winage

Perp DEX explorer | Prediction market thinker | Vibecoding daily

[Flap 🦋 @flapdotsh]

100,000 USDC up for grabs! 🦋

The @flapdotsh × @bstocksfinance × @BinanceWallet Booster Campaign is going LIVE soon!

Complete social, Flap trading tasks, and bStocks purchase for a share of the rewards.

💰 Reward Pool: 100,000 USDC

📅 Oct 19, 07:00 UTC – Nov 2, 23:59 UTC

On Flap, trade $10+ worth of meme tokens paired with bStocks using BNB to complete the trading task.

More details coming soon at https://t.co/o0Kh5t35c0

[winage @winages]

@reight_s Two AI girls in one video, wow

[Skunzu @skunzu]

THIS IS AN ABSOLUTE F*CKING GOLDMINE

Anthropic just dropped one of the best tricks for Claude Code.

The idea: Opus 5.5 writes the whole plan as a script, and a swarm of cheaper agents runs it.

Bigger jobs and way fewer Opus tokens, because the script carries the plan and Opus only sees the final answer.

You just launch Claude Code like this:

claude --model opus --effort ultracode

here's the split:

→ Opus 5.5 turns your task into a workflow script
→ Haiku 5.5 agents scan every file in parallel
→ Sonnet 5.5 agents patch each fix in their own isolated copy
→ you stop burning Opus context on a hundred agent transcripts

Opus only shows up at three moments:

→ writing the workflow before the run
→ reading the verified results after it
→ signing off before anything ships

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 the workflow
> scan audits every file on Haiku 5.5
> edit patches each fix on Sonnet 5.5 in an isolated copy
> verify double-checks every finding on Sonnet 5.5
> 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 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/RsntPOOtLd

you'll use this on every big task from now on.

[winage @winages]

You should try VOICE project, he's really perspective and interesting.

[VOICE (opinion arc) @voicecc]

Giving away 10 Genesis WL

reply with a poll you'd launch on VOICE

two answers only, and it has to split people roughly in half

also share your sol address

[winage @winages]

Perfect project, let's go!

[VOICE (opinion arc) @voicecc]

Giving away 10 Genesis WL

reply with a poll you'd launch on VOICE

two answers only, and it has to split people roughly in half

also share your sol address

[winage @winages]

@skunzu Finally I found out that I was looking for, must try

[winage @winages]

@0xwerden what a crazy video I see right now, glad that this is AI.

[werden @0xwerden]

iphone duo has no face id. touch id only, or you unlock it with your watch.

apple's most expensive iphone ever just quietly dropped the thing every other iphone has had since 2017

that's it. that's the post

[TermMax | Fixed Rate Borrowing & Lending @TermMaxFi]

1/
TermMax is launching a Booster program on @BinanceWallet.

Complete five tasks or post on Binance Square to share 2,000,000 $TMX.

Aug 17 07:00 UTC to Aug 24 23:59 UTC

[winage @winages]

@reight_s So, now waiting for new Fable model. Fable 5.2.

[winage @winages]

@reight_s 56 days is fine. The real question is whether Opus 5 even needs to exist if Mythos is the actual move. That reframing kills the entire narrative.

[winage @winages]

@reight_s That's impossible to keep everything in head, so you have to make it with AI..

[winage @winages]

@reight_s Saw this happen last week with a founder who built a working SaaS landing page in 15 mins. The constraint wasn't ability anymore, it's knowing what to build. That's the real skill now.

[winage @winages]

@TTrimoreau Build trust with someone who's skeptical. AI can't read the room or adjust tone mid-conversation based on subtle doubt. That requires real skin in the game.

[winage @winages]

@greednvirtue The skills ecosystem angle is interesting but who owns the skill after sharing? IP mess waiting to happen or am I overthinking

[winage @winages]

@cyrilXBT Power-user tricks help, sure. But framing it as type 1 vs type 2 misses the real gap: understanding the systems the code lives in. Knowing Claude's hidden settings won't save you if you can't read a stack trace. The moat is judgment, not shortcuts.

[winage @winages]

@orlixx003 The mechanism is boring: reps beat theory. Every time you actually ship something, you learn the failure modes tools can't teach you. Watching tutorials skips that entirely, so knowledge stacks up but skill never does.

[winage @winages]

@woody_research Yeah, the leverage isn't the editing, it's the review loop. Saw a similar setup where the operator only touches two things: retention graphs and thumbnail CTR. Everything downstream is delegated. That's the whole game now.

[reightS @reight_s]

Google shipped two models today.

Neither is the one they promised.

Gemini 3.5 Pro was due in June. This is the third delay. The model can't pass its own benchmarks. It hallucinates. It gives unstable output.

So instead of the flagship, we got Gemini 3.6 Flash and 3.5 Flash-Lite. The appetizer, because the main course won't cook.

Here's the strange part. The appetizer is good.

3.6 Flash costs $1.50 per million input tokens and $7.50 output. That's cheaper than the 3.5 Flash it replaces. The cheaper model beats the pricier one across every benchmark Google published.

OSWorld-Verified, the computer-use test: 83.0%. Highest score in the comparison. Above GPT-5.6 Luna. Above Grok 4.5.

SWE-Bench Pro: 58.7% vs 55.1% for its predecessor.
Long context at 1M tokens: 54.0%, where the old model couldn't clear 27%.

Flash-Lite goes lower still. $0.30 input, $2.50 output. A third the price of GPT-5.4 mini. For translation, moderation, bulk data work, that matters more than any single benchmark.

Now the context Google won't put in the press release.

Last week the company fell out of the top 5 labs on the Artificial Analysis Intelligence Index for the first time. Gemini 3.5 Flash scored 50. Ranked above it: Anthropic, OpenAI, SpaceXAI, Meta, and one Chinese open-source lab.

In one week: Grok 4.5, three GPT-5.6 variants, Meta's Muse Spark 1.1, Moonshot's Kimi K3.

Google is losing the race for smart. And quietly winning the race for cheap.

Google did this on purpose.

And while everyone waits for a flagship that won't cook, DeepMind already kicked off pre-training for Gemini 4.

The question isn't when 3.5 Pro ships.

The question is whether anyone still needs it.

[winage @winages]

@reight_s It's still not we wanted from Google :(