robotics was once said to lag 2-3 years behind AI due to hardware and sensor limitations but the gap has narrowed fast. today's robots can walk, run, handle complex objects, and perform complex tasks that were limited to labs what's missing now is the "brain" a general AI capable of understanding goals, making decisions, and adapting on its own

This makes perfect sense once one thinks about incentives: Binance wants BNB chain to eclipse ETH because CZ thinks he can be the next Ethereum Tether wants BTC to dominate ETH as they hodl a lot of BTC from the early days of Bitfinex Goldman and JP Morgan have no invested chain and are being replaced by Crypto payment rails, so they need to buy ETH, the most decentralized, neutral and reliable blockchain from clients' perspective ------ "Show me the incentive, and I will show you the outcome." - Charlie Munger ------ In the end, both BNB and BTC are going to lose to ETH and DeFi because the core property of ETH is incorruptibility “Wall Street doesn’t care about speed, they want $ETH.”- Tom Lee

The Industrial Revolution automated muscle over a century, giving society time to adapt. AI is automating cognition in under two decades. The crisis isn't the technology, but a transition so rapid it will overwhelm political systems.

With the rate AI/robots capabilities are accelerating, it’s hard to imagine this trend will reverse and we’ll somehow start creating tens of millions of new jobs(for people), no matter what new industries emerge. We’re automating intelligence itself, and it will eventually be able to do everything humans can do. This technological revolution is fundamentally different from all previous ones.

Grok 4 Fast has a massive 2M token context window that unlocks enterprise-scale reasoning It easily remembers ultra-long chats, huge codebases, or hundreds of pages in a single go The best part? xAI just updated it with massive performance gains: Reasoning mode: 77.5% ➝ 94.1% Non- Reasoning: 77.9% ➝ 97.9% Grok 4 Fast delivers near-flagship results with up to 98% lower cost

The top immunologist @DeryaTR_ just said the part everyone has been whispering; not even Grok Heavy or Gemini DeepThink comes close to GPT-5 Pro. And the wild thing is that this actually fits what people inside the ecosystem have been hinting at for months. Here is the uncomfortable truth. OpenAI is running internal models that are far more capable than anything public right now. The gap is real. The other labs are sprinting. OpenAI is leading. You can see it in the pattern. Sudden jumps in reasoning. Tool use that feels coordinated. Context windows that swallow entire bookshelves. Multimodal abilities that move from cute to genuinely useful. None of this happens without a deeper tier under the surface. Most serious estimates say OpenAI is roughly six months ahead. In this acceleration curve that is basically a full era. If every frontier lab really launches new models next month, this winter is going to be the loudest release season in tech history. Feels like we are living in the beta version of the future.

Thirty days in. Here’s what's been built and what’s next ↓

Bearish for the AI bubble: - Chinese tech firms can train frontier models pretty cheaply - They shoot straight to the top of the leaderboards, hugging face downloads, are open source - Get incorporated in wrappers like Perplexity Shows US foundational models have no moat

In a year or so this level of performance should be available on a 32b dense model (K2 is 32b active) at a cost of < $0.2/million tokens I don't think folk have that in their estimates

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

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

jensen huang is right china will likely win the AI race. they're releasing open-source models to weaken U.S. funding, while catching up in compute and next-gen models they already dominate in robotics and if AI levels out, they'll win through hardware strength

AI is eating all the jobs. Job postings fell 37% from the April 2022 peak, landing back at February 2021 levels. The entire pandemic hiring boom got deleted in thirty months while everyone argued about return-to-office policies. They laid off 300,000 tech workers over three years. Then AI made sure those jobs never came back. Stack Overflow cut 28% of staff because Copilot does what junior developers used to do. Duolingo stopped hiring contractors for work AI handles now. IBM froze 7,800 positions to automate instead. Software engineer postings down 49%. Web developers down 60%. Every company runs the same filter now before approving new headcount: "Can AI do this?" Just asking the question freezes hiring for months while they evaluate. New postings are only 4% above pre-pandemic levels even though millions more workers entered the labor force. Do the math. The BLS data is delayed from the government shutdown. When it drops, it will confirm what Indeed has been screaming. They fired the workforce, then AI replaced the hiring pipeline itself. You keep waiting for the AI job apocalypse to arrive. You already lived through it and didn't notice because unemployment stayed at 4.1%.

A few thoughts on the Kimi-K2-thinking release 1) Open source is catching up and in some cases surpassing closed source. 2) More surprising: the rise of open source isn’t coming from the U.S., as was still expected at the end of last year (LLama), but from China—kicked off in early 2025 with DeepSeek, and now confirmed by Kimi. 3)The sanctions against China aren’t achieving their intended effect; on the contrary, the chip restrictions are pushing Chinese scientists to become far more creative, essentially turning necessity into a virtue. 4)Taken together, this leads to the conclusion that China is not just becoming an increasingly strong competitor to the U.S., but is now catching up—and it’s no longer clear who will win the race toward AGI.

At a FDV of $1.26 billion, when ZKsync generates $120 million in fees per year, they will have a ridiculously low PE of 10 With a base 15,000 TPS per Elastic Chain instance, that appears highly achievable by creating Private L2s for Deutsche Bank, Abu Dhabi banks etc and running composable Perp DEXs All those interconnected chains will generate interop fees for ZKsync There's no limit to the number of parallel Elastic Chains that can be created- it's unlimited parallelism Few

No fraking way. These people have truly redefined efficiency. This is a new record and even surpasses the efficiency of DeepSeek’s R1 training. Holy moly.

Is DeepSeek still a startup? Technical validation gave it strategic immunity in China’s AI ecosystem. The company can now focus on pushing architectural boundaries instead of chasing market returns. This only works in China. A distinctly Chinese evolution.

Grok upgrades

Grok

BREAKING: Grok just got a major upgrade. Grok-4-Fast got a big boost. Massive jump in reasoning quality and completion rate. • Reasoning mode: 77.5% → 94.1% • Non-reasoning: 77.9% → 97.9% Grok just got a whole lot smarter. Great work xAI team.