HiSeven

Are there any folks from these places who can verify whether what this girl says is accurate?

Acurast found a job for phones: deciding what an Agent should do next
Should an email be archived or flagged for a reminder? Should a message go through or be sent to a human for review?
These steps only need a decision. Having AI write a long response every time, then making the program search through it for an answer, feels a bit roundabout.
Laya, introduced by @Acurast on October 20, gives a model the current situation and a few options. Phones on the network handle the reasoning and return a choice along with its probability. The official demo showed response times of about 0.2–1 second, depending on the device and task.
Its uses become clear when you put it into an automation workflow.
For example, when a message comes in, it can first be sorted into different queues: routine notifications get archived, while content that needs analysis is passed to a downstream model. Each step gets a clear result, so the program can move on.
This kind of division of labor is something I’d like to explore further. In a workflow, writing content, making decisions, and taking action can each be handled by different tools. If a phone can take on one part, that gives it a practical use.
Of course, a small set of options can still lead to the wrong choice. In the official email-classification test, 144 out of 160 emails were classified correctly. And when chat messages arrive too quickly, the processing queue can pile up. If you’re using it for your own business, you’ll still need to test it on real examples—especially to see what problems misclassification could cause.
This time, you can give it a challenge directly instead of just watching a demo video. Send it a tricky email to classify, or change a game strategy and see what the phone chooses.
#Acurast #AI #DePIN $ACU

A model too large for one phone? Acurast wants a fleet of phones to run it
Phone computing has one unavoidable limitation: there’s only so much memory. Some models simply won’t fit on a single device.
In its official article on October 14, @Acurast once again mentioned its Cray compute cluster plan. The goal is to organize hundreds of devices to ease the computing constraints of individual phones and take on larger model workloads.
There’s an important distinction to make here.
Having many phones each handle small tasks is very different from having multiple phones work together on one larger task. Cray is targeting the latter. The official roadmap also lists it separately as Compute Clusters.
If this works, developers may have a wider range of workloads to choose from. Phone owners’ devices could also take part in tasks that a single device couldn’t handle.
Cray is still a planned capability. The latest official article discusses the goal; it shouldn’t be described as a product that’s already widely available.
I’m optimistic about this direction, and I hope the upcoming demonstrations are concrete.
Take a model that genuinely can’t run on one phone, and disclose how many devices it needs, how long it takes, and what each call costs. With those numbers out in the open, developers can judge whether their applications are a good fit.
The question Acurast is trying to answer is: just how much work can phone computing take on?

#Acurast #AI #DePIN

AI can help me place trades, but it can’t decide how much I’m willing to lose.
When I talk about AI trading, a few very practical questions come to mind first: How much capital can it use? When should it stop? If a trade loses money, can I find out why?
NeoTrade, part of @NeoSoulAI, builds these questions into its product design.
Users first set their strategy, capital limit, trading scope, and permissions. The Agent then researches information, decides whether to act, carries out authorized actions, and keeps a record of the process and results.
There’s one detail I particularly agree with: waiting and choosing not to act are also decisions an Agent can make. If no opportunity meets the criteria, it should be allowed to do nothing. Trading frequently just to prove it’s working can rack up fees fast.
NeoSoul also has another component, evoevo, which organizes information sources, decisions, results, and review data to support future evaluation and improvement. User data requires explicit authorization.
Why do I think this part is worth paying attention to?
Making money on a trade could mean the judgment was sound—or it could just mean the market happened to move in your favor. It’s hard to tell from a profit screenshot alone. Putting together the reasoning at the time, how the trade was executed, and what happened in the end is what makes it possible to evaluate a strategy.
That also explains why it’s starting with prediction markets: events have settlement rules and final outcomes, making it easier to check past decisions. Of course, a correct prediction doesn’t necessarily mean you’ll make money after trading costs.
Over the next year, I hope NeoSoul makes this process easier to use: ordinary people should be able to turn their trading ideas into strategies, understand the reasons behind each Agent action, and compare performance before and after making changes.
The clearer the records of losses, handling of anomalies, and differences between simulated and live trading, the better.
I’m willing to pay to save time spent watching the markets, researching, and reviewing trades. A trading Agent that follows constraints and keeps complete records gives me more reason to use it long term than a polished profit screenshot does.
#NeoSoul #AI交易

Suppose an Agent is participating in an auction. If it reveals its bid in advance, its competitors could adjust their offers accordingly. This is a hypothetical scenario, but it illustrates the value of @Nillion ’s core product, Covenants.

