The Underfitted

jr AI Engineer | Transformers × Financial Data | CS Student

[The Underfitted @SimplySigmoid]

"The model will converge anyway" - a compelling argument, and a costly misconception.

Two models. Same everything.

Model A, step 0: loss = 3.29 → starts learning immediately.

Model B, step 0: loss = 27 → spends the first 9,000 steps just getting back to where A started.

Model B didn't train for 10k steps. It trained for ~800.

The rest was debt repayment.

A bad initialization doesn't slow you down. It steals your training budget — silently, one "optimization" step at a time.
🧵

[The Underfitted @SimplySigmoid]

@CodeEdison if it were that simple

[The Underfitted @SimplySigmoid]

@romerojr__ @claudeai Appreciate!

[The Underfitted @SimplySigmoid]

@tetsuoai king

[The Underfitted @SimplySigmoid]

@vivoplt Also dont forget about this gem

[The Underfitted @SimplySigmoid]

@BTechWalaBanda @karthikponna19 help with?

[The Underfitted @SimplySigmoid]

@RoundtableSpace

[The Underfitted @SimplySigmoid]

@vivoplt having a good foundation is a true leverage in this AI era

[The Underfitted @SimplySigmoid]

It's 6PM on a Saturday.

Karpathy on screen. Handwritten notes on the desk. VS Code open with makemore_from_scratch.

No tutorial. No shortcut. Just activation functions, neuron flow through layers, and the slow satisfaction of actually understanding what's happening inside the network.

Week by week.
Layer by layer.

[The Underfitted @SimplySigmoid]

@Shivam25mishra you need solid foundation to build your house

[The Underfitted @SimplySigmoid]

@Shruti_0810 dont forget about this GEM

[The Underfitted @SimplySigmoid]

@VadimStrizheus thats what a single prompt could do

[The Underfitted @SimplySigmoid]

@quantscience_ if would be that simple

[The Underfitted @SimplySigmoid]

@anthdm quants era incoming

[The Underfitted @SimplySigmoid]

@elonmusk better than my automated system

[The Underfitted @SimplySigmoid]

@bigaiguy a comprehensive transformer breakdown by Claude

[The Underfitted @SimplySigmoid]

@nalinrajput23 math

[The Underfitted @SimplySigmoid]

It's 6PM on a Saturday.

Karpathy on screen. Handwritten notes on the desk. VS Code open with makemore_from_scratch.

No tutorial. No shortcut. Just activation functions, neuron flow through layers, and the slow satisfaction of actually understanding what's happening inside the network.

Week by week.
Layer by layer.