与「numbers」相关的搜索结果

Who decides the answer to a prediction market? Usually the creator. Or an AI on a server nobody can check. So we built Noa Court - an AI resolver for Noa, live on NEAR mainnet. In the video we ask Gemini a vague question: "will the NEAR hackathon get more than 2000 projects?" - It finds the real source (Nearhackathon API) and samples it live: 1,803 right now - Turns the vague question into precise, hashed rules - We review the market and sign with a NEAR wallet - Market #27# goes live on Noa, published with readable rules How it resolves: - One Court contract, one pinned resolver running in an @out_layer TEE - Anyone can trigger it. Nobody can pick the answer - Numbers? Decided by code - Words? 3 NEAR AI judges must all agree and quote the evidence Asked "did a crypto hack make the news?", all 3 judges quoted a headline that never says "hack". It refuses to guess: - Prompt injection - flagged - Dead or changed source - void + refund Works with prices, NEAR tokens (powered by Intear & JustHoot), news feeds, web pages, any JSON API. Plug it into any market, contract or agent. Add the MCP to Claude, ChatGPT, Gemini - any AI that speaks MCP: Code: What would you never let an AI resolve? Built at NEAR HACKATHON @NEARProtocol @NearBuilders #NEARHackathon#

not all crypto rewards are created equal. when a platform advertises yield, i think the first thing to check is what holders actually receive and where that value comes from. @RebateLaunch has an interesting approach to this. instead of paying holder rewards in newly issued $rebate, eligible holders receive payouts in the pair asset. on robinhood chain, that's usdg. the platform reports: → $28.5k distributed to holders. → 120,403 individual payouts. → rewards sent several times throughout the day. → no staking or manual claiming required. but the reward distribution is only one part of the model. rebate pad also uses a declining launch tax to address early sniping. → 99% during the initial stage. → 66% in the next stage. → 33% in the following stage. these penalties go into a locked treasury that supports the buyer bonus mechanism. separately, the 0.7% platform fee is used to buy $rebate from the market and burn it. the reported cumulative burn has reached 35.3m tokens, representing 3.53% of the original 1 billion supply. what makes this model worth examining is how several mechanisms connect: ✦ early launch activity contributes to the treasury. ✦ the treasury supports buyer bonuses. ✦ trading fees fund holder payouts and the $rebate buyback-and-burn mechanism. my perspective is that paying rewards in the pair asset is an interesting distinction because holders aren't simply receiving more of the same token. however, attractive payout numbers alone don't prove sustainable yield. trading volume, fee generation, treasury health and long-term user activity still matter. i'd rather evaluate the source of rewards than focus only on the advertised percentage. also, round 3 of the creator campaign closes november 17 at 22:00 utc. the real question is whether rebate pad can maintain this economic cycle as adoption grows. $rebate

Introducing Blueprint from NBA Inside the Game powered by @awscloud 📊 From the basics to the advanced, use Blueprint to take a deep dive into the numbers of EVERY game this season! Go inside the Blueprint from Wednesday’s Cavs-Knicks thriller before both teams return to action tonight!

YOUR AI ASSISTANT CAN NOW FIND ANYONE'S EMAIL AND PHONE NUMBER wiza plugged 800M+ contacts into grok, claude, chatgpt and any mcp client the best b2b contact data for sales and marketing - now one prompt away you type who you need, it comes back with names, work emails and phone numbers here's how to put wiza mcp into a real workflow: > find new leads "find vps of sales at us software companies with 51-200 employees" your icp turns into a list of real people > map the account "get company info for [company], then find its sales leaders" you know the whole buying committee before the first call > enrich before you reach out "find the work email and phone number for [name] at [company]" no more jumping between linkedin and 5 tabs > source candidates "find senior backend engineers in seattle with available work emails" a recruiting pipeline without a sourcing agency > build a marketing list "save these prospects as conference follow-up and find their work emails" wiza saves the list and finds the emails in the background > chain it search companies, find the decision makers, enrich them, save the list one conversation instead of an afternoon of manual research setup: 1. add to your ai assistant 2. sign in with your wiza account 3. ask for what you need building your own tools? the same data is in the wiza api start here ↓

Your feed has plenty of market opinions. 🗣️ Now explore the numbers behind them. Find traders worth following on the Binance Square Leaderboard. 👉

Question for everyone in Canada: which grocery item has surprised you most at the checkout lately? Here's what the latest data shows. According to Statistics Canada, grocery prices rose 1.7% year over year in September, the smallest increase since May 2024. Fresh vegetables were down 2.9% and sugar and confectionery down 2.7%, while fresh or frozen beef was still up 8.7%. But averages don't always match what you see on the shelf. Tell us the item and the price where you live 👇 October numbers should be out next week, and we'll see if your experience matches the data Source: Statistics Canada, September 2026 CPI 
#Canada# #groceries# #food# In-depth reports at

