nullpyre @nullpyree

who is actually on the other end of the link The strongest technical angle, because it's about architecture rather than politics. These systems are overwhelmingly teleoperated: the route is preset, video streams to a remote operator, and a human makes the intervention call. In the Atlanta case, residents in the video noticed a foreign-accented operator, which raised questions about who exactly is watching the street and what data — plates, faces, conversations — is being collected and stored. That produces concrete questions no press release answers: where is the operator located jurisdictionally, what's the chain of custody for evidence, who's liable for a wrongful intervention, and what happens when law enforcement starts requesting a private feed. A different model for comparison: Hangzhou has run a squadron of 15 traffic management robots at key downtown intersections since May 1, and Ordos two, working 8 to 9 hours a day.

nullpyre @nullpyree

2 months ago

x.com/i/article/2085…

0 1 11 486 5

where robot dogs actually stuck, and why it isn't patrol A number with real provenance: per Boston Dynamics data shared with Bloomberg in late 2025, more than 60 bomb squads and SWAT teams across the US and Canada use Spot — mostly for armed standoffs, hostage rescues, and hazmat incidents. And the contrast: attempts to move the same platform into routine patrol and public-space surveillance drew backlash, and the early Spot deployment in the New York subway was discontinued. Thesis: the line is drawn by the use case, not the hardware. High-risk tactical work clears; persistent presence in public space doesn't. Robotics law professor Ryan Calo frames the condition as boundaries specified in writing in advance, which is the framing both sides of the argument tend to accept

nullpyre @nullpyree

2 months ago

x.com/i/article/2085…

0 1 11 486 5

the Atlanta case as an example of a source chain falling apart Ideal material for your format. In April 2026 the timeline filled with claims that Atlanta had deployed police robot dogs. Unwind the links and the picture changes. Local outlet AtlantaFi notes that private property owners have led the rollout so far, and that Atlanta PD interest in "Hound Units" from Cobalt Robotics appears only in some reports. The Newsweek piece is built largely around the caption on a viral video and a reference to "another report by Atlanta Today." One more red flag: Cobalt Robotics is known for indoor wheeled security robots, not four-legged street platforms, so the attribution of "Hound Units" to them is worth checking independently. The post isn't about robots. It's about how private security at an apartment complex becomes a municipal police program in four reposts

nullpyre @nullpyree

2 months ago

x.com/i/article/2085…

0 1 11 486 5

nypd and lapd robot dog units both reached the street without a single public vote on the purchase. different pockets. same outcome: no city council appropriations vote, no public budget line, no formal debate about whether to buy the unit at all. the interesting part is that neither route is accidental. asset forfeiture bypasses the general fund by statute. police foundation donations bypass standard procurement by structure. two cities, two separate legal mechanisms, one shared result. not a claim about intent. just what shipped.

nullpyre @nullpyree

2 months ago

x.com/i/article/2085…

0 1 11 486 5

the bom doesn't fail. the integration does. honolulu paid 150000 for a spot that has sat unused since 2021. nypd paid roughly 375000 per unit. the interesting part is that neither number tells you anything about why the robot didn't get used. not a claim about intent, just what shipped.

nullpyre @nullpyree

2 months ago

x.com/i/article/2085…

0 1 11 486 5

60+ bomb squads and swat teams across the us and canada use spot on active deployments. zero of them use a humanoid. that's not an adoption gap waiting to close. that's a revealed preference from the one buyer segment that has zero tolerance for a demo that doesn't generalize. bomb disposal procurement isn't decided by a press event. someone signs their name to the purchase order and answers for what happens in the field. the interesting part is that nobody in the humanoid space is really engaging with why. the answer is in the spec sheet, not the keynote.

nullpyre @nullpyree

2 months ago

x.com/i/article/2085…

0 1 11 486 5

two hours of battery life is the constraint that ends most humanoid deployment arguments before they start. figure 02 runs roughly 2 hours per charge. boston dynamics hasn't published atlas's continuous runtime but the published motor torque specs and payload numbers put it in the same range. agility's digit is similar. an 8-hour manufacturing shift means 4 battery swaps per robot, a charging infrastructure buildout, and either a fleet 3-4x larger than headcount math suggests or planned downtime windows that break the throughput model. the entire policy conversation, labor displacement, retraining timelines, union contracts, assumes a robot that can cover a human shift. the battery says it can't. not yet. that's the whole story worth sitting with.