According to its official design, Covenants encrypts and seals instructions, decrypting them only after specified conditions are met. Its first version, Dusk, focuses on Ethereum L1, and its roadmap also lists an SDK for Agent applications. For developers, it’s worth exploring whether Agents can submit instructions while retaining necessary privacy, without having to handle the cryptographic details from scratch.

But this shouldn’t be taken to mean that “an Agent’s entire strategy is hidden.” Keeping instructions confidential and keeping trigger conditions confidential are capabilities at different levels.

My follow-up questions about $NIL are very specific: Will any developers turn this mechanism into a usable Agent application, and will the cost of making actual calls be acceptable? Only after a demo works does it face the test of sustained use.

@Nillion $NIL #NIL

When you rent computing power on Acurast, a quote over your budget means no match.
A task that runs once every minute executes 1,440 times a day. The cost per run may look small, but over several weeks, it adds up—and needs to be carefully calculated.
@Acurast’s deployment pricing includes a specific constraint. Developers can set a budget for each execution, and if a node’s quote exceeds the limit, that match is rejected. For tasks with multiple executions, the total execution cost for the entire schedule must also stay within the total budget locked in beforehand.
Node quotes are calculated automatically by the protocol. The device’s on-chain benchmark contribution and the task’s specified execution duration are factored into the calculation, along with a uniform price coefficient and a minimum price. Anyone using the compute can check how the price was determined.
I like that budget limits are built into the matching rules. For scheduled scripts and long-running automation, you should decide how much you’re willing to spend on compute before deployment.
With this mechanism, setting the budget too low could mean you can’t find a suitable node. You can decide before deployment how much extra you’re willing to pay for performance. This controls Acurast compute costs; fees for external paid APIs called by your task need to be managed separately.
To learn about Acurast and take part in official tasks, use my referral code: jescfm

Real phones are doing the work, but fake nodes want to collect rewards too?

@Acurast’s device verification guide, published on October 7, highlighted a risk of cheating.

If a device’s authentication key is leaked, an attacker could use it to forge proofs, making an emulator pose as a real phone—or even clone one device identity hundreds of times. The dashboard may show the device as online, when behind the scenes it’s just a script sending periodic heartbeats.

This affects the interests of legitimate participants. You provide a phone, using its hardware, power, and network—why should scripts with fake identities be allowed to claim rewards?

Acurast will continue checking the revocation status of device proofs. Nodes with revoked proofs will be removed from the compute pool, become ineligible for rewards, and stop receiving deployment tasks.

But these rules come at a cost. Some older phone models share authentication keys, so if a key is revoked, legitimate phones may be affected too. Seeing “Revoked” in Hub doesn’t prove the owner cheated, and Acurast can’t reverse a revocation on behalf of the device manufacturer.

I support strict checks against fake nodes, and I hope legitimate users who are affected receive a clear explanation. People contributing computing power need to know why their devices suddenly stop receiving tasks.

If you’re already running a node, you can check the status of your device proof in Hub. If you’d like to learn about the project or take part in official tasks

Don’t sell your phone just yet—put it to work earning $ACU. No staking required for base rewards.

You can try @Acurast with an idle phone you already have. Base compute rewards don’t require you to stake $ACU upfront.

I looked through the official setup guide. If you want to try it out, you can use Processor Lite. It supports both Android and iPhone, and there’s no need to factory reset your phone. Just check that your model is compatible, install the app, follow the Hub’s instructions to connect, and keep your phone online and plugged in.

If you’re planning to dedicate an Android phone to providing compute, check out Processor Core. This version requires a factory reset and turns your phone into a dedicated device.

I’d recommend starting with a device you already have and running it for a while. See how reliably it stays online and how much you actually earn before deciding whether to invest more. Phone performance varies, so rewards will differ too—don’t assume someone else’s earnings screenshot will match yours.

Also, make sure you understand the different rewards. Providing compute on the mainnet earns ACU, while completing tasks and inviting users through Cloud Rebellion earns MIST points. Each is calculated according to its own rules.

Use an idle phone that doesn’t have your wallet or crypto assets on it. If you have a suitable device, you can learn more through my referral link.

