Introducing Nested Learning: A new ML paradigm for continual learning that views models as nested optimization problems to enhance long context processing. Our proof-of-concept model, Hope, shows improved performance in language modeling. Learn more: goo.gle/47LJrzI @GoogleAI

another solid breakthrough from Google... they're introducing nested learning: "a new ML paradigm for continual learning" 'Hope' is a self-updating, long-context memory architecture that generalizes 'Titans' beyond two update levels, allowing continual learning without forgetting too early to say, but it seems a real step toward fixing the memory problem

Embrace the future through technology. XPENG meets every expectation with innovation, defining the next generation of mobility with confidence and vision. Join us in turning the future into reality. $XPEV

History was made today — IRON, our humanoid robot, faced its ultimate test: to prove it’s not human. On stage, He Xiaopeng powered IRON back on and cut open its leg, revealing the true mechanical core within. At XPENG, transparency drives innovation — and even when doubts remain, progress never stops. $XPEV

Elon Musk says, Neuralink could capture an approximate snapshot of a person's mind and upload it to an Optimus (robot) body It's not immediate, but it's possible and probably within 20 years

Tesla Optimus pilot production line in Fremont. Elon Musk has said the pilot line is aimed at up to 1mn units per year. Internally, the goal is $20k cost of goods per robot at scale, which would undercut many industrial humanoid efforts if achieved.

Grok Imagine prompt: She smiles and says “I will always love you”

wow.. this is actually insane Freepik just outdid themselves here I restored and colorized a 70-year-old family photo, then even turned it into videos all from just one photo and zero prompts full guide below: ↓

Sam Altman talks about when OpenAI could be led by an AI CEO. In just a few years, certain departments could already be entirely run by AI executives.

Initially, they consist of mechanical parts, later of synthetic nerves, bones, and tendons.

Time to pull a LOT of rabbits out of the hat

I’ve been using Voice in M365 Copilot every day, and it’s one of those features that quickly becomes indispensable at work. Excited for customers to try it out now.

A protein from tiny tardigrades may help cancer patients tolerate radiation therapy When scientists stimulated cells to produce a protein that helps “water bears” survive extreme environments, the tissue showed much less DNA damage after radiation treatment. news.mit.edu/2025/tiny-tardi…

🚨 Google just KILLED N8N. I built 10 AI apps in 20 minutes — no code, no logic, no cost. Google Opal is here… and it’s FREE. N8N is finished. Here’s why 👇 Want the full guide? DM me.

How to use ChatGPT to brainstorm 10x faster:

AI just killed “render time.” Forget waiting on renders or rewrites. Seedance 1.0 Pro Fast in Freepik turns prompts into cinematic 3D clips instantly. You describe the scene camera motion, lighting, vibe and 7 seconds later, it’s alive. Here’s how it works 👇

The paper says AGI should mean balanced skills across all key abilities. Simple averages hide weak spots, so they mislead about real generality. The old way of measuring AGI used a simple average of scores across different skills, like reasoning, memory, vision, and language. This means if a model was amazing at some skills but terrible at others, the good scores could still push its average up, making it look more “general” than it really was. The new method in this paper changes that by checking how balanced the abilities are, not just how high the average score is. Instead of one average, it measures many averages under different “strictness” settings, from forgiving (where strong areas can cover weak ones) to strict (where one weak area drags the whole score down). ---- Paper – arxiv. org/abs/2510.20784 Paper Title: "A Coherence-Based Measure of AGI"

I’ll just leave another remarkable achievement on a math problem here, with help from GPT-5 Pro (more to come from the biomedical front soon ☺️): “Starting from the key idea of the first tweet, we extended the convergence result to several related settings and resolved the main 42-year-old open problem, with ChatGPT doing most of the heavy lifting along the way. Overall, this entire journey took just a week, less than 30 hours of my time. ChatGPT’s assistance provided a significant speedup, and without it, I would most likely have given up after three days of slow progress. (As I did in the past.)”

ChatGPT is now at the level of solving some math research questions, but you do need an expert guiding it. This exercise was a lot of fun and was highly productive. I also feel I'm getting better at prompting ChatGPT. I'll also try other open and unsolved problems. (16/N, N=16)

Conclusion: I will continue to develop this project and publish the results in a respected optimization theory journal. I'll share updates and future parts (II and III) here on Twitter as the work progresses. (15/N)