AlphaFold - The Most Useful Thing AI Has Ever Done

To learn more about AlphaFold and its importance, I would recommend the Veritasium video[0]: AlphaFold - The Most Useful Thing AI Has Ever Done [0]: ▶ this video

Is that really true 120 years ago? We have a lot of hindsight bias, but things really weren’t amazing for the average person in the US even one hundred years ago. Today, the average person has affordable access to a lot of things that may have been luxuries, or simply didn’t exist, 100-200 years ago. Running water, electricity, internet, mobile phones, modern sanitation & infrastructure, and any number of tiny cheap devices that improve QoL. And yes, we also have huge problems with wealth disparity and late stage capitalism. The problem is, we should not throw out the baby with the bathwater. Like, we do not have widespread food scarcity problems today. Extremely advanced medical technology is available, if too expensive. Our tech bubble is only one small part of actual technology. There is still very cool research going on in other fields. Big tech is all about personal computing & social media, and yeah, AI in this area is dubious and hard to get excited about. At the same time, there are SO MANY other areas of technology which can have huge impacts on humanity. For example, the tech behind modern AI models has already produced breakthrough research in the field of protein folding. ( ▶ this video ) This will have a big impact on our ability to solve biological problems. All that to say… let’s stop being so closed-minded about tech. Yes, the big tech social media, app, and SaaS companies are not going to give us a utopian future. But technology is way more than that. It includes

> everything else: not much of anything achieving real, deep value. Lots of anecdotal use cases like drafting emails. Well, this is what people are missing IMO. Yeah, general purpose chatbots can’t solve advanced problems. However, models and pipelines (using similar technology as chabots) specifically built for certain tasks can achieve really incredible results — like in the area of protein folding ( ▶ this video ) I think there is a lot of potential, but most companies are just slapping a chatbot (which uses some other company’s models) on their app and calling it a day. Very few are taking a deep look at how ML could actually solve real problems.

Personal projects: - [GitMentor]( link ) → 300+ users so far - [NurtiLens]( link ) → 50+ users so far Blog post on GitMentor: link - Terminal in Notion: link Dropping Resume Optimizer: link Made a video building it from scratch in 30 mins—completely freestyle. Accidentally exposed some API keys, so had to rotate them since I don’t know how to edit videos :) Video: link Conclusion: You steer the LLM, don’t let the LLM steer you. - [LLM Bootcamp]( link ) → making content Another day, another drop. Not taking this one to production because PineconeDB costs too much for cosine similarity search. Built it anyway, here's the video: link Codebase: link YT Answer Finder goes to prod today: link Only two videos processed so far: - ▶ this video - link More projects: - Real-time flight tracker: link - Music recommendation app: link

> The first snakebite antivenom was made in the mid-1890s and the method has changed little since: snakes are “milked” for their venom, which is injected into horses or sheep and the antibodies that their immune systems then produce are extracted via the animals’ blood. This is expensive, and many people develop allergies to the antibodies. A cool practical application of DeepMind's AlphaFold is that it allows one to "design" proteins because it is now possible to predict how they will fold. [0] This means it is possible to create synthetic protein antibodies that neutralise snake venom, specifically designed for humans [1]. There is a recent Veritasium video [2] that contains a great explanation. The whole video is worth a watch! [0]: link [1]: link [2]: ▶ this video

A similar idea came up in Veritasium's latest video today. Training AI by DeepMind to predict protein folding greatly improved by withholding the most evident information about a protein's primary structure — its linear polypeptide chain — within the The Structure Module step. [0] After asking ChatGPT not to agree with me that your comment and these two different approaches to solving problems are the alike, it concluded there still might be similarities between the two. [0] ▶ this video [1] link