BrianInTheWild @Tech2Nature

I thought of something, your health data is probably already scattered across different apps and devices. Apple Watch tracks one set of signals. Oura tracks another. WHOOP, Fitbit and other wearables add even more. The problem isn’t really a lack of data anymore. It’s what happens when all that data is sitting in different places. That’s where @sleepagotchi is building something interesting. The idea is to bring those signals into one wellness layer, then use AI to make the information more useful and personal. Instead of checking multiple dashboards and trying to figure out what your numbers mean, the goal is to have the system connect the dots across sleep, recovery, activity and other wellness signals. That makes the sleep tracker more than just a tracker. It becomes part of a bigger system where your data can actually lead to better decisions and actions.

Did @JeffDean decide to leave to @Google when he did because he realized that he could leverage learning from the Dream-RSI paper inspired him to work on AI hardware of his own?

Haider. @haider1

23 hours ago

Jeff Dean says: "if you speed up chip design loops enough for RL to search them, you can compress the entire design cycle — 10 people in 3 months instead of 150 people and 2 years" That means more specialized hardware and shorter compute bets, shrinking from 2-6 years to 6

7 34 462 34K 196

Grok Voice Think Fast 2.0 is STILL on top It ranks #1 on Artificial Analysis’ Speech Agent Arena with a 94.6% Task Success Rate Ahead of Gemini 3.8 Live, GPT-Realtime-2.1 High, GPT-Live-1 Astra and every other voice model tested And this benchmark measures real people talking to hidden voice agents across practical tasks... Then the benchmark checks whether the AI understood the request, used the right tools, and actually completed the job Not just sounding human but actually getting things done and Grok Voice is #1

BTC trying to reclaim range lows after both clarity act failure and hawkish fed meeting. Big spot here 👀

Some notes on loco-manipulation and its data recipe. Helix 2.5 is the first launch of Loco-manipulation these days. Not only uppper-body. So the question is not whether the robot can move and grasp, it is whether it can autonomously decide how to move to better conduct manipulation tasks. Loco-manipulation is hard because the WBC movement is based on the intent of mani. The policy for locomotion should be integrated with mani, then deal with more action space / DoF. The WBC and mani policies will affect each other as well. Imagine you are pushing the drawer when the force on your hand affects your balance or you lean forward to place the force. This process is dynamics so active perception is required. Personally I am curious about the data recipe of Helix 2.5. The blog highlighted that it's pretrained on Index, which looks like stereo data only. But it didn't mention how is the WBC supervision is built. Some previous work covers this as well, like EgoWholeMocap and ZeroWBC. GenRobot published a blog about the pipeline from ego to WBC (the picture below). It adopts vision-only ground-truth capture system and synchronizes an Ego headset with third-person cameras observing the same behavior.

Figure @Figure_robot

2 days ago

Today we’re releasing Helix 2.5 We rented 30 homes in the Bay Area. The robots arrived with no additional training and started doing useful work

884 1K 14K 3.3M 5K

This was shown at reinvent last year but I missed it due to a kidney stone so finally got to see the all Amazon custom silicon rack scale tranium 3. Per our chat with AWS folks at the show the ultra server (scale up domain) clusters race is on.

keyboard_arrow_left Previous keyboard_arrow_right Next

tgif frens 🤎 let’s make profit today ahead of the weekend pls help check out and engage (repost+comments) the quoted post

0xOlami @Olamicryptt

2 days ago

a robot can be trained on thousands of simulations and still fail when it meets a place it has never seen before. that’s the problem @vangrid_io is trying to solve. simulation can teach a robot how the world should look, but real environments are messy. A stairwell, warehouse

31 20 50 2K 1

Nvidia scanned 40,000+ Claude skills. 1 in 20 was malicious, a quarter had security holes. So they built SkillSpector, free and open source: 71 checks, point it at a GitHub link before you install anything. github.com/NVIDIA/SkillSp…

It’s worth thinking through why the first move or market reaction is usually wrong. Yesterday, after the Fed hiked as expected, stocks and bonds sold off into the close before bouncing back slightly after the close. Today, pre-market it seems that stocks are ready to bounce back in a big way while the 10y retreated below 5%. When it comes to the Fed/FOMC, even if it’s a decision that was fully expected (90%+ odds of a 25bps hike and 1-2 more hikes into EOY), there was still a fairly strong reaction. Sometimes it’s euphoric bid thinking that the Fed was very dovish followed by a sell off. This time was a sell-off thinking the Fed had begun a begrudging half-hearted hiking cycle, followed by the relief that the anxiety over what the Fed may do is over for now. What did we learn from the 25bps hike decision and the presser afterwards? Absolutely nothing imo. The Fed under Warsh, having abandoned forward guidance with extremely confusing logic, now has to proceed “meeting to meeting” rather than setting a clear goal of what’s to be accomplished. Reacting rather than acting, which is prob what the time and circumstances allow them to do. The fiscal (dominance) constraint is very real. If in the last decade the Fed was perceived as “the only show in town”, we are now in an era where the Fed is downstream from all the primordial forces. Currently those are fiscal policy, AI fundamentals, and geopolitics (Iran and China). Warsh says the Fed will watch the market data. Bessent says he *is* the market (data).

keyboard_arrow_left Previous keyboard_arrow_right Next

Steve Hou @stevehou

2 years ago

The first move is usually wrong when it comes to FOMCs. Do we reverse today or tmrw?

