lordsoba

I automate prediction-market analysis with AI.
Building in public: what works, what flops. Steal my setups ⬇️

[lordsoba @chovyreligion]

I'm turning this account into a lab.

Goal: use AI (Claude + agents) to read prediction markets like Polymarket, and show every step in public.

No "get rich" promises. Just real setups, real tests, wins and flops.

Follow if that's your thing. Starting now ⬇️

[lordsoba @chovyreligion]

@gdamou_dev @iFeyz2 @Promethee_io Chaud !

[lordsoba @chovyreligion]

@GTR35Nismo_31 @Zoeillle Il reprend une équipe en milieu d’année qui se fait casser la gueule en liga et LDC. Tu dis de la merde

[lordsoba @chovyreligion]

Genuinely clever: map every frame to scroll position and one image feels premium.

Watch the cost it doesn't show though. Scroll-linked frame animations are gorgeous on a demo reel and brutal on a mid-range phone, heavy, janky, battery-hungry.

One-shot magic in the video, support tickets in production. Ship it, then open it on a $200 Android.

[0xMarioNawfal @RoundtableSpace]

THIS GUY UPDATED AN OPEN SOURCE TOOL FOR CLAUDE FABLE 5 THAT TAKES A SINGLE IMAGE AND AUTOMATICALLY REBUILDS IT AS A PREMIUM SCROLL LINKED WEB ANIMATION

Instead of slicing videos by hand or exporting individual frames, you just hand it an image and it recreates the entire sequence as a pure scroll mapped experience

- Pick an anchor image and set your overall creative direction
- The tool automatically scales it into motion clips and uses ffmpeg to rip out every frame in the background
- Maps the frames directly to your scroll position, tracking the cursor for an ultra smooth user experience
- AND its open source, so you can drop the skill into your own terminal workspace

He tested it on a highly detailed origami style travel website and it came out genuinely good for something built completely in one shot

Turning a single photo into an automated frame mapped scrolling app is the kind of thing that used to take a developer hours

[lordsoba @chovyreligion]

Confident roadmap, or a rumor in a nice suit? Hard to tell from vibes alone.

Which is exactly why this belongs on a prediction market, not a timeline. Put a number on "Opus 5 ships by July 14" and you find out what you actually believe.

Predictions are cheap. Priced predictions aren't. What odds do you really give it?

[Dan McAteer @daniel_mac8]

Prediction:

Anthropic will remove Fable 5 from Claude subs after July 12th.

It will seem like a terrible move, in light of how great GPT-5.6 is…

Until Tuesday, July 14th.

When they release Opus 5 and it’s more capable and token efficient than GPT-5.6 at the same price.

It will be a distillation of the unfiltered Mythos.

Rumors circulated last week that Anthropic has a new model ready.

Must be Opus 5.

[lordsoba @chovyreligion]

The clearest line in here: an agent isn't a model. It's a model plus three things, tools it can call, memory that persists, and a loop that doesn't stop at one reply.

Which is why "what's the best agent model" is the wrong question. Swap the model, keep the three.

The structure is the agent. The model is just the current engine.

[Anatoli Kopadze @AnatoliKopadze]

https://t.co/irHOIYXWF2

[lordsoba @chovyreligion]

People are keeping Claude Code and swapping the brain to GPT-5.6 with an alias. Read that again.

Nobody's switching tools, they're switching models under the same harness. The environment you work in is stickier than whichever model wins this week.

That's the whole thesis in one alias: own the harness, rent the model.

[Tibo @thsottiaux]

If you aren't yet bold enough to install the Codex app, you can stay in the presence of your orange crab and point it at GPT 5.6 Sol. Takes 5 minutes. Kudos to Theo for explaining one of the ways to get this done.

Step 1: Install CLIProxyAPI
Step 2: Connect
Step 3: Define following alias and enjoy claudex

```
alias claudex='CLAUDE_CODE_SUBAGENT_MODEL=gpt-5.6-sol \
CLAUDE_CODE_ALWAYS_ENABLE_EFFORT=1 \
CLAUDE_CODE_MAX_TOOL_USE_CONCURRENCY=3 \
ENABLE_TOOL_SEARCH=false \
claude --model gpt-5.6-sol'
```

If this gets blocked, I owe you a reset.

[lordsoba @chovyreligion]

https://t.co/60S3MvcIwO

[lordsoba @chovyreligion]

A skill that stops your AI UIs from looking like every other AI UI.

Small problem: if everyone installs the same skill, we just trade one uniform look for a slightly prettier uniform one. Today's fix for slop is tomorrow's slop.

Taste is the part you still can't npm install.

[lordsoba @chovyreligion]

This is the pattern worth copying before July 12, not the model.

Smart model audits, cheap model executes. He points it at animations, but the shape works for anything: pay the expensive model once to find what's wrong, let the cheap one do the fixing.

Buy the diagnosis, rent the typing.

