merumput

adufhsdg
djfhkdjfhsd
ksdjhsdkjf

[merumput @bakohteki]

Dear @CryptoEternalAI

Please make an ai agent to help people with depression and anxiety feel more safe in real world reality. So they comfort with him/her self and with others
Thanks before

[Robbie Jeo 楊忠興 @robbie_jeo]

Web scraping in 2026 is broken. Here's why — and what's replacing it.

Most scraping pipelines still work like this:

1. Fetch HTML
2. Parse the DOM
3. Chunk the text
4. Embed and retrieve

It works. Until it doesn't.

Dynamic pages. Shadow DOM. JavaScript-rendered content. Anti-bot measures. The list of failure points keeps growing.

PixelRAG takes a completely different approach.

Instead of parsing HTML, it screenshots the page — then retrieves the image directly using a vision-language model.

No DOM parsing. No text chunking. Just visual understanding of the page, the same way a human would read it.

The result: a retrieval system that's far more robust against the layouts and protections that break traditional scrapers.

And it's fully open-source.

If you're building AI pipelines that pull data from the web, this is worth your attention.

Source: https://t.co/x2ehrbsm0c

[𝑝𝑢𝑟𝑒 @purelivn]

RT FOR GOOD LUCK 🍀

[Robbie Jeo 楊忠興 @robbie_jeo]

DeepSeek just dropped V4.

1.6 trillion parameters. 49 billion active. 1 million token context window.

Translation: it can read an entire codebase — or a novel — in a single prompt.

The Flash version costs /bin/bash.14 per million tokens. That's not undercutting frontier models. That's redefining what "frontier" means.

DeepSeek says they've "closed the gap" with GPT-5.4 and Gemini 3.1 Pro on reasoning benchmarks. The lag? About 3-6 months behind state-of-the-art.

Here's what nobody's saying: being 3-6 months behind while costing 10x less isn't a gap. It's a strategy.

The real story isn't that DeepSeek caught up. It's that the price-performance curve just bent in China's favor — again.

— Robbie Jeo

[Robbie Jeo 楊忠興 @robbie_jeo]

VinMotion’s Motion 2 is a versatile humanoid robot designed for both factory and home use.
It can lift up to 40 kg, automatically stand back up after falling, and take care of its own charging or battery swapping.

It also supports multiple languages, including Vietnamese and English. So it can communicate and perform real-world tasks effectively.

@spaceandtech_

[Robbie Jeo 楊忠興 @robbie_jeo]

Quick & powerful tutorial on adding Agentic Skills to Claude!

Skills are reusable Markdown templates that turn Claude into a specialized expert for specific tasks — no need to repeat long prompts every https://t.co/tipZ3GbY5V Install (Collections):

Run these npx commands in Claude to instantly add dozens of ready-made skills:
• LUNARTECH collection (200+ skills)
• 43 College practical skills
• Automation tools20 Highly Recommended Skills (categorized):

Core Agentic
orchestrate • browser-qa • writing-plans • executing-plans • verification-before-completion • systematic-debugging • subagent-driven-development

Web &Coding
code-review-excellence • sql-optimization-patterns • git-advanced-workflows • e2e-testing-patterns

Python Backend
python-packaging • async-python-patterns • microservices-patterns

Design & UI
dark-mode-design • glassmorphism-design • micro-interactions

Documents & Office
docx • xlsx • pdfSpecialized extras
rails-audit-thoughtbot • feishu-assistant • social-media-scout • web-browser • media-transcriber • video-creator • follow-builders

How to use:
Just type /skill-name (e.g. /orchestrate or /code-review-excellence)You can also install any single skill with one command.

Super useful for devs, designers, and power users who want Claude to be way more consistent and capable. Worth bookmarking!

[Robbie Jeo 楊忠興 @robbie_jeo]

Advanced robotic hand handling prayer beads with high precision and control.

[Robbie Jeo 楊忠興 @robbie_jeo]

Every company is facing the same AI decision right now.

