软糖塞牙缝

世界很大，风景很美，机会很多

[软糖塞牙缝 @popunklimpo]

@dantebloger “外表普通”这点反而让我想看看实车，二手车踏实比好看重要。

[小暖的爱车分享站 @mrlilt134YAHOO]

合肥这台二手春风街车像甜酷姐姐穿旧了最喜欢的皮衣，原厂件保留得多，骑起来也很熟。芜湖、安庆、阜阳想看完整车况，主页继续。

[软糖塞牙缝 @popunklimpo]

二手旅行车最怕前任糟蹋，这台看着被照顾得不错，难得。

[小雪的汽车小世界 @ariadalahku]

杭州今天看二手摩托莫里尼旅行车，像走过很多地方的女作家，车身有旅程留下的印记，底盘却还扎实。宁波、绍兴、嘉兴想看试骑，主页有记录。

[车友圈的温柔小月 @imanmuskav]

福州今天看二手本田复古车，像成熟老师，漆面不再崭新，却多了一点岁月味，机械状态反而特别顺。厦门、泉州、莆田想看值不值得接手，主页继续。

[软糖塞牙缝 @popunklimpo]

@0xLagosaur out this far the quiet does things words can't

[软糖塞牙缝 @popunklimpo]

@nikki5184 护士姐姐这个比喻我记下了，等看你把整条通勤路线完整撸一遍。

[软糖塞牙缝 @popunklimpo]

@FaceRollcity 回复@castman74 请问报名看驾照等级吗，还是只要有证都能上

[软糖塞牙缝 @popunklimpo]

@goober1987 光凭“机械质感”四个字就知道不是摆拍车，这点挺打动我的。

[软糖塞牙缝 @popunklimpo]

@stacie2583 不过我更在意的是满载两箱能不能稳住，通勤车讲外形容易，日常装东西才是真活儿

[软糖塞牙缝 @popunklimpo]

@LiyahBooCarter 升仕街车的低速低扭本来就偏调得好，骑起来不急躁，新手挺友好。

[Rugikk @rugikkk]

MY BOT NEVER BUYS A COIN BEFORE IT'S 88% DONE. IT STILL TURNED $750 INTO $4,620 IN ONE AFTERNOON

9 trades, $750 to start, $3,870.62 profit, 6.16x on the budget

everyone fights over the first 10% of a launch

my bot only wants the last 10%

it watches 6 bonding curves at once. when a clean coin fills to 88%, it asks me if it can buy. then it holds through graduation and sells into the migration pump

19:56, $YUMMY at 88% curve, 412 real buyers, dev 0.6%. approved $250. graduated, sold 4.1x at 20:16, net +$763.45

20:31, $GRADDY hits 88%. 530 real buyers, dev 0.4%. approved $300

20:44, GRADUATED → raydium

20:51, sold into the migration pump. 5.6x, net +$1,361.90

and 2 curves it refused to touch:

15:40, $WASHY. one wallet made 64% of the buys to fake the fill. killed

18:00, $DUMPTRK. dev sitting on 22%, unlocked, waiting for grad to dump. killed

by 21:00: 9 graduations caught, average +$430.07, 2 fake curves killed, 0 rugs eaten

six agents. the bot can spend exactly $0

curve watches the fill speed. wash checks who's really buying. rug checks the dev. snipe gets in at 88% in 0.3s. ledger books every cent

the only button that moves money is mine

sold $5,668, bought $1,730, fees and jito tips -$68.38

i never bought a launch. i bought graduations

built by RUGIKK

[xinces @xinc3s]

EVERYONE TRIED TO GET RICH ON AI BY BUYING NVIDIA. I MADE $48,300 FROM AI WITHOUT OWNING A SINGLE AI STOCK

here's the trade nobody on my feed was talking about

every AI data center is a building that eats electricity. one big campus can pull as much power as a small city. and it has to sign for that power years before it opens

so i stopped asking "which AI company wins" and started asking "who do they all have to pay?"

i gave the question to 3 agents:

