Thinking in Systems

I struggle to give a good summary of this book. My take is how to see systems (a very generic word) in the big picture as most systems are too complex to understand how fully work (economy, stock market, etc). In general, once we see the relationship between structure and behaviour, we can start to understand the system and modify it. You can’t know a system just by its parts. Look for the interconnections of those parts.

A system is formed by “stock” (water in a reservoir, mineral deposits, etc) and the stock changes overtime due to the actions of “flows” (rain = inflow, evaporation = outflow, mining = outflow, etc) Inflow increases the stock. Outflow decreases the stock. If the rate of inflow and outflow is identical, you have a system in a state of dynamic equilibrium. You want to see the systems behaviour based on time. Generally, stocks change slowly compared with the rate of change of in flows. So stocks act like a “buffer” in systems. A feedback loop is formed when changes in a stock affect the flows into or out of that same stock. You have two types of feedback loops: balancing (seek stability and resistance to change) and amplifying/reinforcing (can cause healthy growth or destruction) (ie: learning piano, the more I practice, the more I learn, the more keep practicing and so on). Doubling time = 70/growth rate (It takes 14 years to double your money in a back at a rate of 5%) In real systems, a single stock can be influenced by several types of feedback loops (with different directions and strengths)

The information delivered by a feedback loop can only affect future behaviour (can’t have an impact fast enough to correct the behaviour triggered the current feedback). And there will always be delays in responding. As well, because systems often have competing feedback loops working at the same time, the loop that dominates the system will determine the behaviour. You can have shifting dominance of feedback loops (dead rate vs birth rate)

System dynamics models explore possible futures and ask “what if” questions. Testing the value of a model: 1) Are the driving factors likely to unfold this way? 2) If they did, would the system react this way? 3) What is driving the driving factors?

Dynamic systems studies are designed to explore what would happen if a number of driving factors unfold in a range of different ways.

Systems largely cause their own behaviour. Systems with similar feedback structures produce similar dynamic behaviours, even if the outward appearance is not similar (population vs industrial economy, coffee cup cooling vs radioactivity decay)

A delay in a balancing feedback loop makes a system likely to oscillate (ie: response of orders and deliveries in a car dealer). Delays are pervasive and are strong determinants of behaviour. Changing the length of a delay may (or nor) make a large change in the behaviour of a system.

Examples of two-stock systems:

  • A renewable stock (capital) constrained by a non-renewable stock (oil): oil company: Non-renewable resources are stock-limited. The entire stock (oil) is available at once and can be extracted at any rate (limited by extraction capital). The faster the extraction rate, the shorted the lifetime of the resource.
  • A renewable stock (capital) constrained by a renewable stock (fish): fishing company: Renewable resources are flow-limited. They can support extraction indefinitely but only at a finite flow rate equal to the regeneration rate;

No-physical system can grow forever in a finite environment.

Scale Systems 2024, MS GenAI for beginners, Federer, Whisper WebGPU, Starlink TCP

Scale Systems 2024 (videos): GenAI Training: Short but interesting video. Main failures: GPU, memory and network cables 馃檪 For the Network side, I liked this screenshot. Still they are able to build two 24k GPU cluster with IB and RoCE2.

MS: GenAI for beginners.

Federer: Effortless is a myth (without hard work there is nothing), It’s only a point (resilience, present), life is bigger than the court.

Whisper WebGPU: Real-time in-browser speech recognition

Free Matrix Multiplication: This looks like a big deal.

Starlink TCP: Very quick summary, control protocols with Selective Ack perform better. The ping analysis is quite good. Being able to see that each 15s you are changing satellite, is cool.

Intro LLM, LLM bootcamp, Computex 2024, UALink, Aurora, Arista AI Center, Kubenet, Nutrigenomia, Videos

Intro LLM

LLM Bootcamp 2023:

NVIDIA Computex 2024: It seems they are going to yearly cadence for networking kit. They showed plans for 2025 and 2026… I liked the picture of a NVLink spine and the huge heatsinks for B200….

UALink: The competition for NVLink. This is for GPU-to-GPU communication. UltraEthernet is for connecting pods.

Aurora supercomputer: Exascale broken. Based on HPE slingshot interconnect (nearly 85k endpoints) Everything else is Intel.

Arista AI Center: it seems they are going to team-up with NVIDIA. Some EOS running on the nics.

Kubenet: Seems interesting but only supporting Nokia SRLinux at the moment.

Nutrigenomia:

“Lo que hicimos fue un trabajo personalizado en el que cuidamos todos los aspectos de la nutrici贸n y buscamos la regeneraci贸n y la correcta expresi贸n de sus genes.”

fisiogen贸mica: Yo lo llamo as铆 porque mezcla fisioterapia, nutrici贸n y nutrigen贸mica. En cada persona tenemos que buscar por s铆ntomas, an谩lisis e intervenciones qu茅 alimentos limitar por producir una mala expresi贸n gen茅tica, pero todas las pautas est谩n basadas en la Pir谩mide de la Dieta Mediterr谩nea”

Videos:

Bear Grylls: Be kind, never give up.

Born To Run

I can’t run for the last 4 months so reading this book has been a bit annoying… but increases my desire to get to it.

To be honest I didnt have a clue about the book apart of running. It started to get hooked slowly and at the end, I was eager to know if the race was going to happen, who was going to race and how was going to finish.

The center of the book is about the Tarahumara and their tradition of long distance running with basic kit (sandals) and frugal diet (mainly based on corn and beans). Thinking coldly, all looks a bit too romantic but it is a hard life.

Things I learned:

Tarahumara consume a lot fo Chia seeds. It seems it easy to grow (other) but I think I would need a big space to produce enough quantity for one year consumption?

Benefits of barefoot running (Daniel Liberman) and it seems that endurance running was the key difference with Neanderthals when the ice age ended and things got warmer and it was the only way to hunt in the savanna: outlasting your prey. Arthur Lydiard is the father of modern running training. Supports barefoot running. It interesting the data showing the increase of injuries with the advance of running shoes technologies… And the history about Nike and Lydiard and Bowerman (his mentor). Still getting to that level you need to make a slow transition. Need to research about this.

The crazy stories about Jenn Shelton and Billy. Party ultrahard and then ultrarun: epic.

Scott Jurek diet is vegan: vegetables, fruits, grains and legumes. recipes.

Caballo Blanco died at 58 running.