小芽的汽车分享站

聊聊车主真实经历，给日常用车多些灵感｜车友交流 @mothio1201

[小芽的汽车分享站 @Dj_Mestizo]

@SolSt1ne equipment decks drifting on that garage set the mix fixed

[小芽的汽车分享站 @Dj_Mestizo]

@Mooncaller56 对，滑动离合轻得有点意外，堵车左手不怎么累。你是刚上手还是在看车？

[小芽的汽车分享站 @Dj_Mestizo]

南京这台钱江赛600像成年校队学姐，车头低，整流罩紧，身体一趴下来气氛马上变。转速越高，性格越明显。苏州、无锡、常州想看动力和日常表现，主页有后续。

[小芽的汽车分享站 @Dj_Mestizo]

@jelly_belly_ily 确实，这车在北京天天堵着骑有点憋屈，拉不起转速性格都闷着。你平时跑哪段山？

[小芽的汽车分享站 @Dj_Mestizo]

@lyntrice88 哈哈对，那种一趴下去整个人语气都变的感觉，仿赛专属。你骑的是哪台？

[小芽的汽车分享站 @Dj_Mestizo]

@2MUCH2MATCH 对，低扭确实比我预想顺，堵车跟车不用一直高转吊着。你是在市区骑得多还是周末才拉出去？

[小芽的汽车分享站 @Dj_Mestizo]

@l0lsmileyface14 低扭还行，市区两三千转就能走，踩深一点反而有点闷。你是想拿它天天通勤还是周末撒野？

[小芽的汽车分享站 @Dj_Mestizo]

@raqueloser 夏天市区等红灯确实有点烤，走起来就好了。要不要我专门去市区跑一圈测测？

[小芽的汽车分享站 @Dj_Mestizo]

@fUCKOFDEATH 哈哈我更偏学姐那种，不是纯御姐，得看油门开到哪一段，转速拉高以后气场完全不一样。

[小芽的汽车分享站 @Dj_Mestizo]

@AIRWALK101 确实，高速上趴下去被包得很稳，基本感觉不到乱流往胸口砸。你也骑仿赛吗？

[Edmunds @edmunds]

The Tesla Roadster is real and it can fly and it looks incredible in person

[Mr. Buzzoni @Mr_Buzzoni]

MemT 1.0 might be the first memory-naitve LLM that actually remembers your project between sessions

@MemOS_dev team dropped it on Oct 20

it's a memory-native MoE
writing, retention, updating and reading all happen inside the model

the architecture bit that makes it work:
every Transformer block gets its own memory slot running next to attention

the model learns which bits to erase, write and keep between turns, so your context lives in the weights, not in a vector store

the real tell for me:
give it a constraint, a past decision, a correction you made last week
it reuses them next task
no re-explaining the whole project

a model that stops forgetting the brief is a model you stop babysitting

tech report + weights are open:
↳ https://t.co/ss0VgLqN2K

[mirku @mirku21]

THIS IS A PURE F*CKING GOLDMINE

Anthropic just shared the sharpest trick for Claude Code

The idea: let Opus 5.5 write the plan as a script, and swarm the new Haiku 5.5 in parallel

Huge sweeps with zero Opus burn: the script holds the loop and Opus only sees the final diff

You just launch Claude Code like this:

claude --model opus --effort ultracode

here's the split:

→ Opus 5.5 turns your objective into an executable workflow script
→ Haiku 5.5 swarms every file in parallel at 340 tok/s to extract candidates
→ Sonnet 5.5 patches each fix and cross-checks findings in its own isolated copy
→ you stop burning Opus context on a hundred agent transcripts

Opus only shows up at three moments:

→ authoring the workflow script before the run
→ reading verified findings after the swarm finishes
→ signing off before anything ships to git

Opus plans · Haiku scans · Sonnet ships

JEV engineering is the same move one layer down: the forks that need no thinker (which file, which tool, retry or stop) go to JEV in milliseconds, and the big models only see the ones that split

the full tree
> Opus 5.5 runs the main session and writes workflow.js
> pipeline() fans out the new Haiku 5.5 to audit every file
> parallel() schema validators drop false positives in V8
> Sonnet 5.5 patches each fix in an isolated worktree
> verify cross-checks all diffs before reporting
> Opus 5.5 reads only what survived verification
> JEV routes micro-forks in milliseconds

paste the tree and this prompt into Claude Code ↓

"Rebuild my Claude Code setup around this tree:
1. Run the main session on Opus 5.5 with ultracode on
2. In every workflow, run read-only scan stages on the new Haiku 5.5 and edit stages on Sonnet 5.5, each edit in its own isolated copy
3. Add one rule to CLAUDE.md: every finding gets an independent verifier agent before it's reported, and fix loops stop after two rounds with no progress
4. 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/PcTdH7dyjb

save it, it'll come in handy

[RetroChainer @RetroChainer]

$200 A MONTH FOR GROK BOT, AND ONE RULE DECIDES WHAT IT DOES WITHOUT ME: IF IT CAN'T BE UNDONE IN UNDER A MINUTE, IT DOESN'T HAPPEN.

draft, file, tag, summarize, research, reconcile. it finishes all of that alone and logs it.

send to anyone outside, spend money, publish, sign up, agree to terms. it parks every one of those and waits.

my sales bot had 36 drafts ready this week. 0 went out until i opened them.

that one line is step 9 of 10. the other nine are in the picture.

save this.

