I love adaptive reuse. At 150 Hayes, they are turning an ugly vacant office building to a beautiful 105-unit residential building (1,2,3 BR's) for ~$350/sf (construction cost) + $225/sf (acquisition cost). The project only pencils because SF waived affordable-housing requirements for office-to-resi conversions. $575/sf (construction + acquisition cost) seems like a pretty good deal for what looks like a nice building. The same guys did a great job on 100 Van Ness.

This robot is skateboarding completely blind. And yes... It’s autonomous. No vision or markers. Just proprioception, joint torques and some very clever state estimation. Aditya Bhatt (PhD at IAS / DFKI under Jan Peters) just dropped one of the wildest robot demos of the year. People in the comments are already losing it over the same questions: How the hell does it know where the deck is? Pure IMU? Contact models? What observations is the policy actually getting at inference time? They only trigger the tricks like a real-life video game. This is coming straight out of @Jan_R_Peters’ lab, @ias_tudarmstadt. The same Jan Peters who just sat down for a long, unusually honest interview about how German red tape cost the field a full five-year head start on humanoids. If you’re into real robot learning (not just pretty videos), both are worth your time. Skateboarder demo and work by @aditya_bhatt. Podcast: Ep 105 | Red Tape Cost Us a 5-Year Lead in Humanoid Robotics (w/ Jan Peters) Huge thanks to Jan Peters for the rare honesty, his humour and insights into the world of robotics and his own! Full Conversation: On Spotify: On YouTube:

Internet video is now robot training data! 🛜 @perceptroninc, the startup from the ex-Meta team behind the Chameleon multimodal models, just released Isaac 0.5, an open-source embodied foundation model. The same weights answer questions about a video, point to objects, track them over time, report task progress, and generate the robot's actions. What it learns from perception directly shapes what it does. To make a big model run fast enough for control, they built on something called Null Experts: each token can pick how many experts it activates, so the model spends more compute on hard parts of a scene and coasts through the easy ones. → 36B-parameter sparse model, trained across 35+ robot systems and 3T multimodal tokens → 62.6 on ScreenSpot-Pro grounding vs 54.8 for the strongest Qwen3-VL run → Does it at ~8.5× lower inference cost, 26.9 TFLOP per request against 228.4 → On a chess-manipulation task, one epoch on a single episode cut action loss 10.5× Perception and action sharing one brain, with compute that flexes to the task, that's a genuinely different shape for a robot model. And it's fully open, weights, training code, inference via LeRobot. Here's the blog: ~~ ♻️ Join the weekly robotics newsletter, and never miss any news →

What is happening with the manufacturing sector in Chicago? The Chicago PMI dropped -10.5 points in August, to 47.1, the lowest level since December 2025 and returning to contraction in business activity. This marks the largest monthly decline since the 2020 pandemic and was also well below estimates of 58.3 points. New orders declined -15.4 points, while production fell -8.8 points, the largest monthly decline since December. Order backlogs plunged -12.1 points, to their lowest since November 2025, and remained in contraction. Meanwhile, prices paid rose +3.8 points, to their highest reading since February 2022, pointing to a stagflationary mix of weakening activity and rising costs. Manufacturing activity may be heading toward its next downturn.

ALIBABA'S LATEST AI DISCLOSURES SHOW A POTENTIAL STRUCTURAL ADVANTAGE OVER TENCENT — A PROFITABLE CLOUD BUSINESS THAT CAN HELP FUND AI MODEL AND APPLICATION LOSSES — WITH FISCAL Q1 AI-RELATED NET COSTS OF 8.2 BLN YUAN VERSUS TENCENT'S 10.5 BLN YUAN, AS ITS CLOUD ARM GENERATED 5.6 BLN YUAN IN ADJUSTED EBITDA AT NEARLY A 12% MARGIN.

