Adelaide → NYC. Our co-founders met former SA Premier Steven Marshall to talk about how the American Australian Association connects Australian companies with US markets and partners.
#Metacognition#ArtificialIntelligence#AustraliaUS
For the first time, an AI agent can follow verbal navigation directions at human-level performance.
Metacognition’s NavGPT-3 hit 90.4% on Room-Across-Room - outperforming leading systems from Alibaba & Mistral AI.
And we’ve released the model + code publicly. https://t.co/TZ2WRf8kNM
Read more here: https://t.co/QAQGpfoJ4i
When code fails, you roll it back. When a robot spills your coffee, you can't ☕
Our Chief Scientist @sgould_au on what reliable-code principles mean for robots acting in the physical world 👇
https://t.co/z2j3Gqt1uU
#PhysicalAI#Robotics#MetacognitionAI
We're headed for San Francisco 🇦🇺✈️🇺🇸
Our co-founders @AntonvdH, @PaulMetacogAI and @sgould_au are in the Bay Area this week as part of Main Sequence's Physical AI Roadshow, alongside six other Australian deep tech companies.
If you're in SF and building, investing in or deploying physical AI, come and say hi at the Physical AI Ecosystem Night on Thursday 15 October for demos, drinks and time to connect with the teams behind them.
RSVP here: https://t.co/628QPxCl2S
We're talking to robots. 🤖
#PhysicalAI #SFTechWeek #AustralianAI #Metacognition
What if the best token to evict is the token the future already knows?
Our new paper, Back from the Future, uses the model's future context to identify redundant tokens.
📄 Paper: https://t.co/1vIc1zRGZ8
💻 Code: https://t.co/8NmpZ0cuFL
#LLM#AI#LongContext
Meet some of the people making it all happen at Metacognition.
L–R: Stephen Gould, Cherie Gauci, Paul Dalby, Anton van den Hengel
We’re talking to robots. 🤖
#Metacognition#ContinualLearning#Adelaide
Why can't AI agents just... remember better? Turns out it's not one problem — it's memory, reasoning, and knowing what to forget, all tangled together. Breaking down the architecture agents actually need to learn over a lifetime.
https://t.co/pzUBtyYPKH
Can an AI be trusted with a safety switch? New EU rules demand independent sign-off for AI safety systems — but the test doesn't exist yet, and aviation's already ruled LLMs out. The fix: a hard-coded layer between the AI and the machine, not a smarter model.
https://t.co/nIXets2wR3
@metacogai just raised a $10 million seed round backed by Main Sequence Ventures and the South Australian Government to build a "Windows for robots" operating system. https://t.co/1z37g7uZYT
founder and chief scientist @AntonvdH is Chief Scientist of the Australian Institute for Machine Learning and a former Director of Machine Learning at Amazon, with an H-index of 92 across 400+ papers, alongside co-founder Stephen Gould, who previously co-founded Sensory Networks, acquired by Intel in 2013.
government backing this early in a hardware-adjacent robotics operating system is worth watching, that kind of institutional money usually signals a longer runway commitment than a typical seed round implies. View all funded startups at https://t.co/gq9A9xTgDw, free, no login
Memory and reasoning are more entangled than people assume. Here's a breakdown of the memory hierarchy behind agentic AI, and why lifelong learning is harder than it sounds.
https://t.co/pzUBtyYPKH
Hello world, we’re here! While today’s LLMs are primarily text prediction machines that are unreliable and have short attention spans, Metacognition is building a new architecture, which includes memory that persists, reasoning that adapts, and trust you can inspect.
Read more:
https://t.co/AGPvvnQope
#DeepTechWeek #AI #AustralianTech #Metacognition