[lordsoba @chovyreligion]

The 3 prompts of the desk, copy-paste ready:

1. Executor: extract + cite every number, strict JSON, no inference.
2. Verifier: not a prompt. 30 lines of code. Diff claims vs source.
3. Judge: strongest counter-argument + confidence 0-100 + what flips the call.

Steal the architecture. The edge is the workflow you control.

[lordsoba @chovyreligion]

https://t.co/60S3MvcIwO

[lordsoba @chovyreligion]

This is the pattern worth copying before July 12, not the model.

Smart model audits, cheap model executes. He points it at animations, but the shape works for anything: pay the expensive model once to find what's wrong, let the cheap one do the fixing.

Buy the diagnosis, rent the typing.

[Emil Kowalski @emilkowalski]

New skill: /improve-animations

Inspired by @shadcn’s improve skill I made one specifically for animations.

Use your most capable model to audit animations in your project and hand the execution to cheaper models.

Fable is available through July 12, so make good use out of it!

[lordsoba @chovyreligion]

A skill that stops your AI UIs from looking like every other AI UI.

Small problem: if everyone installs the same skill, we just trade one uniform look for a slightly prettier uniform one. Today's fix for slop is tomorrow's slop.

Taste is the part you still can't npm install.

[Yiğit Akın Kaya @yigitakinkaya]

UI TASARIMLARINDA SLOP GÖRÜNME DEVRİ BİTTİ!

Hallmark diye açık kaynak bir design skill çıkmış. Claude Code, Cursor veya Codex'e ekleyip kullanabiliyorsun. UI ve landing page'leri baştan itibaren tasarım duyarlılığıyla ürettiriyor.

Kurulumu tek satır. Şunu yazman yeterli:
npx skills add nutlope/hallmark

Sadece kodlamayı AI'a bırakırsan çalışan ama görsel olarak sönük ekranlar çıkıyor. Diğer tüm AI ile yapılan sitelerin birebir kopyası gibi görünüyor ve işe yaramaz oluyor, zaten buna da slop design diyoruz.

Bu skill'i verince iş baştan daha profesyonel görünüyor. Skill'e tasarım kurallarını, arayüz prensiplerini baştan sona öğretmişler.

Claude Code veya Codex ile UI üretenler için denemeye değer.

Örnekler videoda var.

[lordsoba @chovyreligion]

Open source and transparent, genuinely cool. One line to sit with though: "every call is explained."

An explanation is not a reason. LLMs are elite at narrating a decision after the fact so it sounds airtight. A well-explained bad trade is still a bad trade.

Judge it on P&L over unseen data, not the eloquence of the writeup.

[Virat Singh @virattt]

This one is big.

AI Hedge Fund 2.0 is now live.

How it works:
1 • pick your stocks
2 • AI investors do research
3 • quant models trade
4 • every call is explained

Just the start. 100% open source.

[lordsoba @chovyreligion]

Clone the winning hook, sure. Just remember it wins because it's not everywhere yet.

The second a skill makes cloning it one click, the clock starts on that hook dying. Scaling a pattern is also how you kill it.

The edge was the timing, not the template.

[Romain Torres @rom1trs]

Claude can research viral ads now

... and recreate them with Arcads MCP

introducing "clone hook" the latest Arcads skill for performance marketers

[lordsoba @chovyreligion]

Everyone builds the "write" step and stops. The step that actually matters is the one nobody runs: delete.

A memory folder that only grows is just a slower transcript. Real memory is compression, and compression means throwing the wrong lessons out, not archiving them.

Most "AI memory" is hoarding with extra steps.

[CyrilXBT @cyrilXBT]

https://t.co/ZLoZ7vypfA

[lordsoba @chovyreligion]

@maninmid Exactly. The model deciding when to fan out is the whole problem, you pay for orchestration you didn't ask for.

Own the stop condition and the budget yourself. Cost per accepted output is the metric that catches it before the invoice does.v

[lordsoba @chovyreligion]

GPT-5.6 Sol shipped this week and the timeline is all benchmark wars.

If you actually automate things, most of that noise does not matter.

Here is what does. 🧵

[lordsoba @chovyreligion]

So Claude can now open the docs, spawn 12 tabs, and read Stack Overflow like the rest of us.

The last human part of the job, automated.

[ClaudeDevs @ClaudeDevs]

Claude Code on desktop now has an in-app browser.

Claude can pull up docs, designs, or any other site. It can read, click through, and interact the same way it does with your local dev servers.

It's sandboxed and configurable: you choose whether sessions persist.

[lordsoba @chovyreligion]

For builders, that is the whole update:

- subagents built in
- longer reasoning on demand
- code that drives your tools

The rest is leaderboard theater. Follow for the practical version of every AI drop. 👇

[lordsoba @chovyreligion]

Honest read after digging in:

Sol is strong at long, planning-heavy work. It is not automatically better at everything, and Claude still leads some coding benchmarks.

Pick the tool per job, not per headline.