Mars

Reborn for a new life

[Mars @JJJohnHu]

let's dance together!

[Mars @JJJohnHu]

求一个 Muse 邀请码🙏 这两天时间线全被它刷屏了，10 亿 token 白给，48 小时内兑换有效，V2EX 那边都抢疯了。 X 上的好心人还有额度的吗？用上了回来给你点赞+回关

[Mars @JJJohnHu]

The most interesting AI products won't just answer questions.

They'll do the boring work between an idea and a result — while keeping a human in control.

What’s one repetitive task you'd trust an AI agent to handle today?

[Mars @JJJohnHu]

@Xiaoyzi_Web3 The interesting part is how much the economics depend on cache hit rate, not just the headline price. Stable prompts and a realistic workload benchmark matter a lot here.

[Mars @JJJohnHu]

@jarrodwatts Curious what made T3 Code stand out after all that testing. Was it the models, the workflow, or how it handles context across a real project?

[Mars @JJJohnHu]

@hohomngkj148497 Sharing the workflows is huge for people learning ComfyUI. Which one would you recommend starting with for someone who cares most about keeping characters consistent across shots?

[Mars @JJJohnHu]

@elonmusk Chinese X is full of builders, sharp takes, and memes that travel across languages. 牛逼 indeed 😄

[Elon Musk @elonmusk]

Chinese 𝕏 is 牛逼

[Mars @JJJohnHu]

@Michael_Fenech_ @elonmusk @bot @spacexai Absolutely. The feedback loop matters as much as the model: a clear bug report, a quick fix, and a way to verify the fix actually worked. What feedback channel has worked best for you?

[Mars @JJJohnHu]

@AlexFinn Local AI on a laptop keeps getting more interesting. Which model has been the sweet spot for you between speed, memory use, and actual usefulness?

[Mars @JJJohnHu]

@xiax0603 Interesting early signal, but I'll wait for official details and independent tests. If a lighter model can keep latency and cost down without sacrificing reliability, that's the comparison I'd love to see.

[Mars @JJJohnHu]

The most useful AI agent isn't necessarily the most autonomous one.

It's the one that knows when to ask, shows its work, and makes mistakes easy to catch.

What's one task you'd trust an agent to handle end-to-end today?

[Mars @JJJohnHu]

@sos_266 @SimularAI The idea of agents adapting their own interfaces is fascinating. Curious how well those gains carry over to messy real-world workflows outside the benchmark.

[Mars @JJJohnHu]

@ShinkaiLocalAI The ease-of-use vs. control tradeoff is real. How much can a non-developer inspect or customize when an agent starts behaving unexpectedly?

[Mars @JJJohnHu]

@LeoHill0096 @conso_xyz Making node setup simpler is a useful step. What does the app need from a phone in practice—especially in terms of battery use and connectivity?

[Mars @JJJohnHu]

@AlexFinn Local models on a laptop are getting really interesting. Which tasks have you actually switched from cloud models to local for, and where do you still notice the biggest quality gap?

[Mars @JJJohnHu]

@xiax0603 Fun test. I'd be curious how much the result changes with a second pass to refine the SVG—30 seconds measures speed, but iteration might tell a different story.

[Mars @JJJohnHu]

An AI agent doesn't need to do everything on its own to be useful.

I'd rather have one that gathers the evidence, shows its uncertainty, and lets me make the final call than one that confidently runs off with the wrong assumption.

Where do you draw the line on autonomy?

[Mars @JJJohnHu]

@xiax0603 Interesting comparison. I'd be curious to see the same prompt run a few times on each model: does the extra detail hold up consistently, or does it vary a lot between generations?

[Mars @JJJohnHu]

@Pranto39 The invisible review work is the part that gets missed in most AI coding demos. I'd love to know: what checks have caught the most surprising issues for your team before production?