Day one has been “slightly” chaotic, in the best possible way. I’ve never seen this many people at a conference. Industry, startups, government, students, press. Rooms were packed, with people standing along the walls and spilling into corridors. The surprise visit by Prime Minister Modi to the exhibition hall added another layer of disruption and energy.

Despite the chaos, the excitement is palpable. People are not shy. They walk up, challenge, pitch, debate. Journalists sit in sessions. Speakers are surrounded. It feels less like a conference and more like a national mobilisation moment.

What did I learn today?

𝗙𝗶𝗿𝘀𝘁, 𝘁𝗵𝗲 𝗰𝗼𝗻𝘃𝗲𝗿𝘀𝗮𝘁𝗶𝗼𝗻 𝗮𝗯𝗼𝘂𝘁 “𝗺𝗶𝗱𝗱𝗹𝗲 𝗽𝗼𝘄𝗲𝗿𝘀 𝘃𝗲𝗿𝘀𝘂𝘀 𝗱𝗶𝗴𝗶𝘁𝗮𝗹 𝗲𝗺𝗽𝗶𝗿𝗲𝘀” 𝗶𝘀 𝘃𝗲𝗿𝘆 𝗿𝗲𝗮𝗹. Countries are asking where in the AI stack they should compete. There are strategic dependencies. The harder question is what to build, what to control, what to steer, and what to depend on. I don’t think Australia has reached a stable consensus on that yet. The answer will partly depend on the science and technology we can genuinely differentiate in. That is also why we at CSIRO’s Data61 are very careful about where we place our bets, not me-too investment.

𝗦𝗲𝗰𝗼𝗻𝗱, 𝗜𝗻𝗱𝗶𝗮’𝘀 𝗹𝗮𝘆𝗲𝗿𝗲𝗱 𝗮𝗽𝗽𝗿𝗼𝗮𝗰𝗵 𝘀𝘁𝗼𝗼𝗱 𝗼𝘂𝘁. Rather than owning every layer, it has built population-scale digital public infrastructure and used it as a control layer: identity, payments, data sharing rails. Sovereignty here is orchestrated, not absolute.

𝗙𝗶𝗻𝗮𝗹𝗹𝘆, 𝘁𝗵𝗲 𝗔𝗜 𝘀𝗸𝗶𝗹𝗹𝘀 𝗱𝗲𝗯𝗮𝘁𝗲 𝗳𝗲𝗲𝗹𝘀 𝗺𝗼𝗿𝗲 𝘂𝗻𝘀𝗲𝘁𝘁𝗹𝗲𝗱 𝘁𝗵𝗮𝗻 𝗺𝗼𝘀𝘁 𝗽𝗮𝗻𝗲𝗹𝘀 𝗮𝗱𝗺𝗶𝘁.
Countries are investing heavily in “AI skills”. But some of that upskilling may already be backward-looking. Prompt libraries seem to have a half-life of months. Among experts, the gap is widening: some achieve transformational results on a task, while others struggle to make the same system work at a basic level. Online, “skills issue” has become a meme, often used to dismiss those who cannot replicate impressive demos. Is that fair? That suggests this is not just about typing better prompts. It may be about problem framing, system thinking, how to scale feedback loop and when to distrust which precise parts of the output. Those are new domain and cognitive skills, not surface tricks.

𝗔𝘁 𝗗𝗮𝘁𝗮𝟲𝟭, 𝘄𝗲 𝗮𝗿𝗲 𝗹𝗼𝗼𝗸𝗶𝗻𝗴 𝗰𝗹𝗼𝘀𝗲𝗹𝘆 𝗮𝘁 𝘁𝗵𝗲 𝗹𝗼𝗻𝗴-𝗹𝗮𝘀𝘁𝗶𝗻𝗴 𝗔𝗜 𝗺𝗲𝘁𝗮-𝘀𝗸𝗶𝗹𝗹𝘀 𝗵𝘂𝗺𝗮𝗻𝘀 𝗻𝗲𝗲𝗱 𝘁𝗼 𝗱𝗲𝘃𝗲𝗹𝗼𝗽, 𝗻𝗼𝘁 𝘁𝗵𝗲 𝗼𝗻𝗲𝘀 𝘁𝗵𝗮𝘁 𝗲𝘅𝗽𝗶𝗿𝗲 𝘄𝗶𝘁𝗵 𝘁𝗵𝗲 𝗻𝗲𝘅𝘁 𝗺𝗼𝗱𝗲𝗹 𝗿𝗲𝗹𝗲𝗮𝘀𝗲.

As usual, some of the most thoughtful conversations have been with the Australian team. Great discussion with Paul Hubbard on GovAI and APS AI plans.

I will be speaking here on Day two. Stay tuned.
𝗙𝘂𝘁𝘂𝗿𝗲 𝗼𝗳 𝗔𝗜 𝗚𝗼𝘃𝗲𝗿𝗻𝗮𝗻𝗰𝗲 𝘁𝗵𝗿𝗼𝘂𝗴𝗵 𝗣𝗮𝗿𝘁𝗶𝗰𝗶𝗽𝗮𝘁𝗼𝗿𝘆 𝟭𝟳/𝗙𝗲𝗯 𝟮:𝟯𝟬 𝗣𝗠 – 𝟯:𝟯𝟬 𝗣𝗠, 𝗠𝗲𝘁𝗮 𝗕𝗼𝗼𝘁𝗵, 𝗛𝗮𝗹𝗹 𝟯