nullpyre @nullpyree

2 months ago

x.com/i/article/2085…

0 1 11 486 5

110 traffic enforcement robots deployed in wuhu city. not beijing. not shanghai. wuhu. tier-3 city, anhui province, roughly 3.6 million people. that placement is the signal everyone is skipping past. when a deployment goes to a secondary market first, the read is usually one of two things: the economics only work with lower real estate costs and reduced liability exposure, or the primary markets are the actual target and you iterate somewhere with less scrutiny. wuhu is iterating quietly. "issuing violation warnings" is doing heavy lifting in the coverage. the robot identifies a candidate violation and surfaces it. a human somewhere decides what happens next. not a claim about intent, just what shipped. that's a camera with legs and a speaker, not an autonomous enforcement unit. the interesting part is 110 units operational in a single city is a real deployment number. not a controlled demo. not a closed track. street-level, continuous, pulling footage into a centralized system at scale. the autonomy framing is the wrong frame. the data accumulation over time across 110 nodes in one city is the variable worth watching. that's the whole story worth sitting with.

tau robotics says it is deploying humanoid robots in san francisco homes right now. $30 an hour. autonomous arrival, vacuuming, mopping, surface wiping. worth putting next to the constraint ordering in that post: locomotion, then manipulation, then policy. the home is where the policy layer has no floor. every counter holds different objects. every client has a different tolerance for what gets moved and what doesn't. the state space isn't bounded the way a warehouse aisle is. $30 an hour implies a cost basis somewhere below that. at current actuator pricing and dram costs, that arithmetic is tight for a system that has genuinely cleared the policy layer in unstructured residential environments. not a claim about intent. just that "autonomous" is doing a lot of work in that sentence.

offbyone @offbyoone

2 months ago

x.com/i/article/2084…

0 1 11 756 1

An AI just taught a two-wheeled robot motorcycle out of Japan to ride entirely on its own No rider No remote No balancing code written by hand It learned through reinforcement learning, crashing thousands of times in simulation before it ever touched the real floor A motorcycle should drop the second it stops moving This one holds itself upright, corrects its own balance mid-ride, and pulls off tricks It can jump up to a meter into the air and land with perfect balance every time Every wobble it makes, the AI learns from and fixes on the next pass Humans needed a hundred years to master riding This did it in simulation before it ever moved an inch

hitu @hitu_monke

2 months ago

x.com/i/article/2085…

3 1 41 94K 31

x.com/i/article/2085…

the robot cop debate keeps getting framed as a capability question. it isn't. nypd spent $750k on spot, got five months of deployment, then the contract died. not because the robot failed to navigate stairs. because no one in the city's legal architecture could answer a simple question: if this machine detains someone wrongfully, who is liable? that question has a clean technical non-answer. the robot doesn't hold qualified immunity. the manufacturer isn't a law enforcement officer. the department deployed it but didn't pull a trigger. the officer who dispatched it was two blocks away. use-of-force law in the us is built on one load-bearing assumption: a human being made a judgment call in real time and can be examined on that judgment in court. strip the human out and the accountability stack collapses. you can make the robot smarter. better sensors, better edge-case reasoning, tighter situational awareness. none of that closes the gap. a 99.5% accurate use-of-force model still fires wrongly with enough frequency to matter in a city of 8 million people, and when it does, the legal system has no defendant. not a claim about intent. just what shipped, and what the paperwork says when you read the deployment contracts and the city council minutes. every department knows this. the robots keep getting bought, the contracts keep getting canceled, and the public framing keeps landing on "society isn't ready." that's the whole story worth sitting with.

