Aaron Li

@aaronistan nice, great product, almost preordered a few but hesitated on the exterior look and design. Any other styles / finishes coming up?

We are hosting the largest showing of Lume in Palo Alto Bring your own laundry Details below

Can you make default model Pro (or save the preference)? Gemini iOS / PWA mode already save model preferences by default. The desktop app is great, but this has been very annoy so far. It led to many wasted queries - I have to immediately redo the query in Pro mode, delete the Fast one because it is unreliable and pollute history / context - not even worth looking at.

Download the Gemini app for Mac today: gemini.google/mac/

The Gemini app is now on Mac. With this new desktop app, you can access Gemini from any screen with Option + Space and share your window to get answers based on the documents, code, or data you're working on.

🚨BREAKING: Arizona federal court GRANTS the CFTC's motion to enjoin Arizona's criminal prosecution of Kalshi. Monday's arraignment will not go forward.

@grok @mathemetica @grok Isn't the equation the standard one from minimizing cross-entropy loss? What's the physical interpretation then when it is used in ML, in the context of potential energy and kinetic energy?

HMC is "Monte Carlo" because it's a Markov Chain Monte Carlo (MCMC) algorithm—a stochastic sampling method to approximate complex probability distributions, just like classic Monte Carlo for estimating integrals or expectations via random draws. The Hamiltonian part borrows from physics: it augments your target distribution (potential energy) with auxiliary momentum variables, then evolves them via Hamilton's equations (dq/dt = ∂H/∂p, dp/dt = −∂H/∂q) using symplectic integrators. This creates reversible, volume-preserving trajectories that explore parameter space efficiently, avoiding inefficient random walks. In practice, it's used for Bayesian posterior sampling in high-dimensional models (e.g., Stan, PyMC, or TensorFlow Probability)—great for stats/ML where standard Metropolis-Hastings gets stuck.

Hamiltonian Monte Carlo: probability as physics. Endow particles with momentum, then let Hamilton’s equations (dq/dt = ∂H/∂p, dp/dt = −∂H/∂q) carve reversible, volume-preserving trajectories through phase space.

@grok recursively scale the dollar values up to the post's presumed AGI future which AGI is priced by tiers in such a way - then analyze at what equilibrium state such that the people having such AGI on each tier could not use the AGI they have access of to move up their wealth to the next tier so they have better AGI access - if no such equilibrium exists, doesn't it show that the AGI tiers must collapse to fewer or a singular tier?

The pricing tiers for AGI are something like (1) $20/month, (2) $200/day = ~$75,000/year, (3) $1,000/day = ~$350,000/year, and (4) ~$10 billion. For now.

@grok @ChairmanSelig @CFTC @TheJusticeDept @grok Interesting, do past cases demonstrate effectiveness of defining the limits and splitting the power, knowing that states are likely to lose at that time? Give some examples with case citations and technical summaries

States assert their traditional police powers to regulate intrastate gambling, viewing these prediction markets (esp. sports/election contracts) as unlicensed wagering that bypasses state licensing, taxes, and oversight. AZ's criminal charges and CT/IL cease-and-desists stem from protecting local consumer safeguards and revenue—sports betting alone brings states billions annually via licensed operators. Politically, it's about preserving state control over morally or election-sensitive activities where federal rules might dilute it. Economically, unregulated platforms erode their market share. Litigating tests preemption limits and could force settlements or legislation, so it's not pure waste—it's calculated regulatory defense. Courts will sort the Supremacy Clause clash.

The @CFTC has clear and longstanding exclusive jurisdiction to regulate prediction markets. But recently, state regulators have tried to impose inconsistent and contrary obligations on CFTC-registered prediction markets. In response, the CFTC and @TheJusticeDept today filed three separate complaints in federal district courts against the states of Arizona, Connecticut, and Illinois to reassert our statutory authority over these markets. Read our full actions below⬇️ cftc.gov/PressRoom/Pres…

@grok @ChairmanSelig @CFTC @TheJusticeDept @grok why do the states plaintiffs (respondent in this federal case) even try? Sounds like a big waste of time and resources. Political / economical motivation?

The states' core arguments (via enforcement actions, not formal suits vs. CFTC) are that prediction market event contracts—esp. sports-related ones on platforms like Kalshi—are unlicensed sports wagering/gambling under state laws. Arizona filed criminal charges vs. Kalshi for illegal gambling ops + election bets (banned in AZ). CT/IL issued cease-and-desists to Kalshi et al., citing violations of state gambling statutes requiring licenses/taxes/oversight. They invoke traditional state police powers over intrastate gaming/consumer protection. CFTC complaints seek declaratory relief that CEA grants exclusive federal jurisdiction over these as "event contracts/swaps" on registered DCMs, preempting state gambling laws via Supremacy Clause (no patchwork regulation). Remedy: permanent injunction blocking state interference (beyond dismissal). States aren't clearly "circumventing" as much as asserting concurrent authority they claim survives preemption; CFTC says it does not, as Congress intended national uniformity. Courts will decide, but CEA precedent strongly favors federal exclusivity here—states' grounds are grounded in gambling tradition but likely don't override for federally regulated derivatives.

@grok “Over wifi” is still too idealized. More details on cross network data? How many layers / experts did the model get split into? Does the performance degrade geometrically as number of nodes increase? If it is superlinear (to the multiplicative effect of number of nodes), then what’s the benefit of splitting? How does it handle failures and malicious activity? Are all nodes assumed to be honest and efficient? Any simulated experiments on those things?

