Demis Hassabis, CEO Google DeepMind laid out the clearest roadmap to AGI

1/ AGI won’t come from scaling alone. Demis Hassabis says it’s 50% scaling, 50% innovation. Bigger models matter, but new ideas matter just as much.

2/ Today’s AI is powerful but jagged. Gold-medal level at Olympiad math. Yet still fails basic logic and consistency tests. That gap is why we’re not at AGI.

3/ The missing ingredient isn’t intelligence. It’s reliability, reasoning, and self-awareness of uncertainty. AI needs to know what it doesn’t know.

4/ Hallucinations aren’t random. They often happen because models are forced to answer when they should say “I’m not confident.”

5/ AlphaFold showed the playbook. Solve a root problem once, unlock entire industries downstream. Now DeepMind is targeting materials, fusion, and climate.

6/ Fusion is the ultimate root node. Clean, abundant energy would reshape water, food, climate, and even space travel. AI could help crack it.

7/ Language models surprised us. They understand more about the world than expected. But language alone isn’t enough.

8/ That’s why world models matter. To understand physics, space, causality, and action, AI must experience worlds, not just read about them.

9/ Simulation is the next frontier. If an AI can generate a realistic world, it likely understands its mechanics.

10/ Drop agents into those worlds and let curiosity drive learning. Now you have infinite training data, created on the fly.

11/ This could be how AI learns like humans do. Exploration first. Understanding second. Generalization last.

12/ Hassabis believes simulation may also unlock science. Weather. Biology. Materials. Even the origins of life.

13/ Why simulations matter philosophically: If you can simulate something, you’ve understood it.

14/ That leads to the deepest question. Is there anything in the universe that’s non-computable?

15/ So far, we haven’t found one. Protein folding. Go. Complex biology. All computable.

16/ Consciousness might be next. AGI could become a mirror that shows us what, if anything, is unique about the human mind.

17/ If creativity, emotion, or dreaming are computable, machines may have them too. If not, we’ll finally learn where the boundary is.

18/ AGI isn’t just a tech problem. It’s an economic, social, and philosophical one.

19/ The industrial revolution took a century. AGI may unfold in a decade. The disruption will be faster and bigger.

20/ Hassabis’ core belief: The universe runs on information. And intelligence may be the ultimate way to understand it.

Anthropic will add 5 different starting points to its upcoming Tasks Mode: Research, Analyse, Write, Build, and Do More. Tons of granular controls

A new sidebar for tracking tasks' progress and working with Claude's context has also been added.

Meta introduced SAM Audio, the first unified model that isolates any sound from complex audio mixtures using text, visual, or span prompts.

This is a cool model, because always struggled with finding good scenarios that combine audio and vision, where audio plays a larger role than just "like language in vlms, but, you know, as sound wave instead".

VC firms are building their own AI tools to compete for the best startup deals. And for founders, that's changing the relationships game.

This summer, venture capitalist Aubrie Pagano snagged the chance to invest in a buzzy funding round with a major assist from AI.

For their crucial pitch meeting with a frontier science lab, Pagano brought a list of 50 high-value prospects – academics, pharma execs and former FDA leaders – and the exact route by which her firm, Alpaca VC, could connect its founders to each.

The startup made room for Alpaca to invest $1 million. It was only afterward that its founders found out that Pagano had used an agent from the firm’s proprietary AI system, known internally as Gordon, to help secure the deal.

Seemingly every VC firm has that partner (or several) who drafts LinkedIn ‘thought leadership’ posts in Claude, runs meeting notes from Granola through NotebookLM, or calculates market projections in a custom GPT.
But Alpaca, investing out of a $78 million fund, and a growing number of boutique and emerging VC firms are looking to compete – and punch above their weight with founders – by making outsized bets on building and investing through their own advanced AI tools.

These firms are fine-tuning their own models and setting up MCP servers, and managing long-running agents that automate entire processes, from back office reporting to their investment memos and content calendars.