[Mr. Buzzoni @Mr_Buzzoni]

Claude Code tip: once Opus 5.5 is your main model, stop letting your Fable 5.1 quota go to waste

put it on call with /advisor

run /advisor fable

Opus 5.5 does all the building
Fable 5.1 follows the session in the background and cuts in three times:

-> before a plan: does this hold up?
-> when one error keeps returning: am I fixing the wrong thing?
-> before "done": what slipped through?

Fable 5.1 checks the direction. Opus 5.5 does the typing

Jev engineering takes it one step further:
the small forks (which file, which tool, retry or stop) go to Jev in under half a second, so the pricey models only see the calls that can change the result

where this setup pays off most right now: long reverse engineering runs with REA, the open source MCP that lets Opus 5.5 take any app apart down to the binary

• the full setup
> Opus 5.5 on high drives the main session
> subagent one reads the code
> subagent two edits and runs tests
> subagent three checks the docs
> all three on medium
> Fable 5.1 on call

give Claude Code the tree plus this prompt 👇

"Configure Claude Code to match this tree:

1. Reuse suitable subagents from ~/.claude/agents and .claude/agents.

> Draft new ones only where a role is missing
> Give each model: opus, effort: medium
> List any that pin another model and leave them

2. Set main session effort to high with effortLevel in ~/.claude/settings.json

3. Look for env vars that turn off the advisor (CLAUDE_CODE_DISABLE_ADVISOR_TOOL, DISABLE_TELEMETRY, anything that blocks flag fetching) and CLAUDE_CODE_EFFORT_LEVEL, which overrides subagent effort. Report them and leave them be

4. Add a rule to ~/.claude/CLAUDE.md: check with the advisor before a big plan, when an error repeats, and before marking a long task done

Show me the changes first, edit nothing until I confirm."

↳ https://t.co/pkgl3RZv47

[Kaiju @kaijucode]

F*ck YES... I made a prompt that stops Claude from making things up while it reverse-engineers a binary

the shift: Sonnet 5.5 maps the binary for cents → Opus 5.5 rebuilds the hard logic → Fable 5.1 only reviews what already passed tests

here’s 3 step guide:

step 1 → drop the CLAUDE.md from the image into your project root (every claim now needs an address or a test behind it)
step 2 → run Sonnet 5.5 and ask for a function map: ADDRESS | NAME | ROLE | EVIDENCE | CONFIDENCE

step 3 → switch to Opus 5.5 with that map and say something like:

“Reconstruct the behavior of [FUNCTION]. Write a spec plus two alternative interpretations and the minimum tests that tell them apart”

the result: you pay ~$0.30 per pass instead of $1.50 on Fable, and you catch the “this loop is CRC32” mistake before it lands in your code

Paste it now. Then read article below to master Claude reverse engineering:

[Flandermaxx @Flandermaxx]

WHOEVER LEAKED THIS JEV DECISION GLOBE SHOULD BE SETTING THE STANDARD

I thought running a decision through an agent meant prompting the model, parsing its text, and praying the next step understood what came back.

Apparently not.

This globe's setup is much more interesting:

→ STATE packs the current case into a typed input before any decision call, no raw chat, no "figure it out from context"
→ QUESTIONS pins the Choice schema to a fixed set of options, nothing outside the schema can come back, not even on retry
→ JEV returns a choice with confidence in 3ms, $0.042 per million input tokens, output is free, no LLM burned on routing
→ GATE compares the confidence against a 0.85 threshold, anything below drops straight to human review, no silent pass-through
→ CODE acts on the gated choice deterministically, flags 5 action-rate drifts the author never saw against the baseline

so you get:

States typed. Schemas pinned. Gates thresholded. Actions deterministic.

the interesting part isn't making the agent smarter.
it's giving every decision a gate the code can't argue past.

Instead of running 10,000 decisions through a full model at $8-12 and reading the JSON yourself, you run them through JEV at $0.42 total, 256ms median, and let the gate send anything under 0.85 to a human.

Type the state. Gate the choice. Code the act.

[小芽的汽车分享站 @Dj_Mestizo]

@Arindam_1729 the feel of cumbia groove is not fine as explained percussive down truly underrated without going wild at pitch full

[小芽的汽车分享站 @Dj_Mestizo]

@leopardracer up this in cheap speakers honest new budget headphones quality more capable all can than wild tune stuff always think i'm sitting at world

[Deni @0xDeniAi]

this is getting f*cking dangerous...

REA creator just dropped 20 GitHub repos to build a f*cking reverse engineering studio:

[01] analysis core
[02] mobile & apps
[03] binary & low-level
[04] network & runtime
[05] rebuild & verify

how it runs: binary -> REA -> SPEC.md -> rebuild -> oracle -> retry

point it at any app and watch it explain itself.