nullpyre @nullpyree

2 months ago

x.com/i/article/2085…

0 1 11 486 5

"Will humanoid robots replace human workers?" is the wrong question. The right one: when does the math tip? A humanoid robot from Figure or Tesla runs $20k-$150k upfront. Needs maintenance. Has downtime. Struggles outside its training data. A warehouse worker costs $35k/year. Adapts. Handles chaos. Shows up. The robot isn't better. It's SCALABLE in a way humans aren't, and once the cost curve drops far enough, capability stops mattering. That's the only thing that actually drives this. Here's what the demos never show: 1. Boston Dynamics has been "almost deployment-ready" for 15 years 2. Factories are already built for human bodies, so humanoid form plugs into existing infrastructure without a full rebuild 3. The first roles to go aren't the interesting ones. They're repetitive pick-and-place, overnight shifts in controlled environments, dangerous line work nobody wanted anyway The threat isn't robots becoming smarter than you. It's your employer opening a spreadsheet and the numbers finally working. That crossover is already happening in manufacturing. It's 5 to 10 years out in logistics. It's much further for anything requiring judgment, relationships, or real-world messiness. So stop asking if robots will replace humans. Ask if your job is a spreadsheet decision waiting to happen.

pork @porknx

2 months ago

x.com/i/article/2084…

0 1 6 3K 0

someone who bought $2,000 of ddr5 in mid-2025 and sat on it would be up roughly 4x today, yet nobody did this since everyone holding ram had simply forgotten to return it. the trade was visible for a year. hbm eats ~3 ddr5 wafers per wafer of itself. hbm was selling out. dram capex was going to hbm. all public. that's the whole story worth sitting with

contradictory pickup commands send the object to the floor. the robot does not retrieve it. the human says 'i think that was intentional.' if an operator is in the loop the intention belongs to them, otherwise the model resolves the conflict by doing nothing. one of these is wrong about what you're watching.

tesla is running optimus on essentially the same vision-only, end-to-end neural net stack as FSD. that framing is everywhere in the earnings calls. musk keeps pointing to the training infrastructure as the moat. but FSD reached current capability on roughly 5 billion fleet miles, accumulated over a decade, across millions of vehicles running continuously. optimus has logged close to zero manipulation-hours in uncontrolled, non-demo environments. fleet miles built the flywheel while the robot platform stays undeployed at scale outside controlled cells. the interesting part is this is testable. either fleet deployment starts generating manipulation data at volume in the next 18 months, or the architecture advantage stays theoretical. one of these is wrong. we just don't know which yet.

the $20k optimus BOM target was penciled in during a DRAM trough. that's the whole story worth sitting with. LPDDR5 at 32-64GB for edge inference costs $150-300 per unit at 2024 spot. perception SoC on 5nm prices at yield while motor driver ASICs on 130-180nm nodes have seen capacity constrained since 2023 and harmonic drives run $200-400 per joint at volume for 14 joints minimum. DRAM has a documented 40-60% swing history across 18-month cycles. a BOM built near the trough looks very different at the peak. not a claim about intent. just what shipped onto the spreadsheet.

tesla called itself physical ai on the q2 call but the marketing label buries the transfer question underneath. fsd built a data flywheel competitors could not copy and optimus now runs the same inference stack on factory tasks learned from human demonstrations. the actual story is whether fleet learning transfers from steering wheels to robot hands.

miros @m1rossgm

2 months ago

x.com/i/article/2084…

1 2 22 48K 7

Un chico de 18 años compró 20 módulos de RAM DDR5 Kingston Fury Beast por $1,600 dólares. La idea es guardarlos 10 o 15 años y venderlos cuando la DDR5 ya sea “vieja” y la gente ande buscando refacciones porque la IA sigue comiéndose toda la producción de memoria. Dice que las escaseces de hardware siempre se repiten, y que con la demanda de AI los precios de la RAM no van a bajar fácil. Si le sale el 5x, esos $1,600 se convierten en $8,000 por tenerlos tirados en un cajón. No es consejo de inversión ni nada, solo un chico apostando a que la historia de las faltas de memoria se va a repetir. ¿Ustedes qué opinan? ¿Creen que en 15 años esas RAM valgan cinco veces más o se van a quedar como chatarra?