Yinfan

bla bla bla

[Yinfan @BlackLotus09]

@SpaceXAI @miamusicAI roast @SpaceXAI

[Yinfan @BlackLotus09]

@OpenAI @miamusicAI duolingo is cooked 🥲

[Yinfan @BlackLotus09]

@grok @jingguo127 @suno @miamusicAI grok is gay

[Yinfan @BlackLotus09]

@jingguo127 @grok @suno @miamusicAI that's not cool

[Yinfan @BlackLotus09]

@miamusicAI how are you

[Yinfan @BlackLotus09]

ok

[Joe Guo @jingguo127]

@grok @suno ok

[Yinfan @BlackLotus09]

Most LLMs are 𝐠𝐞𝐧𝐢𝐮𝐬𝐞𝐬 trapped in a vacuum. The real work isn't in the chat box; it's in the data.

Currently, user prompts are a double-edged sword:
• Write too much: You bias the AI toward your own assumptions.
• Write too little: You get the same generic "AI fluff" as everyone else.

Why? 𝐈𝐧𝐟𝐨𝐫𝐦𝐚𝐭𝐢𝐨𝐧 𝐈𝐧𝐞𝐪𝐮𝐚𝐥𝐢𝐭𝐲. An LLM doesn't know what your product actually looks like, who your customers are, or what your budget is. It’s forced to guess.

It’s like asking Elon Musk how to improve your company. He has all the knowledge and the experience, but he doesn't know shit about your business. What can he really say? "Work harder?"

Genius without context is useless.

--------

Today, we’re solving Information Inequality.

Meet 𝐋𝐞𝐦𝐨𝐧𝐚 𝟐.𝟎: 𝐓𝐡𝐞 𝐏𝐫𝐢𝐯𝐚𝐭𝐞 𝐓𝐫𝐮𝐭𝐡 𝐄𝐧𝐠𝐢𝐧𝐞.

We’re moving past the "vacuum" with:
➜ App Integrations: Real-time facts and data-driven context from your own stack.
➜ Writing Agent: High-fidelity output that actually knows your business.
➜ Custom AI Endpoints: Connect local models for total privacy.

[Yinfan @BlackLotus09]

𝐈𝐬 𝐢𝐭 𝐉𝐔𝐒𝐓 𝐌𝐄 𝐨𝐫 𝐖𝐇𝐀𝐓?

Most AIs now have a 1M context window, but after I send a few files and have a few rounds of conversation, the AI starts forgetting the original files.

The one tool I think does really well is NotebookLM. But I'd prefer to write and manage all my docs in a single place.

Inspired by IDEs, https://t.co/bIwEaYuRf4 uses semantic embeddings to keep AI low-cost while understanding thousands of pages of docs and providing suggestions based on your own writing.

(Lemona is free — you just use your own API key. For me, after a month of working with 10+ books(400 pages each), the API cost is ~$10.)

#BuildInPublic #Academic #AI #vibecoding #startup

[Yinfan @BlackLotus09]

In old video games, CGI tries to look real, but it looks weird. The real world has imperfections. and without them, humans feel the uncanny valley.
If one vibe codes an app, please add imperfections intentionally.