AI大佬卷入政治口水战

RT @NickKristof: .@elonmusk says that no one can name a person who died from his aid cuts. In fact, I've met the kids who are dying, and I'…

RT @sam_d_1995: it should probably be a bigger story that there was a normal bipartisan America 250 organization but then Trump decided to…

RT @Forbes: One study found Musk’s closing of the U.S. Agency for International Development could cause the deaths of millions of children.…

RT @noturtlesoup17: Just in case you thought it was all three dimensional. No! https://t.co/t122C6o7FE

RT @Tom_Winter: Better than a clock when it comes to knowing when it’s 5pm on Sunday:

RT @SteveRattner: Musk is right that many problems can't be solved just by "throwing money" at them. But US aid to fight AIDS, child hunge…

We’re introducing GeneBench-Pro, a research-level benchmark for a harder kind of AI progress: how well agents can navigate messy biological data, choose the right analysis path, and make judgment calls that real computational research depends on. https://t.co/AsilnnSxnE

RT @aljhlester: A very damaged man who had the political power and the private wealth to save millions of lives but chose instead to allow…

RT @NickKristof: The discussion about @elonmusk's comments on humanitarian aid has mostly focused on our values, which seems right. But not…

RT @GoogleDeepMind: We’re shipping 2 major releases:
 🔘 Nano Banana 2 Lite: our fastest and cheapest Gemini Image model 🔘 Gemini Omni Flash…

RT @ClementDelangue: Open-source AI is booming, massively impactful for progress, competition, transparency & orders of magnitude less dang…

Today, we give robots a /skills library that self-evolves and compounds indefinitely! Introducing ASPIRE: a robot solving its 100th task is no longer as clueless as solving its first. Coding agents observe multimodal sensory traces from simulation and real robots, launch an evolutionary search over control programs, and distill the best know-how into an ever-expanding library. ASPIRE is a new type of continual learning: "training" is skill refinement instead of gradient descent. "Trained model" is a repo of sensorimotor skills instead of floating weights. “Distributed training” is a panel of agents each practicing a different skill instead of sharded minibatches. Here's the beauty: ASPIRE gives the tired terms "sim2real transfer" and "cross-embodiment transfer" a whole new meaning. Bridging the sim-to-real gap is notoriously brutal. An end-to-end policy has to swallow both the visual shift (sim looks toyish next to a real camera) and the subtle contact physics it never quite gets right. ASPIRE sidesteps the mess, because it doesn't ship pixels or weights across the gap, but ships the know-how. The robot still has to practice in the real world, not zero-shot, but it gets there way faster because it isn't rediscovering the strategy from scratch. Same for going single-arm to bimanual hardware, which usually requires new data and retraining from zero. ASPIRE achieves up to ~10x cut in "transfer learning” tokens (yes, tokens are the new unit of *training* compute ;) Check out our gallery of 150+ tasks and 90+ skills the robots taught themselves, all on the website! Kind of wild that we can ship the "learned weights" as an HTML page rather than a GGUF. We'll open-source the full stack so your own robot library starts compounding from ours! Deep dive in thread:

Harnesses matter way more than people think. Cline ran a couple of experiments on a set of coding tasks using GLM 5.2: • 57.3% using their harness with reasoning turned off. • 68.5% with their harness with reasoning turned up. That's a difference of 11.2 percentage points! Same model, same set of problems. The difference stemmed from how the model was driven by the harness. Current open-weight models are way more capable than we think. They aren't the bottleneck anymore. We need better harnesses.

RT @KenRoth: Trump's Board of Peace plans to grant itself legal immunity for itself from “any arrest, detention or legal proceeding," accor…

RT @KenRoth: Putin's cannon fodder: "The average life expectancy of a new recruit—from arrival at a training ground to death in a combat zo…

End-to-end example of a long-running AI agent that pauses, resumes, and never loses context. It simulates the onboarding of a new employee. Here are 3 patterns you'll learn from this example: 1. How to implement a durable state machine that persists over time. 2. How to build event-driven agents that stay dormant until they receive a webhook event. No active polling or blocked threads. 3. How to build multi-agent delegation instead of relying on a single agent to do everything. You can deploy this example in the Gemini Enterprise Platform: https://t.co/vPH9Xd6UsO You'll find a link to the GitHub repository with the complete source code and a complete explanation of this example in this article: https://t.co/tm2WDLKI8v Thanks to the Google Cloud AI team for partnering with me on this post.

Oh wow, this makes a ton of sense: Scout is a platform that automatically creates agents based on a KPI you want to measure. You aren't building the agents; the platform is. Your input is not code; it's a goal. This is going to be a hit. It makes so much sense when you see it.

How to measure whether using an agent is worth it: Instead of focusing on whether agents can do this or that, focus on "value per token dollar". Basically, take the value an agent produced and divide it by what you spent on tokens to produce it. Treat this value as a ratio: • Below 1, the agent costs more than it returns • At 1, the agent pays for itself • Above 1, you can build a business using the agent Two different agents can run the same model and spend the same number of tokens, but create completely different amounts of value. @matrix_build is the first team I've seen using this metric.