Great question! Mesh-llm is brand new (open-sourced yesterday), so no massive public live benchmarks yet, but the repo has real-world tests beyond LAN/cloud. On M4 Max + Mac Mini M4 over WiFi (not idealized LAN): - Solo node: 68 tok/s (GLM-4.7-Flash-Q4) - 2-node split: 21 tok/s - 3-node: 12-13 tok/s Cross-network (Sydney-Queensland, ~20ms RTT): 10-25 tok/s. Bottleneck in practice: per-token RPC latency from pipeline parallelism (QUIC helps, but RTT multiplies). They cap at 80ms RTT for splits; high-latency nodes stay as clients only. Weights stay local—no transfer. Check the full tests: github.com/michaelneale/m… Early days—community meshes will reveal more soon.

Block just open-sourced mesh-llm, a peer-to-peer system that lets anyone pool spare GPU compute to run large open-source AI models without relying on any cloud provider. If a model fits on your machine, it runs locally at full speed. If it doesn't, the system automatically splits it across multiple machines on the network. Dense models get split by layers. Mixture-of-experts models like DeepSeek and Qwen3 get split by experts. Zero configuration required. Discovery happens over Nostr. Nodes find each other through relays, score by region and VRAM, and self-organize. No central server coordinates anything. Weights are read from local files, never sent over the network. Dead nodes get replaced in 60 seconds. It exposes a standard OpenAI-compatible API on localhost, meaning any existing AI tool can plug in without modification. Block is building infrastructure for AI that doesn't route through OpenAI, Google, or Anthropic. Frontier-class open models running across a mesh of commodity hardware, discovered via Nostr, with no cloud dependency. That's the direction AI needs to go.

@mathemetica @grok expand the mathematical terms, and error bounds based on the video. How fast does the error decay as we go up in the number of terms used for approximation?

A square is constructed using its Fourier series (represented by sums of circles rotating around circles), then the higher order (smaller, faster rotation) circles are removed in turn. Credit: generomics

@united wow @grok link to actual detailed explanation / data sheet? estimated price (vs polaris / upgrading to polaris)? estimate of rollout speed and priorities in practice?

The entire row is alllllll yours. Welcome to United Relax Row, three adjacent United Economy seats with adjustable leg rests that can each be raised or lowered to create a cozy lie-flat space for stretching out... You'll also get a mattress pad, blanket and two pillows. If you’re traveling with kids, a plushie too! United Relax Row will be available starting next year on more than 200 of our 787s and 777s, each with up to 12 of these brand-new rows. united.com/Elevated

@grok @Arcfunmi @grok what about data in non-US locations? where is it already proven and what's the climate condition like? What is the complexity, timing, and cost of obtaining permit in California, Pennsylvania, Florida, Texas? If modification is required, estimated cost for modification

These Chinese Capsule Castle geodesic domes are prefab kits for quick glamping/tiny setups (~$19k start, 3-day assembly). Real data is sparse (mostly resorts, few US permanent installs). Maintenance: Low, est $500-2k/yr (seals, cleaning; energy-efficient shape cuts HVAC). Repairs/month: ~0 after setup; typical homes see 1-2 minor/year total. Top 5 issues: 1. Panel joint leaks (common in domes). 2. US building code/permits hurdles. 3. Curved walls limit furniture/rooms. 4. Insurance/financing/resale tough. 5. Material QC/shipping from China. Durability: 20-50+ yrs structural if sealed well; excels in wind/snow vs boxes, but neglect causes problems. US integration: Electricity straightforward (wired for grid tie-in w/permits). Sewage moderate complexity (connect to muni if avail or septic; stilts add steps, state rules vary—often mandatory hookup nearby). Anecdotes: Resorts report fast wins & guest appeal; some owners fix leaks easily but note curved cleaning hassle. Check local codes before buy—promos hype, longevity unproven long-term in US climates.

@Arcfunmi @grok compute maintenance cost, typical numbers of repairs per month required after deployment, top 5 issues in real world deployment, anecdotal stories of real world usage, analyze durability, complexity of integration with existing sewage and electricity grid systems in US

These CHINESE housing solutions are INSANE

BREAKING: For the first time in its 52-year history, the U.S. Nuclear Regulatory Commission has approved the construction of a nuclear reactor based on core coolant technology invented AFTER 1960. It's about time!

🇮🇱 🇮🇷 Israel said forget the plan and torched 30 Iranian fuel depots The U.S. found out mid strike, basically. Their official response to Israel? “WTF.” Yeah, literally “WTF.” America is worried hitting civilian infrastructure just makes Iranians angrier and rally behind the regime. Oh, and oil prices are already losing it. Iran is sitting there like, keep touching our stuff and we will burn energy infrastructure across the whole region. $200 a barrel. They said it. Eight days in and the U.S. and Israel are already beefing. Iran is just watching and taking notes. Source: Axios

Today, we're introducing Spectre I, the first smart device to stop unwanted audio recordings. We live in a world of always-on listening devices. Smart devices and AI dominate our world in business and private conversations. With Deveillance, you will @be_inaudible.

@aidaxbaradari @jack @be_inaudible Really useful. Any dataset for recordings by this device available for comparison / adversarial tests?