ZEROBASE WEEKLY 8.24-8.30 ZBT traded in a relatively tight $0.075–$0.082 band through most of the week before a late bounce toward $0.083–$0.086 on August 30. Crypto market capitalization held in a $2.62T–$2.78T range after the prior week’s sharp expansion, settling near $2.63T–$2.70T by the weekend. The week was a classic digest-and-test sequence: residual bid from the prior 22% Bitcoin surge, a probe of resistance above $81,000, then a hawkish-policy fade. Bitcoin opened the week near $77,700 on August 24, advanced through $80,000, and printed a three-month high around $81,300–$81,455 on August 25–28 before reversing. The Friday close near $77,840 reflected a roughly 3% session drop after Fed Chair Kevin Warsh’s Jackson Hole remarks. By Sunday, Bitcoin had recovered into the $78,200–$78,800 zone, leaving the week modestly higher from Monday’s open but well off the highs. Ethereum moved in parallel, starting near $2,460, tagging the mid-$2,500s, sliding toward $2,420–$2,440 on August 28, and finishing near $2,450–$2,480. Derivatives confirmed the late-week de-risking. Friday saw about $488 million in liquidations, overwhelmingly longs, across nearly 98,000 traders. Open interest stayed elevated after the prior week’s short-squeeze, while funding on major pairs flipped from constructive to more cautious as hike odds repriced. Macro and geopolitics dominated the tape. The Jackson Hole symposium (August 27–29) was the focal point. Warsh, in his first keynote as Fed chair, called the 2% inflation target a “firm, fixed” objective, said forward guidance had “overstayed its welcome,” and warned that policymakers “have work to do” if underlying inflation does not move to target “clearly and at sufficient speed.” July PCE remained sticky at 3.7% year-over-year (core 3.3%). CME-implied odds of a September rate hike jumped from about 35% to around 57%. Two-year yields rose, the dollar strengthened, and risk assets faded into the Friday close. Other data mixed the picture: initial jobless claims fell to 203,000, supporting labor resilience, while July new-home sales dropped 10.5% under high borrowing costs. Trade friction intensified after the U.S. moved toward 50% tariffs on roughly $20 billion of Canadian goods following collapsed talks. Washington also widened Iran-related sanctions; oil still posted its first weekly decline in three weeks, with WTI near $83.40 and Brent near $89.30 as Hormuz-related risk was reassessed. Gold firmed toward $4,500. U.S. equities finished the week modestly higher despite Friday’s pullback. The S&P 500 closed Friday at 7,711.76 (−0.25%), the Dow near 53,560 (essentially flat on the day), and the Nasdaq weaker. Nvidia’s fiscal second-quarter print was the offset: revenue of $96.2 billion, data-center revenue of $89 billion, and guidance that produced an 8.7% single-session jump and a roughly $442 billion one-day rise in market value—one of the largest on record. That AI bid kept Communication Services and Tech among the week’s better sectors even as rate-sensitive names and the Russell 2000 lagged. Institutional flows remained the structural support, then cracked at the margin. U.S. spot Bitcoin ETFs took in $337.6 million on August 24, $314.4 million on the 25th, $232.1 million on the 26th and $242.2 million on the 27th, before a $201.8 million outflow on August 28 ended a nine-session, roughly $3 billion inflow streak. The August 24–28 trading week still netted about $924.5 million. ARK 21Shares led Friday redemptions (−$114.9 million), followed by Bitwise and a modest IBIT outflow. Ethereum ETFs did not follow: they added about $102 million on August 28 and extended a 10-day inflow run. Solana and XRP products also stayed in positive flow on the reversal day. Crypto Fear & Greed spent the week in greed rather than the fear readings of mid-August, oscillating roughly in the mid-60s to high-70s and ending near 69–76. On-chain data continued to show a holder split. Wallets with 100+ BTC added more than 39,000 BTC over the recent week, extending a 60-day accumulation of about 43,000 BTC (larger cohorts above 10,000 BTC added even more over that window). Retail-sized wallets (0.1–1 BTC) remained in distribution, with accumulation-trend scores near −0.98. Exchange flows flipped positive on the Friday dump (net coins onto venues) and quieter thereafter. A large share of supply remains dormant; long-term holder behavior stayed more constructive than the short-term cohort that sold into strength. In summary, August 24–30 was a consolidation week after Bitcoin’s strongest weekly dollar advance in years. Spot prices tested $81,000, ETF demand stayed heavy until Friday, and whales absorbed retail supply. The Warsh speech reintroduced rate-hike risk, ended the BTC ETF streak, and forced long liquidations—without breaking the higher range established the week before. Sticky 3.7% PCE, U.S.–Canada tariff escalation, and residual Middle East energy risk keep the macro overlay two-sided.

Rivian has just unveiled the 2027 model year R1S and R1T, and they have introduced the most requested interior feature ever. Here's what's new: • R1S now has optional second-row captain’s chairs for the first time ever. Both seats are heated, fold flat, have armrests that stow flush, along with a push-to-release cup holder that folds down from the inside edge of the seat, and full car seat tie-downs. • Gear pass-through: Long items such as skis and surfboards can fit through the center while all six seats remain occupied • New Esker Silver exterior color comes to both R1S and R1T after debuting on R2. • New simplified trim structure: Rivian is replacing the Dual/Tri/Quad motor-based naming with Premium, Performance, and Quad trims. • Premium trim: Dual-motor AWD with 533 hp, 610 lb-ft, and 0–60 mph in 4.5 sec. • Premium power upgrade: Optional upgrade boosts output to 665 hp, 829 lb-ft, and 0–60 mph in 3.4 sec. • Premium Long Range: Up to 410 miles EPA-estimated range for R1S and 420 miles for R1T. • Performance trim: Tri-motor setup producing 850 hp and 1,103 lb-ft, with 0–60 mph in 2.9 sec and up to 371 miles of EPA-estimated range. • Quad remains the flagship trim: 1,025 hp and 1,198 lb-ft, with the R1T hitting 0–60 mph in 2.5 sec and the ¼ mile in 10.5 sec using Launch Mode; up to 374 miles of range. • New interior naming system designed to match Rivian R2. • Slate Sky Signature interior: New grey/plaid interior with walnut wood accents for Performance and Quad. • Black Mountain Signature interior: Warm ash wood accents for Performance and Quad. • Ocean Coast Signature interior: Light coastal-inspired interior with weathered driftwood accents for Performance and Quad. • Premium interiors: Black Mountain and Ocean Coast are offered with dark ash wood accents. • Seven-seat R1S remains available with the traditional second-row bench on Premium interiors and Black • Mountain Signature on Performance and Quad. • New 20-inch Adventure All-Season wheel/tire package for R1T, designed to improve off-road traction while maintaining strong on-road range. • 20-inch Adventure wheels are standard on R1T Premium and a no-cost option on R1T Performance. • New Goodyear Adventure All-Season tire delivers EPA-estimated ranges of 300 miles on Premium, 380 miles on Premium Long Range, and 348 miles on Performance. • Esker Silver replaces LA Silver as the included/no-cost paint color. • Esker Silver is available across all R1 variants and interiors. • 2027 R1S now starts at $83,990. • 2027 R1T now starts at $79,990. More photos in thread below:

105 years ago, the Communist Party of China had fewer than 60 members. Today, that number has grown to 101 million with 5.4 million primary-level Party organizations. To keep this vast Party united and effective is a miracle. We’ve done it, we’re doing it, and we’ll do even better.