USD.AI just scored backing from YZi Labs, fueling its push to turn AI hardware into yield-bearing stablecoins. Definitely one to watch out for.

The AUBRAI token, powering the world’s first decentralized AI agent for longevity research, is now live on Base.

Maiga has partnered with BNB Chain for a $45,000 giveaway, with 1,000 winners set to claim rewards.

Nvidia beat expectations again, with Q2 revenue up 56% to $46.7B and profit at $26.4B. It also guided for ~$54B next quarter, showing AI demand is still surging.

The only soft spot was data center sales at $41.1B, slightly under forecasts because of blocked H20 chip exports to China. With export restrictions now easing, analysts expect Nvidia could quickly unlock another $2–5B in revenue.

Meanwhile, Blackwell chips now dominate data center sales, gaming grew 49%, robotics nearly doubled, and the board approved a $60B stock buyback.

The buyback seems bearish to me, a sign Nvidia has too much cash and not enough places to deploy it. But the underlying growth story remains explosive.

Numerai just landed $500 million in capacity from JPMorgan Asset Management.

Ok to be clear this doesn’t mean $500M was wired. It means JPM can put up to $500M into Numerai’s strategy, now or later.

Numerai is not your typical hedge fund with PhDs locked in a Bloomberg dungeon. Numerai runs an open tournament where thousands of data scientists around the world submit stock predictions on anonymized financial data.

No CFA required. Anyone can download the obfuscated dataset, train a model, and send predictions through an API. If your model is good, you stake Numerai’s token $NMR on it and earn. If it’s bad, you lose your stake. Skin in the game, quantified.

All those predictions get fused into one giant “Meta Model” that drives the actual trades. Right now, about 1M NMR (~$17M) is staked, and on an uptrend, which means there’s a live, on-chain market measuring confidence in different models.

The experiment is working. Numerai returned +25.45% net in 2024 with a 2.75 Sharpe, which beats plenty of elite quant funds. Assets under management have exploded from $60M three years ago to $450M today, and the team just did a $1M token buyback to tighten alignment with its contributor community.

The rule is simple: if a model performs, it stays. If it fails, it vanishes. No committees, no politics.

After almost ten years of quiet building, Numerai is looking less like a quirky experiment and more like a serious new architecture for quant finance.

Reinforcement learning needs environments. They’re the “worlds” where agents learn by doing, whether it’s playing a game, writing code, or holding a conversation. Without them, there’s nothing to interact with. Just math in a vacuum.

Big labs know this. That’s why they’ve spent millions building private, high-quality RL environments, and kept them behind closed doors. If you’re not sitting inside DeepMind or OpenAI, good luck.

This week, a new option dropped.

Prime Intellect launched the Environments Hub, a platform where anyone can build, share, and run RL environments. In just the first week of private beta, 30+ contributors joined, including Arcee AI, Hud, WhyPhy Labs, and Groq.

The Hub gives you the infra to build environments, run evaluations, and plug everything directly into Prime’s open-source training stack.

It’s designed to support their long-term plan: training INTELLECT-3, an open, agentic model built entirely from community-driven environments.

Why this matters: environments are the new bottleneck in open AGI. The best models won’t get smarter without better places to learn.

Small teams and indie researchers can finally access the kind of tools that used to be proprietary. It’s still early, but the direction is clear: open-source AGI just got a new playground.

Customer service eats billions every year and nobody likes it — not customers, not companies, not the humans stuck in the call centers.

Sierra thinks it can kill the queue entirely.

It’s a full-stack Agent OS that replaces call centers with AI agents built to deliver outcomes, not endless “your call is important to us” loops. Backed by the ex–co-CEO of Salesforce and Google’s former Labs chief, it’s already landing billion-dollar enterprises.

This week, our deep dive breaks down how Sierra works and what makes it one of the most serious applied AI contenders right now. 👇

It highlights the top picks for images, video, audio, and 3D.

Now in its fifth edition, the a16z Top 100 is the Billboard Hot 100 for AI apps. People fight to be on the list.

He tells us how to squeeze more from tiny data and give the model memory long enough to remember what you said last week

Good time to buy.. soon?

David Beckham’s shop dropped an ad that featured an AI Aryna Sabalenka and pulled 35 million views in a day!