☀️🛰️ Google’s “Project Suncatcher”: exploring AI compute… in space

Google outlined a research design that puts solar-powered TPUs on satellite constellations with free-space optical links to one day scale ML off-planet. Early-stage, but the architecture paper is public.

Why it matters: If space-based compute matures, expect shifts in latency, cost, resilience—important for startups and creators relying on cloud AI.

🍏🤝 Google Gemini rumored for the new Siri… Again!

Bloomberg reports Apple is nearing a deal to use a custom Gemini model (~1.2T params, per reports) to power next-gen Siri/Apple Intelligence. Apple would still run its own models for some tasks.

Why it matters: If confirmed, expect a faster pace of assistant features and more demand for private cloud compute.

🦾 XPENG unveils next-gen “IRON” humanoid robot

At its AI Day, XPENG demoed a highly human-like IRON prototype with smooth gait and life-like proportions; part of a broader “physical AI” push.

Why it matters: This signals that embodied agents are moving from lab to public demos - opening new product/service ideas at the digital-physical boundary.

🔐💘📱 Tinder is testing AI that learns from your Camera Roll (opt-in/Or Out)

“Tinder Chemistry” analyzes your photos plus interactive prompts to improve match quality; tests start in New Zealand/Australia.

Why it matters: This is the next wave of personalization - and it raises real questions. If your app touches user photos, you’ll need to be explicit about consent, scope (which albums, which signals), and retention (how long, where stored, how to delete). It’s a preview of what many consumer apps may try next.

Think it through (and tell us what you think):

- Where’s your line between useful and too invasive personalization?

- What would “clear consent” look like to you - one big toggle or granular album-level controls?

- Should apps analyze images on-device by default, and only upload with extra consent?

Read more

Hit reply: Would you opt in? Why/why not? 👇

⭐ Kimi-K2 Thinking (Moonshot) 🧠🔧

🧠 A new open-source “thinking” model tuned for multi-step reasoning and stable tool-use across long chains (claims: 200–300 calls), positioned as an agentic backbone.

✅ Strengths vs. many OSS baselines

Built for deeper step-by-step reasoning, tool calling, and long-horizon tasks; Moonshot cites strong scores on HLE and BrowseComp benchmarks.

🎯 Why it matters: If you’re building agents that must reason, browse, call tools, and keep context over many steps, this is worth a trial alongside your current stack.

Link: Project announcement + repo details.

Write to us, what did you think of today's edition?

🤖🤖🤖🤖🤖 - Love it!

🤖🤖🤖 - It's ok...

🤖 - Needs Improvement on …