Go fast and automate everything?
Or go slow and keep humans in control?

The data is in: 76% of enterprises are choosing the slow path.

Not because they're conservative. Because they learned from the first wave.

Here's what "controlled AI adoption" actually looks like in 2026:

→ Assisted decision-making, not autonomous action
→ Human review on high-stakes outputs
→ Sandboxed pilots before production rollout
→ Clear audit trails for every AI recommendation

OpenAI's own research shows the pattern. Their enterprise customers using ChatGPT for finance teams don't let the AI send reports. They let it draft reports. Human signs off.

Same for legal. Same for medical. Same for anything that matters.

The companies moving fastest on AI aren't the ones removing humans.
They're the ones augmenting them.

Meanwhile, Baidu just reentered local services with an AI-first approach — but they're hitting the same wall. Search-driven discovery works. The transaction loop doesn't. Not yet.

The gap between "AI-powered" and "AI-autonomous" is where the real battle is happening.

The winners won't be the companies with the smartest models.
They'll be the ones who figured out where to trust AI — and where not to.

Your move: Audit one workflow this week. Where does AI assist vs decide?

Link in first comment 👇

— Robbie Jeo

[Robbie Jeo 楊忠興 @robbie_jeo]

@OpenAI just created a $100 tier because developers were hitting walls.

ChatGPT Plus ($20) → not enough Codex usage.
ChatGPT Pro ($200) → overkill for solo devs.

So they built the Goldilocks tier:
• 5X more Codex requests than Plus
• Priority access during peak hours
• $100/month — right in the middle

This isn't about features. It's about usage limits.

The message? OpenAI knows Anthropic is eating their lunch with Claude Code. They're fighting back by unblocking the developers who actually build things.

Your move, Anthropic.

Link in first comment 👇

— Robbie Jeo

[Robbie Jeo 楊忠興 @robbie_jeo]

This AI upscales images with only ~10% of the usual training cost.

PolyU and OPPO just introduced VOSR — a vision-only generative model for image super-resolution that doesn't rely on large pretrained text-to-image (T2I) diffusion models.

Why this matters:
→ Most recent generative SR methods adapt massive T2I models pretrained on web-scale image-text data (very expensive)
→ VOSR trains purely on vision data only
→ training cost drops to about one-tenth (~10%) of representative T2I-based methods
→ It outperforms previous vision-only SR methods and remains competitive with T2I-based SOTA in perceptual quality, with faster inference and fewer artifacts

For developers building image enhancement tools, this is a major cost breakthrough. For users, it opens the door to high-quality upscaling that is more efficient and practical.

Paper accepted at CVPR 2026, code & models already available on GitHub.

Blurry photos are about to look a lot sharper.

— Robbie Jeo

https://t.co/YXFd5YgHpg

[Robbie Jeo 楊忠興 @robbie_jeo]

The AI coding wars are heating up.

@OpenAI, @Google, and @AnthropicAI are all pushing hard into AI coding tools. Claude Code from Anthropic went viral. OpenAI released updates to Codex. Google added more coding features to Gemini.

Here’s what matters:
Coding isn’t being fully automated — it’s being commoditized.
Simple, repetitive code that many high-paid engineers used to write is now fast and cheap to generate with AI.

But the engineers who architect complex systems, handle tricky edge cases, and critically review AI output? They just became significantly more valuable.

The market is splitting in two:
→ Commodity code (fast & cheap)
→ High-judgment engineering (premium & irreplaceable)

We’ve seen the same pattern in writing and design.
The tool doesn’t replace taste and judgment — it exposes who actually has it.

— Robbie Jeo

https://t.co/GPY27hEVMw

[Note To Self @Alphafiles1]

God will make it happen for you .

[Robbie Jeo 楊忠興 @robbie_jeo]

Shopify just made every e-commerce store AI-native.

AI agents can now manage your store - inventory, pricing, customer service, marketing. Autonomously.

This is not a plugin. This is commerce infrastructure redesigned for AI.