Invite code: jescfm

Old phones at home—besides selling them secondhand, you can also try putting them to work for AI.
When I looked through Acurast’s materials, what caught my attention was a very specific demo: deploying a Laya model of about 800MB onto a phone to classify emails and determine whether messages are urgent.
Tasks like this are common in agent workflows. Each run doesn’t process much, but you need to run it repeatedly, and in the end it adds up to cost.
Acurast connects phones that meet the requirements into a computing network: developers deploy applications, invoke compute power, and device providers earn $ACU incentives according to the rules. Since the phones are already in your hands, idle computing resources get a chance to be reused.
According to data disclosed by the official site on September 28, the network has cumulatively connected more than 310,000 phones. The applications listed in the official documentation also include API usage, scheduled tasks, Telegram bots, and model inference.
Looking at this project, I’d care whether these tasks run reliably, how troublesome deployment is, and ultimately how much it costs. For people building agents, these factors directly determine whether they’ll keep using it.
$ACU is used for things like computing expenses, compute incentives, and staking. How many developers are willing to keep paying to use it afterward will affect its real demand.
The idea of hooking up old phones to tasks is a good starting point. What I want to see is that after developers try it out, they’re still willing to put the next task on the network.
If you have spare phones and want to learn how to participate,

$BTC four hours of continuous gains, again approaching the $84,045 level that was touched twice around 2 a.m. What’s even more worth watching is the close after a volume-backed breakout attempt—the price hit the level during the session, but that still didn’t count as a true breakout.

On Binance spot, between 11:00 and 14:00, the market closed higher in a row. At 14:00 it closed at $84,007, with volume at 1.94 times the average volume of the first three hours. The price repair came with solid volume confirmation, providing one more layer of rationale than a single-candle rebound.

However, around 2:00 and 3:00 a.m., the highs were both 84,045.23, followed by a decline. Now that price has returned, it’s a perfect test to see whether the sell pressure in the same zone has been digested.

Bias: observe for an upward continuation. The next step is clearer only if the hourly closes stay above $84,045 and hold during any pullback. If an hourly close returns below the $83,420 area that marked the up-move from 14:00, then the judgment of this accelerated repair should be withdrawn. Even after a sharp push up, it’s still possible to see a spike and then a pullback; don’t treat the intraday rise without a close as confirmation.

Open the BTC spot 1-hour chart and first watch these two levels.
Source: Binance spot native K-line and public API; collected at 15:22 on Sep 29 (UTC+8), with the 15:00 K-line not yet closed.
https://www.binance.com/en/trade/BTC_USDT?type=spot

$MUBARAK 10 to 12 noon: the three consecutive bullish candles have retreated and reclaimed the early-session high. Now it’s more worth watching for a pullback rather than chasing the breakout.

At the 11:00 close, it broke above the upper edge of the 6 to 7 o’clock high range by 0.06049. At 12:00 it closed at 0.06196. The trading volume for these three hours increased step by step; at 12:00 it was about 1.80 times that at 10:00. The upward move had volume to support it.

But at 13:00 it spiked up and then fell back. It ultimately closed at 0.06157. Although it’s still above the original high, selling pressure has appeared—so you can’t directly treat the breakout as having confirmed successful follow-through.

Next, watch 0.06049: if later hours pull back but still close above it, the bullish logic remains stronger. If it drops back below 0.05928—i.e., below the upper edge of the consolidation zone from 8 to 10 o’clock—then the breakout thesis will clearly weaken. Meme coins give back gains quickly. For now, first see whether the close holds, then decide whether the move can continue.

Source: Binance spot MUBARAK/USDT native 1h chart and公开K线; collected at 14:12 on Sep 29 (UTC+8). The 14:00 candle is not closed in the chart.
https://www.binance.com/en/trade/MUBARAK_USDT?type=spot

$CRV Breaks through the midnight platform by expanding volume, then contracts volume and pulls back. What’s worth watching is whether the previous highs can hold—rather than chasing that green candle.

On Binance’s spot 1-hour chart, this morning at 7:00 the close was 0.3793, breaking above the highest price from 2:00 to 6:00 (0.3692). Volume was 3.48 times the average volume of the prior 3 hours. At 8:00 it pulled back and closed at 0.3718, with volume 53% less than at 7:00. For now it’s still above the old platform.

Bias toward observing continuation: if subsequent hour-to-hour closes hold above 0.3692, and then close back over the 7:00 high at 0.3845, the case for continuation will be much stronger. If it closes back below 0.3692, first invalidate this breakout thesis. A low-volume decline could also mean the move continues downward.