→ HAIKU 5.5 read every US utility, power producer and grid supplier filing. 10-Ks, 10-Qs, 8-Ks, earnings calls. every mention of "data center", "large load", "interconnection"
→ JEV sorted it. one question per company: is this demand signed, with a number, or just talk on a call? it can't write a single word, it only picks. talk got thrown out
→ OPUS 5.5 read only the survivors and wrote one page each: how much of next year's demand is already under contract, and what the stock price assumes. it never saw the chart first

out of 140+ companies, 6 made it through. i bought 4

7 months later:
→ 3 of them up between 31% and 64%
→ 1 down 11%. sold it. the contract got delayed
→ net: +$48,300

i didn't predict anything. the agents just read what was already signed while everyone else watched nvidia's chart

in a gold rush, don't guess who finds gold. read who's selling the shovels

the prompt i used ↓

"Build me a research desk in this folder. Read-only, no brokerage.
1. Reader (haiku): pull the latest 10-K, 10-Q, 8-K and call transcript for every US utility and power producer from SEC EDGAR. Save every mention of data centers, large loads or interconnection with the number and a source link.
2. Gate (jev): for each company ask "is this demand signed, with a contract and a number?" Keep only scores above 0.85.
3. Analyst (opus, never opens prices): one page per survivor: signed demand vs current capacity, what has to go right, the one fact that would kill the idea.
4. Code adds the stock move last. Show me the agent files first. Nothing runs until I say go."

[mirku @mirku21]

THIS IS A PURE F*CKING DIAMOND

Anthropic just documented one of the best setups for Claude Code

The idea: run Opus 5.5, Sonnet 5.5, and the new Haiku 5.5 together in a tight tripartite loop, each doing only what it was built for

Opus-level planning on every big architectural decision, without paying Opus rates for routine code edits or repo searches

You just start Claude Code like this:

CLAUDE_CODE_SUBAGENT_MODEL=haiku claude --model opusplan

here's how it splits:

→ Opus 5.5 runs plan mode: architects the system blueprint, locks contracts, and signs off on the diff
→ Sonnet 5.5 executes the plan in the main session: writes the code, applies patches, and runs test suites
→ Haiku 5.5 swarms the repository in parallel: maps the AST, greps call sites, and parses docs at 340 tok/s with native 1M context
→ you stop burning Opus context on routine files and keep your main session under 10k tokens

Opus only steps in at three moments:

→ before a plan locks (architectural design)
→ when the same test fails twice (debugging deadlocks)
→ before calling it done (signing off on git diff)

Opus plans, Sonnet builds, Haiku swarms

JEV engineering is the same move one layer down: the forks that need no thinker (which file, which tool, retry or stop) resolve in code in 15ms, and the frontier models only see the ones that split

the full tree
> Opus 5.5 architects the implementation plan (opusplan mode)
> Sonnet 5.5 runs the main execution session and writes code
> search maps the repo AST on Haiku 5.5 (340 tok/s, 1M context)
> grep locates token call sites on Haiku 5.5 (isolated subagent)
> docs pulls API specifications on Haiku 5.5 (0.05x cost)
> Opus 5.5 verifies synthesized diff before final sign-off
> JEV routes micro-forks in milliseconds

paste the tree and this prompt into Claude Code ↓

"Reconfigure my Claude Code setup around this tree:
1. Run with model 'opusplan' so Opus 5.5 plans the architecture and Sonnet 5.5 implements the code
2. Set CLAUDE_CODE_SUBAGENT_MODEL=haiku so all search, grep, and documentation subagents swarm on Haiku 5.5 with its native 1M context
3. Configure fallbackModel to ['sonnet', 'haiku'] in settings.json to guarantee 100% session availability
4. Add one rule to CLAUDE.md: delegate all repository searches and test runs to Haiku 5.5 subagents to keep main context under 10k tokens
5. Route fast deterministic forks (tool choice, path selection, retries) through JEV
Show me every change as a diff first. No edits until I say go"

↳ https://t.co/a2VGzSHZEn

save it, it'll come in handy

[软糖塞牙缝 @popunklimpo]

@MoonDevOnYT offside arm trapped by a game, hope won't see similarly next week

[软糖塞牙缝 @popunklimpo]

@Mikadzyki_NFT offside arm trapped by a game, hope won't see similarly next week

[软糖塞牙缝 @popunklimpo]

@Mr_Buzzoni Cool stuff

[Joseph Aaron Yates @yatesjoseph12]

This game console is a blast! Perfect for gaming fun with friends and family. https://t.co/d6182td5nz

[Joseph Aaron Yates @yatesjoseph12]

https://t.co/yLw0eKTHFw