Felix Kang 🔻

Founding AI RAAS for marketers & e-commerce business. Distributed team across US & Asia.

[Felix Kang 🔻 @hereFelixK]

Stop trying to look "Professional."
It is killing the startup brand.

"Perfect" = "Fake." Humans have evolved a new biological filter: "The Anti-AI Defense Mechanism"

Instead, we need to master "The Human Glitch."

0/ Here is the 5-step playbook to weaponize imperfections:

The "Mint Mobile" Effect (Weaponized Transparency)

The "dbrand" Effect (Hostility over Politeness)

The "Nike" Rule (Strategic Suicide)

The Anti-Persona (Who to exclude)

The Open Kitchen Policy

https://t.co/eJ4PSeKIQY

[Felix Kang 🔻 @hereFelixK]

This is a pricy lesson.
Infinite Claude for everyone is not a good idea in 2026.

[Warren H. Cohn @WarCo1]

Here’s the email to employees:

Team,

Hope everyone’s doing well and enjoying the productivity enhancements from our AI tooling initiative.

Unfortunately, Finance has asked me to clarify a small issue.

It appears someone, and by “someone” I mean apparently all of you simultaneously, managed to spend $500,000,000 on @claudeai usage in a single month.

For context:
•NASA landed on the moon for less.
•We are now the proud owner of approximately 14% of Anthropic.
•Claude personally sent us a thank-you fruit basket.
•Our CFO has entered a fugue state and only communicates through Slack emojis.
•The electricity usage from your prompts briefly dimmed parts of Northern Virginia.
While we appreciate innovation, there are concerns that:
•“Can you make this email sound slightly warmer?” did not require 11,400 generations.
•Asking Claude to “rewrite this in the style of Succession, Hemingway, and Tony Soprano combined” may have been excessive.
•One employee appears to have used Claude to generate “a quick list of lunch options” that somehow consumed the GDP of a small island nation.

Going forward, please observe the following guidelines:
1Do not upload the entire internet into Claude “for context.”
2If your prompt begins with “simulate every possible outcome,” reconsider.
3Claude should not be used to:
◦settle fantasy football disputes,
◦write your wedding vows 97 times,
◦generate revenge edits of your ex’s LinkedIn bio,
◦or ask “what if Rome had WiFi?”

Most importantly:

If you see the message:

“This request may require additional datacenter construction”

…please stop immediately and contact IT.

Thank you all for your cooperation during this challenging yet technologically groundbreaking time.

Warm regards,
Management

P.S. Whoever prompted:
“Generate every possible PR angle for every company founded since 1983”
…we just want to talk.

[sudox @kmcnam1]

jfc....

[Felix Kang 🔻 @hereFelixK]

everyday work like

[Zara Zhang @zarazhangrui]

Frontend Slides skill now has 19k stars on GitHub ✨

I upgraded it with a new design brain: it can now pull from my Beautiful HTML Templates library, pick a visual direction, and generate slides using that template’s design language

Also new:
- Works better with coding agents beyond Claude Code - Export to webpage or PDF for easy sharing
- Inline slide editing
- Fixed 16:9 deck stage to avoid overflow, cropping, and weird responsive chaos

Try now: https://t.co/CiQpficxcD

[Andy @PositivFuturist]

Why Object Oriented Programming sucks.

[Felix Kang 🔻 @hereFelixK]

https://t.co/NqOFU1RhoN

[Felix Kang 🔻 @hereFelixK]

I cloned a female voice using Qwen3 model for my podcast.

With a 24.7s reference audio. Long segments sounded right. But on short replies, the voice "drifted" to a male voice.

Why:
The reference audio gets encoded into audio tokens and prepended to the model input as a "zeroprompt" prefix. With a 24.7s reference (~1500+ tokens) and only 2 seconds of target output (tens of tokens), the prefix dominates attention. The model can't sustain timbre lock for such a short generation window and falls back to its default speaker prior — which, in a Chinese pretrain corpus, is more often male than female.

Switched the reference to a shorter 16s clip from the same speaker, drift fully gone.

Takeaway:
In voice cloning models that encode the reference as a prompt prefix, reference length should roughly match the target length scale.

A "richer" longer reference isn't always better — for short conversational utterances, it actively hurts.

https://t.co/ZmAvp3wJHT

[Felix Kang 🔻 @hereFelixK]

Very interesting. MiniMax audio do not give you a clear link to cancel your subscription.

it took me 10 minutes, and i don't know where to cancel it.

@MiniMax_AI teach me how to do it.

[Felix Kang 🔻 @hereFelixK]

What will be, will be
Stay with that mindset makes u free

[vittorio @IterIntellectus]

it’s funny because this is genuinely the best approach

Simply not thinking about stressful situations makes you less stressed

[Felix Kang 🔻 @hereFelixK]

@BTCxiaoyu1 @NFTCPS 等你更新效果review

[Felix Kang 🔻 @hereFelixK]

This is the best part of this podcast:
"working on AI DO NOT NEED brain, it is more about being responsible"

Yao has served as a research scientist at both #Anthropic and #Google #DeepMind, contributing to the development of key models including Claude 3.7, 4.5, and Gemini 3.

[Felix Kang 🔻 @hereFelixK]

Claude + Youtube Claude live

= vibe coding package

[Felix Kang 🔻 @hereFelixK]

I am struggled in hiring
The rarest talent in tech right now:

1⃣ Engineers with real product taste
2⃣PMs who can use AI to ship, not just write PRDs

Doesn’t matter if you’re in SF, London, China, or Singapore.

[Felix Kang 🔻 @hereFelixK]

@1knflnba @CodeWithZeee @TheGeorgePu API is way too expensive 😅

[Felix Kang 🔻 @hereFelixK]

@CodeWithZeee @TheGeorgePu it is the first plan they will choose to try claude code.

After realizing pro plan can't even finish a small project. They will have to upgrade

[Om Patel @om_patel5]

stop spending money on Claude Code.

Chipotle's support bot is free:

[Felix Kang 🔻 @hereFelixK]

I will ask why the heck it so long to process my task.

[Thariq @trq212]

We just added /btw to Claude Code!

Use it to have side chain conversations while Claude is working.

[Felix Kang 🔻 @hereFelixK]

@trq212 they just replaced @typelessdotcom

[Elon Musk @elonmusk]

You can now select topics for your timeline, so if you’re tired of political ragebait, you can select something else 😂

Just tap on For You.

[Felix Kang 🔻 @hereFelixK]

这只是长期博弈的开始

现实是：有小公司，有开源模型，有太多方式可以绕过去。

技术挡不住，最后的底线还是在人

让跨线的人付出不可承受的代价，不只是丢合同，是整个行业孤立你，整个社会不跟你玩

说白了跟国际关系一个逻辑。大国之间为什么相对稳定？不是大家都善良，是核威慑——动了就完了

AI 治理最后也得靠这种"威慑"。不是靠一家公司的良心，是靠共识：有些事，做了就没有回头路