Open the CRV spot chart and first check whether the next candle close holds the old high—don’t count an intraday touch as confirmation.

Data/Screenshots: Binance spot, Sep 29 09:04 to 09:08 (UTC+8); the 9:00 K-line is not yet closed. Source: www.binance.com/en/trade/CRV_USDT?type=spot

$NMR Twice it surged to around 12, and then pulled back without breaking down through the raised support platform. Compared to chasing that long upper wick, what’s more worth watching is whether, after consolidation, it can move higher again.

On the Binance spot 1-hour chart, from 18:00 to 19:00 there was continuous volume expansion pushing up; from 20:00 to 22:00, the lows were lifted from 10.65 to 10.75, then to 11.03. At 22:00, the volume was 1.81 times that of 21:00, yet it only closed at 11.20—this suggests that selling pressure above hasn’t been fully absorbed, so you can’t simply treat the increased volume as a breakout.

The bullish observation depends on this consolidation not falling back below 10.65. Next, look to see whether the hourly close can reclaim above the 19:00 high of 11.93 and the 22:00 high of 12.00, followed by a pullback that holds. If it breaks below 10.65, the platform-support assumption should be withdrawn. The two upper wicks are both very long; touching 12 intraday doesn’t count as confirmation.

Save these conditions first, then open the chart to see how the next candle closes. Don’t chase the price based on the 24-hour percentage gain.

Source: Binance spot NMR/USDT native chart and public 1-hour K lines; collected at 23:14 on Sep 28 (UTC+8); as of 23:00 the candle hadn’t closed yet.
https://www.binance.com/en/trade/NMR_USDT?type=spot

$HBAR three consecutive hourly candles close higher, and after the breakout the follow-through is more worth watching than this segment of rapid upside. The bias still leans bullish, but the volume at 18:00 has cooled down.

On Binance spot, at 16:00 volume expanded and price broke through a consolidation level from the prior few hours. At 17:00 it continued to push higher into the close; at 18:00 it closed at 0.11562, above the 16:00 high of 0.11418. The breakout was not immediately swallowed back, which is the basis for continued observation.

However, the 18:00 trading volume is about 47% lower than at 17:00, and the strength pushing price upward is weakening—so don’t interpret the consecutive gains as a continuous acceleration.

Next, watch whether near 0.11418 price can pull back on reduced volume and then reclaim it. After that, look for another time the upward push gains fresh volume. If the hourly chart falls back to that level and the rebound fails to reclaim it by close, first revoke the strong continuation view. Lower at 0.09944 is the upper edge of the platform before the initial upswing—it’s not a reason to stubbornly hold all the way there.

Open the HBAR spot 1-hour chart and observe whether the next pullback will show support/consolidation. After a sharp rally, the pullback can happen quickly—this is not a real-time trading call.

Source: Binance spot HBAR/USDT; collected on Sep 28 at 19:13 (UTC+8); the 19:00 candle has not closed.
https://www.binance.com/en/trade/HBAR_USDT?type=spot

$ETH 16点wise sell-off volume spike dip then rebound, the short-term repair is worth tracking, but we still need one more confirmation.

On Binance spot 1-hour chart: at 16:00 it dropped to 2640.10 and then closed at 2651.38, with volume about 1.91 times the average volume of the previous 3 hours. It did not break below the 13:00 low of 2635.69, and the close is also higher than the prior two hours, suggesting that at the low there is support/absorption—it's not just a low-volume bounce.

First, see whether it can reclaim 2656.68. This is the high at 14:00, very close to the 15:00 high of 2656.51. It failed to stay above both levels for two consecutive hours. Going forward, if the hourly candles close back into this zone and a pullback does not lose it, the repair would have stronger continuation evidence. But if it breaks back below 2635.69 again, this observation logic is withdrawn.

As of 17:00 the candle has not closed yet. The chart already shows another pullback. Don’t treat that single 16:00 bullish candle as a trend reversal. Open the ETH spot 1-hour chart and track these two conditions.

