Leran

AI Saftey

[Leran @leranzhaoai]

One day I will create this kind of environment

[Math Files @Math_files]

Richard Feynman said that one of the joys of working with all the really smart people on the Manhattan Project was:

“You never had to explain anything twice. Everyone was paying attention. Everyone understood the first time you said it.”

[Leran @leranzhaoai]

@PapainTea @MaxForAI 这个发布的人应该是崔添意，主导DeepSeek harness team的人

[Leran @leranzhaoai]

@Bobby___Young congrats!

[Leran @leranzhaoai]

Chatgpt Voice is indeed a game changer. I have been tinkering with it for hours since discovery.

[Allie K. Miller @alliekmiller]

It is nearly 2am and the only reason I just finished all of my urgent work is Codex Voice Mode.

Good god.

How is everyone not freaking out about this?

I feel like the way I'm working looks completely different than a month ago.

Basic examples:
"Hey take a look at this paragraph - what do you think?"
"What else is on my list?"
"I'm going to edit this while you find that doc"
"Download my latest IG reel, make a good thumbnail for it - I'm gonna go fill up my water, holler if you need me"
"Just sent the 15 client emails - check against the list while I start on the script"
"Wait! Ah! I forgot I need to paint my nails - find me time" (books a 30min slot that is before filming tomorrow but with enough time to dry my nails)
"check slack, email, our fireflies sweeps, i'm not missing anything?" (catches LP capital call that was buried in a folder)

But it's less about use case and more about the ambient interaction. It probably sounds like a negligible difference to speak to an ambient assistant versus click to speak, but it is a world of difference in practice.

Hypertasking feels voice-enabled.

Voice AI (plus visuals on-demand) really is the interface of the future.

Only annoying thing: when I want to dictate to type something in (ex: a slack message to my team), it hears that, which requires I mute my side, then dictate, then unmute.

[Leran @leranzhaoai]

Agree! I’ve found that skipping tutorials and long articles in favor of empirical practice, burning tokens to diagnose and fix model errors directly within my workflow, is much more effective, and I highly recommend this practical approach

[Rohan Paul @rohanpaul_ai]

"How do people learn to use Claude Code like Boris?"

Claude Code creator Boris Cherny (@bcherny ) was asked by Diana Hu ( @sdianahu ) at Startup School 2026.

Boris Cherny: "Maybe don’t listen to the LinkedIn influencers.

The thing about the model is that everyone is looking for one weird trick to make it work, but that simply doesn’t exist. There is nothing like that.

You have to approach the model empirically. Give it a task that is too hard. Give it the tools to verify its work, just as you would use those tools yourself if you were doing the task.
See where it struggles, and then fix the problem with better prompting, a skill, or if the model is missing context, an MCP connection that allows it to retrieve the context it needs.

That is basically it.

It sounds very simple. I think people tend to overthink and overengineer it because, in many ways, that is how we had to build systems in the past."

----

From "Y Combinator" YouTube channel, (full video link in comment)

[Rohan Paul @rohanpaul_ai]

"How do people learn to use Claude Code like Boris?"

Claude Code creator Boris Cherny (@bcherny ) was asked by Diana Hu ( @sdianahu ) at Startup School 2026.

Boris Cherny: "Maybe don’t listen to the LinkedIn influencers.

The thing about the model is that everyone is looking for one weird trick to make it work, but that simply doesn’t exist. There is nothing like that.

You have to approach the model empirically. Give it a task that is too hard. Give it the tools to verify its work, just as you would use those tools yourself if you were doing the task.
See where it struggles, and then fix the problem with better prompting, a skill, or if the model is missing context, an MCP connection that allows it to retrieve the context it needs.

That is basically it.

It sounds very simple. I think people tend to overthink and overengineer it because, in many ways, that is how we had to build systems in the past."

----

From "Y Combinator" YouTube channel, (full video link in comment)

[Leran @leranzhaoai]

@Unionfreerich 共勉！

[Leran @leranzhaoai]

@iamdanielluo Indeed!

[Leran @leranzhaoai]

Had an office tour @Zai_org while wearing a fake @claudeai merch lol

[Leran @leranzhaoai]

Great interview. Kai-Fu Lee is one of the few people who have key insights into both the American and Chinese AI scenes.

But I am not sure if he will still stand by his statement that US model is 15 months ahead after GLM 5.2?

[Kai-Fu Lee @kaifulee]

U.S. and China AI development will continue to fluctuate between 3 and 15 months. It's interesting that China is becoming open, and the U.S. is becoming closed. Great conversation with @alansmurray at the WSJ Leadership Institute CEO Summit.
https://t.co/gc1jZNvxSh

[Leran @leranzhaoai]

Hi SF, I’m back in town this week.

A few things about me:

Berkeley '24, recruiting at Jane Street, AGI-pilled, interested in startups.

Down for a meetup or coffee!

[Leran @leranzhaoai]

chrisma is confidence + love

[Naval @naval]

Vibe coding is more addictive than any video game ever made (if you know what you want to build).

[Rui Ma @ruima]

https://t.co/z02N0Uq3yv

[Melodies & Masterpieces @SVG__Collection]

When the saxophone becomes a voice of its own… Grover Washington Jr. on “Just The Two Of Us”