At DVC, a $75 million fund that’s an early backer of Perplexity, AI recommendations have helped the firm write preemptive checks into some of the firm’s fastest-growing companies, like Higgsfield AI, just before revenue or valuations soared.

And at Topology Ventures, a frontier tech firm that raised a $75 million fund last year, managing partner Casey Caruso believes her internal AI CRM, called Fiber, is so good at predicting founder movements that it could raise millions in its own right.

Gemini 3 flash is out. The fast mode from the model picker in the GeminiApp - it’s shockingly speedy AND smart.

What an OP model. Also mind blowing how even just flash is competitive with the best models.

Science announced Vessel, a project focused on rethinking perfusion from the ground up, extending how long life can be sustained, and expanding what’s possible in transplantation and critical care.

Life-support technologies like ECMO can keep patients alive when the heart or lungs fail, but they aren’t designed for long-term use.

Vessel exists to close the gap between what perfusion technology is fundamentally capable of and how it is deployed in daily practice.

More about Science here.

Shunyu Yao, a rising star in AI agents and one of the key minds behind OpenAI’s Deep Research and Computer-Using Agent (CUA), has just been appointed Chief AI Scientist at Tencent.

Anthropic introduced a first-party plugins marketplace, making it easier to discover and install popular plugins.

Run /plugins to browse and batch install available plugins from the directory. You can install plugins at user, project, or local scope.

If you maintain a Claude Code plugin that you’d like to see in the marketplace, you can submit it to the team here.

Max users can now share guest passes with friends.

All Max users have 3 guest passes to share, and each can be redeemed for 1 week of free Pro access.

Run /passes to access your guest pass links. All 3 features + guest passes are now available. Run claude update for the latest.

It turns out that VLAs learn to align human and robot behavior as we scale up pre-training with more robot data.

In a new study at Physical Intelligence, team explored this "emergent" human-robot alignment and found that researchers could add human videos without any transfer learning.

AI agent Gauss autoformalized the proof of the Kakeya conjecture for finite fields

The proof Gauss wrote was surprisingly efficient, in just 6 hours.

The Kakeya conjecture was originally posed in 1917. It was solved in 2 dimensions almost 100 years ago. But just this year in 2025, Wang & Zahl solved it for 3 dimensions.

The Kakeya conjecture for finite fields was solved in all dimensions simultaneously by Dvir in 2008.

Dvir’s proof came as a shock to the math world.

Before, both Terence Tao (Fields medal 2006) and Jean Bourgain (Fields medal 1994) had been stuck making minor improvements on the problem for years.

At the time, Dvir was just a PhD student!

Google introduced Gemma Scope 2

-Largest open release of interpretability tools (over 1 trillion parameters trained!)

-Works as a microscope to analyze all Gemma 3 models' internal activations

-Advanced tools for analyzing chat behaviors.

Paper.
HuggingFace.

Anthropic released Bloom, an open-source tool for generating behavioral misalignment evals for frontier AI models.

Bloom lets researchers specify a behavior and then quantify its frequency and severity across automatically generated scenarios.

Google introduced A2UI: Agent-to-User Interface

- Protocol for agent-driven interfaces
- Enables agents to generate interactive user interfaces
- Open source

Researchers from U. Michigan, NYU, Princeton & U. Virginia presented Next-Embedding Prediction (NEPA).

Instead of reconstructing pixels, the model learns by predicting the next "embedding" in a visual sequence.

It outperforms complex methods, hitting 85.3% accuracy on ImageNet and excelling at segmentation, all with a simple, scalable approach.

GitHub.

Stripe Atlas 2025 Startups: Year in Review – Key Insights

Into Stripe's latest report on startups via their Atlas platform, and it's packed with exciting trends for 2025. Here's a breakdown of the essential highlights:

1. Explosive Growth in Registrations: Startup formations surged 36% YoY. Europe led the charge with a whopping 48% increase, likely due to easier US incorporation amid local red tape. Time for EU reforms?

