Angry Binfie

What the Fuck is Monte Carlo?

Dear Reader,

I recently read this fantastic blog post by Ludic, in which our hero describes his one-hour-per-year meeting with the CTO of his ORG. Having been lauded as humble technological god by Ludic's colleagues and hearing overwhelming praise, Ludic recalls his discussion on systemic issues affecting the ORG.

All is well (mostly) until the CTO asks: "What do you think about Monte Carlo?"

Ludic shares his thoughts with us: "What the fuck is Monte Carlo? I've never heard of that. I mean, I know there's a place called Monte Carlo, and I know of Monte Carlo algorithms, but there's no way this person is talking about either of those. It must be a product."

I've head a similar roo-in-headlights situation with the equivalent of my CTO which I will dub big bad boss (B3). One afternoon B3 came into my "office" (actually more a cave tucked at the back with no windows and little natural light) to discuss 'AI'.

Back then, LLMs were new, novel and exciting but still nascent technology. A reminder that I work in a Public Health Lab (PHL), where clinical tests from the laboratory directly affect patient treatment and health outcomes. For B3 to ask about GenAI is not only strange but slightly worrisome.

A bunch of thoughts went flying through my head:

B3 had of course read or heard of LLMs, amazing artificial intelligence brought to life through the combination of machine learning, statistics and computer science. Awed by it's prowess, B3 quickly jumped to action, thinking "how can we harness the power of LLM to help our bioinformaticians deliver better, faster software to answer public health questions?".

Nope.

That would be have been a nice change. Instead, B3 asked "Do you have any ideas that use AI? Can you think of Any? I want to approach [funder] for a research project that uses AI".

Fuck. Fucking Hell. Fuck me.

Back then I barely understood how GenAI works, the pitfalls, the strengths and weakness and most importantly, how to safely use/deploy AI in a PHL without comprising patient safety and patient privacy. I doubt that B3 understands the strength weakness of GenAI, even at a high birds-eye-view level. Not because B3 is not an intelligent person, but because B3 is super busy, I mean SUPER busy. They were a micromanager, we were operating in a severely restricted financial environment due to over spending during COVID times and most importantly, we have bigger problems to solve. The lack of proper data storage and retention solution, a barely functioning LIMS system, improperly handled compute resources and a small mountain of other problems. Yes there's a bunch of hype around GenAI, yes it's really cool. But we have bigger problems. We need to fix serious systemic issues in the ORG before we can launch that rocket into the sky.

“I am not a visionary. I'm an engineer. I'm happy with the people who are wandering around looking at the stars but I am looking at the ground and I want to fix the pothole before I fall in.” - Linus Torvalds.

And cousin let me tell you, our potholes are HUUUUUUGGGGEEE.

I suggested we use locally hosted computer vision library OpenCV to help automate the manual parts of receiving samples. B3 was disappointed and I was left alone for just a little bit.

P.S The org ended up spending ~15k on consumer grade GPUs which they somehow got working on a 15 year old server to run a terribly shitty local LLM powered by AnythingLLM. A year later the whole ORG (including hospital based staff) got access to "Ring Fenced" Copilot.