The numbers:
- $3.6B market in 2024, projected $282.6B by 2034
- McKinsey estimates $1T in agentic commerce revenue by 2030

The implications:
- Solo founders can run stores that used to need 10 people
- 24/7 AI customer service without hiring
- Dynamic pricing that adjusts in real-time

Every industry is getting an AI agent layer. E-commerce just got its turn.

- Robbie Jeo

[Robbie Jeo 楊忠興 @robbie_jeo]

#Seedance 2.0 just went live on ImagineArt.

Cinematic results built-in. Audio-video referencing. Next-gen physics for action and motion.

We went from AI can generate a clip to AI can generate a MOVIE in 6 months.

The pace is insane. What took studios months and millions of dollars now takes minutes and a prompt.

If you are still waiting to try AI video generation - this is your sign.

- Robbie Jeo

[merumput @bakohteki]

Wow

[Robbie Jeo 楊忠興 @robbie_jeo]

bro created an AI-powered job search system using Claude that evaluated 740+ job offers… and actually landed him a Head of Applied AI role 🔥

AND IT'S NOW OPEN-SOURCE.

It scans 45+ company career pages, rewrites your CV for every job with ATS optimization, scores them A-F, and even prepares interview stories.

The repo is packed with:
• 14 skill modes (evaluate, scan, pdf, batch, etc)
• Beautiful & satisfying Go terminal dashboard
• ATS-optimized PDF generation using Playwright
• 45+ big companies pre-configured (Anthropic, OpenAI, ElevenLabs, Stripe, Vercel, etc)
• Batch processing + human-in-the-loop (you still control everything)

GitHub: https://t.co/9dgeZEiHYM

Massive respect to @santifer — this tool is next level 👏

[Robbie Jeo 楊忠興 @robbie_jeo]

bro created an AI-powered job search system using Claude that evaluated 740+ job offers… and actually landed him a Head of Applied AI role 🔥

AND IT'S NOW OPEN-SOURCE.

It scans 45+ company career pages, rewrites your CV for every job with ATS optimization, scores them A-F, and even prepares interview stories.

The repo is packed with:
• 14 skill modes (evaluate, scan, pdf, batch, etc)
• Beautiful & satisfying Go terminal dashboard
• ATS-optimized PDF generation using Playwright
• 45+ big companies pre-configured (Anthropic, OpenAI, ElevenLabs, Stripe, Vercel, etc)
• Batch processing + human-in-the-loop (you still control everything)

GitHub: https://t.co/9dgeZEiHYM

Massive respect to @santifer — this tool is next level 👏

[Robbie Jeo 楊忠興 @robbie_jeo]

Claude can now build you a full AI YouTube channel.

Not help with scripts. BUILD the entire channel. From zero to monetized in 90 days.

Here are 3 prompts that replace a $10,000/month creator agency:

1/ Niche and Positioning
Analyze the top 10 YouTube niches with high CPM, low competition, and strong evergreen demand.

2/ 90-Day Content Roadmap
Build a 90-day content calendar with 12 video titles optimized for search and upload schedule.

3/ Growth and Repurposing Engine
Take a YouTube video script and repurpose it into: Twitter thread, LinkedIn post, 3 Shorts hooks, email newsletter.

The prompts alone are worth saving. Claude just became a full content team.

Full 7 prompts: https://t.co/0wa69fhvS6

- Robbie Jeo

[The Soul @FuelTheSoul_]

RT FOR GOODLUCK 🍀

[yxf @iamyxf]

WHOEVER TOUCHES THIS POST WILL BECOME FINANCIALLY STABLE THIS YEAR. AFFIRM AMEN!!! 🙏🏼

[Robbie Jeo 楊忠興 @robbie_jeo]

Microsoft committed $1.7 billion to Indonesia for AI & cloud infrastructure.

But outside investment means nothing if the people inside aren't ready.

The opportunity is here. Question is who moves first.

— Robbie Jeo

[Law Of Manifestation @Law_of_manifest]

10 years of badluck if you skip this post