2. Global Teams on the Rise: Multi-national founder teams are up 79% since 2017, thanks to remote work magic. Borders are blurring – talent knows no limits.

3. Faster Monetization Than Ever: New startups hit revenue milestones quicker: Median revenue in the first 6 months jumped 39% YoY. 20% snagged their first customer within 30 days, and top performers reached $100K revenue 11% faster (around 108 days). AI and stablecoins are supercharging this.

4. Polarization in Success: While averages are up, the top 1% grew revenue 67% faster. Shoutout to rockstars like Cursor AI and Lovable for insane traction. Winners are winning bigger.

This report shows 2025 as a year of acceleration in the startup world – more companies, quicker cash, and global vibes. If you're building something, Stripe Atlas is making it easier for founders worldwide.

First large-scale empirical study of how developers actually use AI agent frameworks.

Over 100 open-source agent frameworks have emerged on GitHub, collectively accumulating 400,000+ stars and 70,000+ forks. But 80% of developers report difficulties identifying which frameworks best meet their needs.

Researchers analyzed 1,575 agent projects and 11,910 developer discussions across ten major frameworks, including LangChain, AutoGen, CrewAI, and MetaGPT.

Here are the findings:

96% of top-starred projects use multiple frameworks together. Single-framework solutions no longer meet the complex demands of real-world agent applications.

The dominant patterns: orchestration + data frameworks (LangChain + LlamaIndex) and multi-agent + orchestration combinations (AutoGen + LangChain).

Not surprisingly, GitHub stars don't predict real-world adoption.

NVIDIA launched ALCHEMI Toolkit-Ops to accelerate chemistry and materials science simulations using machine learning interatomic potentials (MLIPs).

ALCHEMI combines the accuracy of quantum chemistry methods with the scalability of AI to enable large-scale atomistic simulations that were previously impractical, helping run faster, more accurate simulations for materials discovery and molecular modeling.

Artemis published an empirical analysis of stablecoin payment usage on Ethereum

The core insight: volume belongs to business.

P2P transactions account for 67% of all payments by count — but only 24% by volume. The remaining 76% of dollar volume flows through business-involved payments: B2B, B2P, and internal corporate transfers.

This isn't surprising if you think about it. Retail users send $50-500 transfers. Institutions move millions. But it does reframe the narrative: stablecoins are currently infrastructure for large players, not yet a mass retail payment rail.

Top 1,000 wallets account for 84% of total transfer volume. Despite all the decentralization rhetoric, stablecoin activity is heavily concentrated among exchanges, market makers, and corporate treasuries.

Payment activity consistently drops on weekends — a pattern typical for business operations, not peer-to-peer retail transfers. This further supports the institutional dominance thesis.

There's a notable spike in transactions below $0.10 — which makes no economic sense given Ethereum gas fees. Researchers flag this as likely bot activity and wash trading. Any serious analysis needs to filter this noise.

The report acknowledges a major limitation: fiat-to-stablecoin-to-fiat flows through intermediaries aren't captured on-chain. Payment processors bundle transactions, so the true P2P payment volume may be higher than what blockchain data shows.

What this means?
Stablecoins have found product-market fit — but primarily as B2B settlement infrastructure. The retail payment use case is growing (transaction counts doubled year-over-year) but remains a fraction of total value moved.
The open question: will stablecoins evolve into a true P2P payment layer, or will they remain primarily institutional plumbing?

The data suggests we're still in the infrastructure phase.

Physical Intelligence demonstrated a series of humanoid robot tasks like making peanut butter sandwiches, cleaning windows, peeling oranges, and washing pans modeled after Benjie Holson’s “Robot Olympics.”

Using their fine-tuned model π0.6, the robots autonomously tackled high-dexterity challenges that highlight Moravec’s Paradox: tasks humans find trivial are still incredibly hard for machines.

Their results show that fine-tuning large embodied models is essential training from scratch failed on all tasks.

Google DeepMind just released DeepSearchQA

A 900-prompt benchmark that evaluates AI agents on complex, multi-step web research tasks across 17 fields.