myun wang

[AlloomiAI @AlloomiAI]

Claude Tag is a strong signal:

The next AI interface is not a blank chat box.

It is the context layer behind the work:
people, projects, decisions, docs, issues, and follow-ups.

Agents don’t become useful just by getting better prompts.
They become useful when they inherit the context of the work.

[jack friks @jackfriks]

claude is down with a major outage for everyone

except for the government

[AlloomiAI @AlloomiAI]

Your AI assistant doesn't need a bigger memory.

It needs a garbage collector.

Most memory systems just keep appending context and hope retrieval will sort it out later.

That works until stale notes, old decisions, and random leftovers start showing up like facts.

Useful memory needs time, decay, and a way to separate signal from residue.

That's a big part of what we're building in OpenLoomi: local-first, inspectable memory for AI agents.

Repo: https://t.co/t7Sk3aW0rD

[AlloomiAI @AlloomiAI]

OpenLoomi takes a different path: build a global context, then retrieve the right pieces when needed. Open source, inspectable, designed for better accuracy and fewer tokens. Try it: https://t.co/Kg8Xq5SxT8

[AlloomiAI @AlloomiAI]

omg, it happened again: the thread finally understood the work, then everything got stuffed back into the context window.

[Tianyi Cui @tianyi]

发现了一个对于 Harness 开发者似乎很有用的网站 https://t.co/zpYSH0hj6J

里面有 Claude Code、OpenCode、Codex、Pi、OpenClaw、Hermes 等各种 Harness 的各个版本的 system prompt 以及 tool schema。

Tip: 点击右上角“Trace”是看上去比较舒服的另一个视图。

[myun wang @wang_myun]

@AlloomiAI So what you're saying is that your product can organize my historical information into a 'reusable memory,' and then draw on it every time I use it? That sounds great — I'd be happy to give it a try.

[Mike Futia @mikefutia]

Google Nano Banana Pro 🍌 is crazy good at static ads...

But it only generates one image at a time.

This n8n AI Agent generates 100s of ad variations in minutes, fully automated.

All inside n8n + Nano Banana Pro.

Perfect for DTC brands & agencies who need massive creative volume to test and beat ad fatigue.

Let's be honest:

Manual ad creation takes forever.

You're generating one image at a time, tweaking prompts, downloading files, trying different angles.

And by the time you have enough variations to test, you've wasted hours.

This n8n automation solves it:

→ Upload ONE product reference image via n8n form
→ OpenAI Vision analyzes your product automatically
→ AI generates custom image prompts (you choose quantity: 50, 100, 1000+)
→ Nano Banana Pro creates all variations in parallel
→ 4K studio-quality output with perfect text rendering
→ Images auto-stored in Box for instant access

No manual prompting. No one-by-one generation. No wasted hours.

What you get:

→ Hundreds of ad variations from one upload
→ Different angles, backgrounds, compositions automatically
→ 4K resolution with perfect text adherence
→ Production-ready static ads for testing at scale

Built 100% in n8n.

Want the complete n8n template?

> Comment "NANO"

> Like this post

And I'll send it over (must be following so I can DM)

[Angry Tom @AngryTomtweets]

no words

[Demirdjian Twins @demirdjiantwins]

Nano Banana + Linah AI + n8n is legitimately insane 🤯

This workflow generates hundreds of UGC-style product-in-hand videos from a single product photo.

Fully automated inside n8n.

Perfect for DTC brands & agencies that need constant creative volume for paid social.

Here’s the bottleneck every brand hits:

You need 50+ ad variations to fight creative fatigue.
But coordinating creators, filming, and editing takes weeks and costs thousands.

This automation removes all of that:

→ Upload ONE product image into n8n
→ Linah AI generates 20–50 product-in-hand visual variations
→ Each variation is turned into UGC-style video scenes
→ AI auto-creates demos, testimonials, reactions & lifestyle shots
→ All outputs save directly to Box / Drive / Airtable for instant launch

No creator sourcing.
No shoot days.
No revision delays.

What you get:

→ Dozens of ready-to-run UGC ads from a single product photo
→ Costs pennies per video
→ Full commercial rights
→ Perfect for rapid A/B testing & scaling campaigns

Built 100% in n8n + Linah AI.

Want the exact workflow setup?

Comment “BANANA”
Like this post

And I’ll DM you the full automation (must be following).

[Daniel San @dani_avila7]

That blue node moving through the codebase?... That's Claude Code.

Found this video on youtube of the entire claude-code-templates development on GitHub - every commit, community PR, and file change from the git logs mapped out.

Seeing an AI agent navigate the dependency graph and ship features is wild 🤯

[Poonam Soni @CodeByPoonam]

It’s over. Wan 2.2 Animate with Nano banana is absolutely genius

FREE Tutorial + Nano Banana prompt👇

[myun wang @wang_myun]

Thanks to tariffs, a DJI Mavic 3 Pro costs $4,750 in the U.S.

In China? $2,118 after tax refund.

Same drone. $2,632 saved.

That’s 8,772 eggs. 🥚

With 15-day visa-free entry, you’ve got one more reason to visit China. 🇨🇳
#Tariffs #DJI #ChinaTravel #WhatToBuyInChina #Drone

[myun wang @wang_myun]

@elonmusk Elon, was it your idea to have the xAI engineers add a world model to Grok 3? I can't wait to see it!

[Kyle Kabasares @kylekabasares]

Following @hendrycks, I also gave the @OpenAI o1-pro model the 2024 Putnam Math Exam. Based on my (non-expert) understanding of the questions and the solutions, it appears that the o1-pro model scored somewhere in the range of 80-90/120, which based on past Putnam exams would put it in the top 1-2% of all participants, though we still await the results for this year's competition.

A Google Doc featuring links to my chats is here: https://t.co/uzcoT5prns

What impressed me even more was the time it took to produce its answers. Below is a breakdown of the "Think time" for o1-pro on each individual question:

A1 (3m 47s)
A2 (6m 25s)
A3 (6m 18s)
A4 (3m 41s)
A5 (9m 59s)
A6 (12m 1s)
B1 (2m 41s)
B2 (2m 31s)
B3 (6m 49s)
B4 (2m 13s)
B5 (4m 37s)
B6 (2m 19s)

In total, it spent 1 hour, 3 minutes and 21 seconds thinking on this exam compared to 6 hours (3 hours across 2 days) that students would normally get.

o1 pro's average time per problem was about 5 min, though there is some variance depending on the question.

As I write this, I'm still not entirely sure how to feel. On one hand, I find myself impressed, but on the other, I feel almost as if...I was expecting this? Based on my experience with o1-mini and o1-preview, this performance from o1-pro doesn't feel all that surprising... and it makes me wonder if I'm becoming more desensitized to these AI models' capabilities. It's as if with each model progression, I have unreasonable expectations.

Still, I have to admit that perhaps it won't be too long before those think times go down even more, maybe by a few more factors or even a whole order of magnitude, and its performance on Putnam 2025 and beyond will be undoubtedly No. 1 in the world compared to the thousands of math undergraduates.

Only time will tell, but I think we're in for some big surprises in 2025.

[myun wang @wang_myun]

I got into #Sora

[myun wang @wang_myun]

@AzmatKhanTDP @ChatGPTapp me too