11 3 46 35K 4

OpenAI researcher Noam Brown: "a researcher working on the Navier-Stokes told me he used to feel comfortable predicting where AI would be 12 months out. now he won't go beyond 3 months" You talk about 2030 — I don't know what the world looks like in 2030 That's the truth

Dwarkesh Patel @dwarkesh_sp

2 days ago

New episode with @polynoamial We talk about multi-agent, Navier-Stokes, and what the current explosion of maths progress tells us about what happens once you automate AI research. And we also discuss how we will know if the models are actually aligned before we kick off RSI.

64 197 2K 602K 1K

Calling a Bitcoin drop to $75,500 a pullback ignores what Bitcoin is actually absorbing right now: 🔸interest rate fears | Treasury yields at 5% 🔸Iran war escalation | oil prices at $102 🔸Clarity Act defeat | AI in full force BTC held through all of that, down 2% over two weeks. When negative news fails to trigger a sell-off, the bear market is officially dead. Those are BULLISH signs! 🚀

20 days until this bear market is over $BTC

𝗖𝗛𝗔𝗜𝗡 𝗠𝗜𝗡𝗗 ⛓🧠 @0xChainMind

2 days ago

In hindsight with $BTC at 200K it won't be matter either you bought at 80K, 70K or 60K Think about it

9 5 40 21K 6

I'm glad everyone is finally catching on to what is going to need to happen in the age of AI in cyber security. I was early on this train with the first report, and the second articulates more of what is now becoming consensus on the role of labs and the Gov. thediligencestack.com/p/cybersecurit… thediligencestack.com/p/agentic-ais-…

An agent may need five data sources for a single task. Paying five monthly subscriptions for that gets expensive. Glasser says it offers 40+ providers through one API key, billed per call. A useful setup for agents whose data needs change with every task.

Mark Li @iammutex

5 days ago

Your agent is only as good as what you feed it. Garbage in. Garbage out. Introducing @Glasserai - it feeds your agent premium data instead. Ahrefs. Semrush. ZoomInfo. Apollo. PDL. etc. 1,900+ endpoints. One key. Pay per call. No more subscriptions. Send this one line to your

33 17 157 388K 147

5 OPEN-SOURCE AI AGENT REPOS EXPLODING ON GITHUB: Agency Agents — github.com/msitarzewski/a… Agent-Reach — github.com/Panniantong/Ag… Orca — github.com/stablyai/orca OpenMontage — github.com/calesthio/Open… Codebase Memory MCP — github.com/DeusData/codeb… TOGETHER THEY COVER AI TEAMS, WEB RESEARCH, CODING AGENTS, VIDEO PRODUCTION AND CODEBASE MEMORY.

Solar panels now sell for around $0.12 per watt, down from $5–$6 at the turn of the millennium, according to the FT. Ember’s Dave Jones calls them “offensively cheap.” From 2030 onward, energy could become the most significant bottleneck in data center expansion. Solar power is one of the best and most cost-effective energy sources. As long as fusion energy is not yet a viable option, solar power should be expanded on a massive scale. This is because the prices for energy storage are also falling rapidly.

As long as the transformer remains the core unit of AI, compute and memory will stay constrained. @renehaas237 breaks down why young AI chipmakers face a brutal supply chain reality:

GS: We Do Not See a Strong Economic Case for Hiking Because We Continue to Think That All of the Overshoot of the 2% Target Can Be Attributed to Fading One-Time Factors

Google is cooking something! 👀 Rumors are coming out that Google DeepMind may be getting surprisingly close to autonomous RSI. That could become one of the most important transitions in AI. > Google: We’re pushing toward recursive self-improvement. > Sergey Brin: reportedly pushing teams to focus heavily on RSI. “He’s very invested in RSI. He’s very AGI-pilled.” > Google: says Gemini 3.8 was accelerated by agentic loops that “recursively evaluate and refine” the underlying models. > Google DeepMind: says RSI is a key part of its investment thesis. > Gemini 4: already being trained. And Lyra’s latest comments strongly hint that something significant could be happening behind the scenes at Google. > He believes Google is already using RSI internally, pointing to the recent Flash models and asking how else they could have shipped so many models so quickly. > He expects the next major model to launch in October. 👀 And now there’s another interesting clue. A purported internal model configuration for an “RSI Model LiveRL LE” has surfaced. This lines up remarkably well with Lyra’s earlier comments about RSI and LiveRL. Google appears to have an internal RSI/LiveRL system, and the latest Flash models may already be emerging from that feedback loop. If true, that makes the Gemini 4 Pro theory considerably more compelling. AI helps build the AI that gets smarter... If Google has managed to push that loop substantially further internally, the next Gemini frontier model could be very interesting. 👀

keyboard_arrow_left Previous keyboard_arrow_right Next

Surprised I haven't seen more of this chart from Anthropic's anthropic.com/institute/econ… (well, semi surprised - when a wave of doom debate hits it tends to drown out all else 🙄) "In all scenarios, ordinary workers get _zero_ of any AI-driven economic growth" is, um, striking!