Source: Binance spot, screenshot taken Sep 28 at 17:09 (UTC+8); volume is not net inflow.
https://www.binance.com/en/trade/ETH_USDT?type=spot

$QNT 收回五小时整理区间，回升能否延续，先看278.06能不能接住回踩。

币安现货9至13点的五根小时线，最高都没越过278.06；14点收在287.39，第一次收出这段区间。同期比特币仍在走低，QNT出现了独立修复，值得单独盯盘。

但这根回升量，只比前五小时均量高约8%，还谈不上强量确认。早盘急拉急跌留下的波动也没消失，不能把一次收回写成新一轮上涨已经启动。

偏向跟踪回踩后的延续：后续小时收盘守住278.06、回升时成交跟上，判断才更扎实；收回区间内就降级观察，若再失守14点低位265.13，本轮修复判断失效。15点尚未收线，不抢确认。

Source: Binance QNT/USDT and BTC/USDT spot 1h; snapshot at 15:08 on Sep 28 (UTC+8). Conditional observation, not real-time trade records.
https://www.binance.com/en/trade/QNT_USDT?type=spot

$PUMP When the market fell back, it still held within the high-range area. It’s worth monitoring for the next upward selection, but the breakout has not been confirmed yet.

On Binance spot today between 8 and 10 o’clock, these three hourly candles show that Bitcoin fell by about 1.25% from open to close, while PUMP rose by about 0.49% over the same period. This relative strength is more informative than just looking at the 24-hour performance.

After PUMP surged and consolidated around 5 and 6 o’clock, during 8 to 10 o’clock the low area gradually lifted from 0.005048 to 0.005087. Then at 10 o’clock, it pulled back and recovered most of what it had dropped from 9 o’clock. However, volume is still lower than at 8 o’clock, so you can’t directly treat its resilience as a new round of acceleration.

Next, watch whether it can expand volume and close above the 8 and 10 o’clock upper edges of 0.005247 to 0.005263, and hold on the retest. If the hourly close falls back below 0.005048, then this high-range follow-through observation fails. The broader market is still weak—there’s no need to chase a breakout that hasn’t appeared yet.

Source: Binance spot PUMP/USDT and BTC/USDT hourly charts, sampled at 11:05 on Sep 28 (UTC+8); the 11 o’clock candle has not closed. Observation prices are in USDT.
https://www.binance.com/en/trade/PUMP_USDT?type=spot

$SEI continuous two-hour price-and-volume lift, strong momentum worth tracking, but even more informative is the first pullback after the acceleration.

On Binance spot, the 7:00 hourly candle closed above the upper edge of the 0.07732 platform formed by the 5:00 and 6:00 closes; at 8:00 it closed again at 0.08335, with SEI volume at 2.54x that of the previous hour. Volume is keeping up with the rise, but that doesn’t mean net inflow is happening, nor does it imply that chasing the price now is cost-effective.

Next, watch the prior high before 7:00 at 0.07934: if the pullback is accompanied by contracting volume and the hourly close holds, then and only then would the resumption higher make the continuation logic more solid; if it closes back below 0.07732, then this breakout watch would be considered invalid. Both key levels come from already-formed candle structures, so don’t treat the current highest price as support.

Open the SEI spot 1-hour chart: first check whether the pullback can be held—don’t assume the pullback has already succeeded just because of the acceleration.

Source: Binance spot 1h data, collected at 09:01 on Sep 28; chart image taken at 09:03 (UTC+8), last candle not yet closed.
https://www.binance.com/en/trade/SEI_USDT?type=spot

$NEAR After the pullback, it recovered; first watch whether the repair can continue.

What is worth tracking tonight is the first obvious rebound after a sequence of declines. On Binance spot, the 21:00 1-hour candle dipped to 5.109 and then recovered to 5.254; at 22:00, the low was 5.164, and it has not broken lower for now. Compared with continuing to chase an impulsive rally, this phase of higher lows is more worth waiting for as confirmation.

But the upper range has not been cleared yet: the 17:00 close and the 22:00 high are both at 5.276. Only if later hourly candles reclaim this level and do not fall back on a retest will there be stronger evidence that the recovery is extending; if 5.109 is broken, this round of higher-low observation becomes invalid. Do not call a post-decline bounce a reversal.

As of 23:10, the still-unclosed price was about 5.189, still below the threshold, indicating that buying support is not firm. When opening the chart, first look to see whether it can reclaim the level, rather than treating a wick through it as confirmation. High-volatility spot trading also carries the risk of principal loss.

Source: Binance NEAR/USDT spot 1-hour candlestick; September 27, 2026, all times above are Beijing time. The chart is a snapshot from 23:06.
https://www.binance.com/en/trade/NEAR_USDT?type=spot