Task 1: ChatGPT¶

Prompt: Give me some cool ML, computer vision and 3D Geometry puns and jokes

Answer:

Absolutely 😭 Here's a mix of nerdy, actually-clever, interview-room, and painfully bad ML/CV/3D geometry jokes.

🤖 ML / Deep Learning¶

  • Why did the neural network break up with its optimizer? It just wasn't converging anymore.
  • I told my model to stop overfitting. It said, "But I know you so well."
  • My model and I have a lot in common. We both perform better with regularization.
  • Why did the neural network go to therapy? Too many unresolved gradients.
  • I asked my model what it wanted to be when it grew up. It said, "A foundation model."
  • My loss function and I have a toxic relationship. No matter how much I minimize it, there's always something wrong.
  • Why was the gradient depressed? It felt like it was always descending.
  • What does an ML researcher say after a breakup? "It wasn't you. It was the local minimum."
  • Why don't neural networks ever get lost? They always follow the gradient.
  • My model doesn't have imposter syndrome. It has validation loss.

👁️ Computer Vision¶

  • Why did the image go to therapy? It had too many unresolved pixels.
  • I tried to explain computer vision to my parents. They said, "We see."
  • Computer vision researchers don't need glasses. They just increase the resolution.
  • Why did the CNN get promoted? It had excellent feature extraction skills.
  • What does a vision model say when it recognizes you? "I've seen you before."
  • My object detector is terrible at dating. It keeps generating false positives.
  • Why did the detector fail the interview? It couldn't distinguish signal from background.
  • I asked the segmentation model to draw the line. It gave me a mask.
  • Why did the optical flow researcher get arrested? They were caught tracking someone's movement.
  • My optical flow model is very emotional. It sees motion everywhere.

🧊 3D Geometry¶

  • Why don't 3D points ever argue? They always try to see things from another perspective.
  • I asked a point cloud how it was doing. It said, "I'm feeling a little sparse."
  • Why did the point cloud get rejected from the party? Too many outliers.
  • My point cloud and I have something in common: We both need better alignment.
  • Why was the 3D point so confident? It knew exactly where it stood.
  • Why did the point cloud go to therapy? It had too many unresolved correspondences.
  • What did one coordinate frame say to another? "You complete my transformation."
  • I met someone who works in SLAM. We got lost together.
  • Why did the camera break up with the world? There was too much projection between them.
  • 3D vision is basically dating. You spend most of your time trying to establish correspondences.

📐 Geometry / Linear Algebra¶

  • Why did the matrix break up with the vector? It needed more dimensions.
  • Why was the vector so attractive? It had great magnitude and direction.
  • I asked the rotation matrix for commitment. It said, "I can transform you, but I can't change you."
  • Why don't rotation matrices ever panic? They know how to stay orthogonal.
  • What did the homogeneous coordinate say? "I feel like there's more to me than meets the eye."
  • Why did the homography get promoted? It knew how to transform the situation.
  • Why did the affine transformation get hired? It was great at scaling relationships.
  • I tried to make a joke about eigenvectors... But it didn't have enough of an impact.
  • Why are eigenvectors so self-confident? They know their direction doesn't change.

📸 Camera Geometry¶

  • What did the camera say to the 3D point? "I can finally see your perspective."
  • Why did the camera fail its relationship? It had commitment issues with the extrinsics.
  • The camera and I have the same problem: We both struggle with depth.
  • Why did the intrinsic matrix go to therapy? It had some internal issues.
  • My camera has trust issues. It doesn't know whether you're in front of it or behind it.
  • What's a camera's favorite pickup line? "You look different from every viewpoint."
  • Why did the camera marry the coordinate frame? They had great extrinsic compatibility.

🧠 VLM / Multimodal¶

  • I asked a VLM to describe our relationship. It said, "Strong visual and semantic alignment."
  • Why did the image and text fall in love? They had good embedding compatibility.
  • My VLM is terrible at relationships. It keeps confusing semantic similarity with actual compatibility.
  • What did CLIP say on Tinder? "Let's see if our embeddings are close."
  • Why did the image refuse to date the text? It said, "We're just not in the same embedding space."

🔥 ML Researcher Humor¶

  • "How's your model doing?" "Training loss is down, validation loss is up, so basically I'm raising a child."
  • My PhD advisor asked me to think outside the box. So I used a point cloud.
  • I don't make mistakes. I make ablation studies.
  • My model achieved SOTA. Unfortunately, only on my laptop.
  • I spent six months improving the model by 0.3%. I call that a meaningful contribution.
  • What's the difference between a researcher and a magician? A magician says, "Watch this." A researcher says, "The results are in Table 3."
  • Why do ML researchers love reproducibility? Because otherwise it's called "a one-time optimization."
  • My model works perfectly. Please don't change the random seed.
  • "Can you explain your method?" "It's complicated." "Mathematically?" "Emotionally."

🥲 Especially good for a CV/ML person's bio¶

  • "I have a PhD in finding local minima."
  • "Currently looking for someone who appreciates me as much as my model appreciates a learning-rate scheduler."
  • "I see the world in 3D, but my camera still thinks I'm planar."
  • "Emotionally overfit, professionally underfitting."
  • "My type? Someone with good gradients."
  • "Looking for a stable relationship. Preferably one with low variance."
  • "I don't chase people. I track their optical flow."
  • "I believe in love at first sight—and multi-view geometry."
  • "Trust me, I'm a professional at minimizing loss."
  • "Dating me is like bundle adjustment: lots of iterations, occasional divergence, hopefully better reprojection error."

Task 2: 3D Shape Images¶

3D Shape

*Source: https://parametrichouse.com/curve-folding-polyhedra/*


Task 3: Funny GIFs¶

Funny GIF

*Source: http://galimateka.tumblr.com*