Shi Dong

A dull person, living a bland life, interested in boring things.

[Shi Dong @ShiDong14]

Finished Ishiguro’s Never Let Me Go today, and the novel kept bringing me back to what Ed said in the interview. Creativity, the faculty to ask novel questions and meaningfully break existing conventions, is what distinguishes us from the soulless. Such an invaluable human asset is arguably also the hardest thing to teach an AI.

Both the novel and the interview are strongly recommended. Both are dense: the novel emotionally, the interview intellectually.

[Edward Hughes @edwardfhughes]

Tomorrow’s greatest scientific discoveries will emerge from collaboration between creative humans and creative machines.

Done safely and responsibly, humanity will benefit from magical new inventions. Here’s how I think we get there.

https://t.co/FHLo7H0fr8

[Shi Dong @ShiDong14]

Robert Frost knew all along: the beauty of RL.

"The woods are lovely, dark and deep,
But I have promises to keep,
And Miles to go before I sleep,
And Miles to go before I sleep."

[RadixArk @radixark]

Today we're launching Miles v0.1, an open-source RL framework for LLMs and multimodal models.

RL training is easy to start and hard to debug. Miles helps you ensure your run is correct, use hardware efficiently, and keep RL running at scale.

Over the past 9 months, 72 contributors have landed 1,326 commits, 85 GPU E2E CI tests, battle-testing Miles on frontier open models like Kimi K3, DeepSeek V4, Qwen 3.8, GLM 5.2, Inkling, MiniMax H3, etc.

Miles powers frontier-model development and production RL workloads at @humansand, @periodiclabs, @modal, @DecagonAI, @Eigent_AI, @nebiusai, @IBM and more, on both @NVIDIAAI and @AIatAMD hardware.

Here is what we built, and why teams picked Miles🧵

[Shi Dong @ShiDong14]

I wonder how Baudrillard would respond to a world in which, on top of mass media, there is an extra layer of AI that assimilates and disseminates all information. AI will create a simulacrum for the hyperreality. An “ultrareality”.

[Edward Hughes @edwardfhughes]

Proud to announce the launch of @inherent_labs. We’re reinventing the scientific research factory for the age of AI agents.

I’m joined by co-founders @kallyaleksiev, @LouisKirschAI and @TantumSCollins; all are deeply technical operators.

Time to live within the experiment.

[Shi Dong @ShiDong14]

@Yuhu_ai_ Sorry to see you go 😥 Wish you all the best!

[Shi Dong @ShiDong14]

@simon_zhai Will surely miss you!

[Shi Dong @ShiDong14]

@FrankYueWu1 Will miss you for sure 😢

[Shi Dong @ShiDong14]

@larrabearr I like what you said, but I’m also curious about the particular circumstance that motivated this tweet 😂 Is it something at work?

[Shi Dong @ShiDong14]

@veggie_eric Love in the Time of Cholera. Timeless classics.

[Shi Dong @ShiDong14]

A wild take on Rich Sutton's recent interview.
For centuries we take for granted that humans are more intelligent than all other animals. Could it be possible that all animals (or at least vertebrates) are equally intelligent, but only humans are driven by the goal to understand the universe?
We could be essentially just the same as... ducks, except with higher ambitions, more hubris, and less nonchalance (which is not a bad thing since these qualities took us to the moon and gave us semiconductors).

[Shi Dong @ShiDong14]

A few months ago, I thought of the following brain twister for LLMs:

Design an MNIST classifier whose accuracy is as close to 67% as possible.

Both DeepSeek-R1 and o3-mini didn't find the optimal solution (though R1 got really close in CoT).