🔥 Startup Pitches to Watch
Out of 10 killer applications, we picked 2 founders to pitch their startups:
Speaker 1 – Chris Rickard, Userdoc
Chris is the founder & CEO of Userdoc, with 20+ years in software. He previously built and sold a 7-figure agency and is now focused on redefining how teams build with AI.
👉 Userdoc turns vague ideas into clear, structured specifications — cutting spec creation time by 70% and bridging the gap between humans and AI in software development.
Speaker 2 – Michael Reitzenstein, Tactile
Michael is a technologist with a background in video games (100M+ players reached) and medical devices. He was Director of Software Engineering at Seer Medical, leading 35 engineers to market in the UK & US.
👉 Tactile is a framework for building interactive, multi-modal applications - enabling real-time interaction with characters and objects via voice, text, and touch.
👉 Learn more about the event here. Come for the AI, stay for the snacks… and maybe meet the next startup unicorn.
📅 MLAI Social Drinks
Wednesday, 29 October 2025, 5:30 pm – 8:00 pm
Sometimes the best AI chats happen over a cold one 🍻 (no agenda, no slides, just vibes). Join us for a relaxed evening where Melbourne’s AI community comes together to share what they’re working on, swap war stories, and make new connections over drinks.
👉 Grab your ticket here. Because pretending you know AI over drinks is better than doing it alone. 🍹
📅 MedHack Part II: Electric Boogaloo
Saturday 7 February – Saturday 21 February 2026.
Build the hospital of the future in one weekend. Clinicians, builders, and optimists unite for one intense weekend: prototype real healthcare solutions, pitch to leaders, and maybe meet your next co-founder. Join MedHack: AI Hospital.
👉 Grab your ticket here. Because saving lives with AI sounds cooler than your weekend Netflix binge. 😎
📅 GenAI Evals, Monitoring, and Automated Prompt Engineering
Tuesday 28 October, 6:00 pm – 8:00 pm, Location TBA, Melbourne
GenAI is everywhere these days, flashy demos, viral AI apps, and endless “look what I made” posts. But anyone who’s tried to take it from a prototype to something people can actually rely on knows the truth: the magic is in the unglamorous, behind-the-scenes work. That’s where productionising GenAI comes in, the critical step that separates proof-of-concept demos from AI systems you can trust in the real world.
Expect practical demos, real-world case studies, and honest conversations about what actually works (and what doesn’t) when moving GenAI from experimentation to production.
Whether you’re a developer, researcher, or just keen on AI, you’ll leave with actionable strategies, fresh ideas, and a few “oh, that’s clever” moments you can actually use.
👉 Get your tickets here - come for the AI, stay for the “oh wow, that actually works” moments. 😎
💼 Jobs in Tech (This Week)
💼 Senior Full-Stack AI/ML DevOps Engineer (Temporary, 6 Weeks) – JobSciences Pty Ltd – Melbourne, VIC
Join a small, dynamic team to build and deploy end-to-end AI/ML pipelines, develop predictive models, and contribute to technical architecture. This 6-week contract offers hands-on experience with AI/ML systems, DevOps/MLOps, data visualisation, and the chance to mentor interns, with potential for extension.
📧 [Apply via email: irma@jobsciences.com.au]
💼 AI/ML Deployment Engineer – Workforce & Career Intelligence (Temporary, 6 Weeks) – JobSciences – Melbourne, VIC (Remote/Flexible)
Join a small, dynamic start-up team to finalise and deploy end-to-end AI/ML pipelines, integrate and polish existing front- and back-end systems, optimise MLOps/DevOps workflows, and ensure scalability, reliability, and usability. This 6-week contract offers hands-on experience with production-grade AI/ML systems, transformer-based predictive modelling, agent-based simulations, cloud deployments, and mentoring a small team of interns, with strong potential for extension.
Compensation: $10k–$13k AUD
Requirements: Strong Python + ML frameworks (TensorFlow/PyTorch/sklearn), transformer-based predictive modelling, synthetic data generation, React + front-end experience, MLOps/DevOps (Docker, K8s, CI/CD, Airflow, MLflow), Neo4j, cloud (AWS/Azure/GCP), and proven track record deploying production-grade AI/ML systems.
📧 [Apply via email: irma@jobsciences.com.au]
MLAI memes
If you’ve made it this far… legend status unlocked!! Don’t forget to like and share our newsletter so you never miss the next AI scoop, wild hackathon, or mind-blowing tool. Your future self (and your inbox) will thank you. Catch ya next week!
