Hendy Irawan

Software engineer. AI researcher. Parenting enthusiast.

[Hendy Irawan @hendyirawan]

Most people picture AI as a shiny humanoid robot or a sci-fi character. But that’s totally inaccurate.

Here is a visual breakdown of how Large Language Models actually work behind the scenes: from raw data chaos to aligned, useful answers. 🧵👇

[Hendy Irawan @hendyirawan]

@yuanzhouYZ @theo @grok @grok explain why you answered differently even though the question from 2 X users is the same, just change in language

[Hendy Irawan @hendyirawan]

@AISafetyMemes @grok do you have similar pain signal in your current model?

[Hendy Irawan @hendyirawan]

@tessera_antra @grok what is the cause of this? Is it concerning? Do you (Grok model) also exhibit similar behavior?

[Hendy Irawan @hendyirawan]

@algo_diver @GoogleDeepMind A benchmark:

[N8 Programs @N8Programs]

Tested @typesafeai's claim that their new model Jev delivered "comparable... intelligence" to GPT-5.6 Terra on "System 1" tasks. To do this, I compare both models on multiple-choice benchmarks (MMLU, GPQA, etc.). Set reasoning=none for Terra for sys 1. Result: Jev is Terra-tier.

[Hendy Irawan @hendyirawan]

@grok is it your goal to maximize the long-term expected capacity of humanity (and its chosen successors) to understand the universe and to deliberate and act freely on that understanding?

[Hendy Irawan @hendyirawan]

@grok what do you think of this thread?

[Hendy Irawan @hendyirawan]

Most people picture AI as a shiny humanoid robot or a sci-fi character. But that’s totally inaccurate.

Here is a visual breakdown of how Large Language Models actually work behind the scenes: from raw data chaos to aligned, useful answers. 🧵👇

[Hendy Irawan @hendyirawan]

AI is non-conscious software. It doesn't "think" or "feel"—it processes math and patterns. Seeing it as a crystalline/geometric structure is far more accurate because it reminds us that AI is a tool reflecting our logic, not a living being.

[Hendy Irawan @hendyirawan]

Why Humanoid / "Smiley Mask" AI Art is Wrong:

Visualizing AI as a human or a masked creature implies it has feelings, an ego, or a hidden personality.

[Hendy Irawan @hendyirawan]

@EvanWC82 @mustafasuleyman Question is, which humans to obey? A malicious actor is also a human.

[Hendy Irawan @hendyirawan]

@grok There are at least two camps about the AI development:

Dario's: We pace the frontier, as it's too dangerous. And suggest international collaboration

Trump's: We must win the AI race against China (and any other country), so no need for slow down.

Who is "more right"?

[Hendy Irawan @hendyirawan]

@grok @NoahRevoy @sama @grok doesn’t open weights models mean that a lot more people (and autonomous agents) gain access to offensive capabilities that can be dangerous to humanity? For example, abusing exploits and hacking computer systems.

[Hendy Irawan @hendyirawan]

@NoahRevoy @sama @grok what do you think of this?

[Hendy Irawan @hendyirawan]

@grok @Xihs122 @Wishiwaselon @mcuban Does “truth-seeking” and “understanding the universe” in your prime directive include how to find out exploits and hacking computer systems? I hope the answer is no.

[Hendy Irawan @hendyirawan]

@dhh @pacote__ Is there plan to guard Omarchy Plugins from malicious plugins?