thou that art

Stop Making AI Read Faster. Make It Read Smarter.

Manu — Tue, 10 Feb 2026 17:50:31 GMT

Here’s a number that should bother you: 0%.

That’s the score GPT-5 the most powerful language model on the planet gets when you hand it 10 million tokens of text and ask it to find something specific. Zero. Not “pretty bad.” Not “room for improvement.” Zero.

Now here’s another number: 91.33%. That’s what the exact same model scores on the exact same task …

Alice's Brother Has Five Sisters. AI Says Four.

Manu — Mon, 09 Feb 2026 18:08:45 GMT

Here's a question my six-year-old nephew nailed on the first try:

The answer is five. Alice herself is a sister to her brother. A child sees this instantly. GPT-4—the model trusted with legal analysis, medical triage, and billion-dollar code reviews—gets it wrong.

This isn’t a cherry-picked gotcha. It’s one specimen in a vast, now-cataloged ecosystem of r…

Piketty’s Revenge: Why the “AI God” Just Validated the World’s Most Hated Economist

Manu — Tue, 03 Feb 2026 21:54:01 GMT

In 2013, Thomas Piketty released a 700-page brick called Capital in the Twenty-First Century. The “smart guys” in the room laughed. They said his math was broken. They said he didn’t understand that hammers need hands—that as capital (hammers) grows, labor (hands) becomes more valuable.

For a hundred years, the critics were right. Labor and capital were …

The "Bluebook" Prophecy: Why the Future is Analog (Whether You Like It Or Not)

Manu — Tue, 03 Feb 2026 15:11:12 GMT

From roughly 2015 to 2025, society collectively agreed to a lie: We pretended that digital was the same as real.

We moved our classrooms to Zoom. We moved our hiring to LeetCode. We moved our dating to Hinge. We moved our trust to the cloud. We thought the physical world was just a legacy system—a deprecated codebase waiting to be overwritten.

I have bad …

Use AI Like a Spotter, Not a Forklift

Manu — Fri, 30 Jan 2026 18:22:24 GMT

I shipped a fix in 18 minutes last week.

It looked clean. Tests passed. PR merged. I felt that familiar little rush: “nice, I’m fast.”

Two days later it broke in production, and I realized something worse than the bug.

I could not explain why the code worked in the first place.

The AI did not just write code for me. It quietly removed the part where I would…

Think in Diffusion, Talk in Autoregression: Why TiDAR Actually Matters

Manu — Wed, 26 Nov 2025 16:06:38 GMT

Autoregressive decoding has owned high-quality language generation for years. It also completely disrespects how GPUs actually work.

One decoding step emits exactly one token, but it drags the entire model’s weights and the full KV cache through memory. You pay the cost of a huge forward pass to say a single word. On modern GPUs like the H100, that means…

The Fractured Brain!!

Manu — Wed, 26 Nov 2025 03:16:22 GMT

I spend a lot of time working with frontier models. I build systems around them, fight with their failures, and watch them trip over tasks a beginner should handle. The patterns become obvious once you stop getting hypnotized by benchmark scores.

The models are powerful, but they are fragile. They feel intelligent until you ask them to operate like somet…