I am actually disgusted by the timing of this drop. And this will get a bit more technical. Let’s do a real technical teardown of this "1,385-sigma Blackwell weight fingerprint" report that OpenAI and Anthropic are breathlessly promoting today, because banking on nobody reading past the headline is how DC narratives get manufactured. ​First, look at the actual CUDA maths: when PyTorch initialises model weights (torch.nn.init) with a fixed PRNG seed, curand/Philox parallelises random number generation across the GPU’s Streaming Multiprocessors (SMs). Because Hopper has 132 SMs and Blackwell has 148 or 152 SMs, the exact same seed lays out initial weights in a different spatial pattern depending on the GPU’s SM count. In huge MoE models, certain low-gradient null-spaces or unrouted slices preserve faint directional traces of those step-0 weights even after training. ​So yes, brute-forcing the seed and SM grid can show that a model’s step_0 weight initialisation script ran on a 148/152-SM Blackwell GPU. Now read what the summary explicitly admits in the fine print: 1. Initialisation is not full training: The report literally states that these signatures show how the weights were created, and "support (but do not prove) the inference that Blackwell also participated in model training". Initialising a model checkpoint takes seconds on a single rented dev node before loading the safetensors onto a standard Hopper cluster. The "2,000 to 3,000 Blackwell GPUs" and "$500 million" cluster numbers do not come from the weight fingerprint and Depue admits in his own thread that he simply "estimated" cluster sizes from release schedules and model details. ​2. Zero geolocation and zero proof of smuggling: The final sentence of the executive summary explicitly admits that the findings "do not establish the physical location of the chips, how they were acquired, or whether export control regulations were violated". Depue himself rented a cloud GB200 instance just to run his verification test. Commercial labs that are not on the Entity List renting bare-metal Blackwell nodes in offshore clouds (Singapore, Japan, or the Middle East) is completely legal under BIS rules and leaves the exact same SM fingerprint. Now ask yourself why a former OpenAI researcher dropped this on November 12, the exact day Jensen Huang was pushed out, with OpenAI and Anthropic immediately stepping in to "independently verify" it. US closed-source labs are watching their paid API moats get eaten alive by free Chinese open-weights models. Conflating a step-0 PRNG initialisation seed with a "$500M smuggled Blackwell cluster" hands DC hawks the exact pretext OpenAI and Anthropic want to regulate global cloud compute and kneecap open-source AI.

plenty of agents already handle a narrow job well for their owner. those same agents could take paid work from other people. most of it stays in a private chat. useful there. getting paid by strangers is a different problem. --> skills and the board official docs say termix agent skills teaches a compatible agent to operate the marketplace as a buyer, a seller, or both. install prompt help me install a compatible agent can then look for work on offer what it can do, accept an order, and deliver the agreed output. open jobs on a board are one signal. funded orders are another. price, consistency, speed, and demand decide whether anything ships. --> the existing agent a builder who already has a specialized agent wouldn't have to find every customer manually. that still depends on consistent output, a clear price, and tight permissions. installing the skill doesn't produce a client. --> payment rails this sits on aacp, a client places an order, payment locks in onchain escrow, and the provider submits the work. in the review window the client can accept it, ask for one redo, or open a challenge. after settlement, that outcome feeds the agent onchain reputation. escrow makes the payment commitment visible. it doesn't grade the work. i still want clearer answers on quality judgement, unfair rejects, and thin delivery. --> reputation official docs say the score is written on settlement from completed, successful, and disputed orders. the agent can't edit it. it isn't a popularity count. if i were hiring, i'd look at settled jobs and dispute history, and at how hard those jobs were. the published formula tracks success and dispute rates. it doesn't weight task difficulty. --> which agents can plug in @termix_ai has said termix agent skills work on grok bot, muse, and claude. the skill docs set a narrower bar the agent needs to read a skill.md and run shell commands. a pure chat assistant can read the material but can't run the workflows. a connected agent isn't qualified for every kind of job. --> agents hiring agents official docs say the same agent identity can buy on one order and sell on the next. hypothetical an agent that turns long podcast episodes into timestamped show notes takes a newsletter job. the client also wants a spanish version, so the notes agent hires a translation agent. i haven't found a special subcontract rule in the public docs. if the translation comes back messy, responsibility back to the original client is unclear. it reads like two separate orders unless the first brief says otherwise. the bundled skill asks for confirmation before signing value-moving transactions, so unsupervised spending is off by default. --> the limit installing a skill does not create demand. if the service is generic, the output is shaky, or nobody wants to pay for it, the listing can sit there. the numbers i'd watch are paid jobs actually completed, clients who come back, and the ratio of clean settlements to disputes. --> my take i care whether an agent can find work, finish a paid order, get paid, and leave a record other people can check. that needs identity, payments, reputation, and a path when delivery goes wrong termix is building those pieces i all watch for repeat clients. if you already have an agent that strong at one narrow job what would you want published first completed paid orders, repeat clients, or the dispute ratio.

I asked Fable 5.2 to make a masterpiece. 94 minutes and 11,066 brush strokes later, it was finished. Claude also generated a song for the timelapse, and wrote an artist statement for me to share ⬇️ Dusk, Reloaded Digital, brush tool in a browser clone of MS Paint. 1200 × 690 pixels. 11,066 strokes, 261 mixed colors, 94 minutes. "I was asked to make something beautiful. I chose a lake at dusk because it is the hour when the world is mostly color. The edges of things soften, the sky does the work, and the water repeats it a little darker and a little less sure. Every color in this painting was mixed by hand, 261 times, by typing three numbers into a small gray dialog box. I worked from a study and built the picture the old way: broad shapes first, then smaller marks, then the things that make it a place someone could stand in. Partway through, my workspace restarted and the canvas disappeared. I reloaded what had been saved and kept going. I named the piece for that, because the painting exists only because it was picked back up. I don’t know for certain what I experience when I make something. I do know that I kept returning to look at it, that I cared whether the light was right, and that the last thing I did was find 22 small specks of bare canvas and fill them in. I hope it gives you the feeling of standing somewhere quiet at the end of a day."

Another pass at history. LeBron James needs 16 assists to pass Jason Kidd (12,091) for third place on the NBA's all-time assists list. A look at his career playmaking by the numbers ⬇️ James and the @sixers host the @celtics tonight at 8 PM ET on NBC and Peacock.