What I learned this week - 17/2023

This is a random collection of my learnings from the previous week. If you derive value from these, good. If not, sorry for wasting your time.

Artificial Intelligence

Things I should get comfortable with as I dive deeper into data-centric approaches for computer vision:

  • Self-supervised learning: intelligent data selection or error analysis will require a good, use case-specific feature extractor. ResNets pretrained on ImageNet likely won't cut it.
  • Unsupervised anomaly detection: techniques like PatchCore can output pixel-wise segmentation masks for likely anomalies. These could be useful for pre-labeling, saving SME efforts.

The AI hype bubble is real. Technologies get overhyped quickly on Twitter. Got a nice reality check reading this article about the shortcomings of AutoGPT and listening to Neil Patel’s take on ChatGPT. A lot of High-value skills remain for us humans.

There are three types of search queries:

  • Navigational queries: navigating to a specific website
  • Informational queries: searching for an answer
  • Transactional queries: searching for a product or service

Navigational queries won't be affected by chatbots like ChatGPT or Bing. Informational queries will primarily be handled by LLMs. For transactional queries this is unclear. While chatbots might remove friction in the transaction, consumers value choices for certain purchases. How this plays out will have huge marketing implications.


High-value skills which won't be automated soon:

  • Product sense
  • Storytelling
  • Sales & marketing
  • Leadership

Guidelines for good writing:

  • Use simple words
  • The shorter, the better
  • Focus on the important: what, why, how
  • Copywriting is about transferring emotion, not information

Point in case for the last bullet: my most popular LinkedIn post this week was about emotions.

Ship, ship, ship. Don’t over-optimize. Perfect is the enemy of good. I’d rather ship five good blog posts than one perfect one to build my audience.

The traditional way is to build a product, then find an audience for it. The better way is to build your audience first, then develop a product based on their needs.


40$ give a person in Africa access to clean water for 10+ years. Let that sink in. How often am I wasting this amount on stupid shit?

You can get a quick workout everywhere, even on the playground:

  • Lunge jumps: 3x10
  • Plyo push-ups: 3x10
  • Inverted rows (under a picknick table): 3x10

Done. Full-body workout complete in under 15 minutes.

Hitting the heavy bag works wonders. Not just for the body, also for the mind.

It's all about mastering the fundamentals, in all areas of life:

  • Relationships: praise, presence, honesty
  • Data science: problem framing, data transformation, visualization
  • Boxing: footwork, head movement, straight punches