The last few months have been busy, as I’ve been organising my time across multiple projects. Conveniently for me, I’m working with collaborators across time zones (IST, GMT and US) and so I’ve settled into a rhythm in which I can time-box my meetings and have time to read and think and code as well.
I’ve come across some excellent reading and listening recently. In no particular order (mainly, how quickly I was able to find this across Kindle and my Chrome history):
An excellent article by Pararth Shah on the topic du jour: Agents!
How diffusion models (e.g. Stable Diffusion GenAI) work, by Andrew Chan (link): This is a very technical post, and I’m not going to pretend I actually followed all the math. BUT, it felt good to try, and the he explains well, and I came away feeling better informed and my grey cells got a good workout
The LLAMA 3 paper: I read it only because of the hype. There are very good summaries of the takeaways available online, and it would have worked just as fine. My main takeaway was that spending a lot of time on quality of data and eval is leading to significant model improvements, and there is still a lot that researchers are trying to understand about how this all work. It strengthens my thesis that there is a lot of value to be created by building very specific applications on top of LLMs right now, since the ability to be stable and ‘guaranteed good’ is much more important to companies vs. ‘well sometimes it’ll do really cool stuff but sometimes it’ll upset y our customer’.
Scott Belsky’s posts, which I’ve subscribed to now:
. The one in question was this oneThe Prompt Report (paper) and easier way to consume some takeaways, the Latent Space podcast (which is how I came across it)
How to succeed at Mr. Beast (link): Apparently a leaked document from Mr. Beast Productions (Youtube’s biggest star, and quite a phenomenon in many ways). This one is interesting - a breezy read, several things that made me nod along vigorously and several that I was like “naah, pass” i.e. works for that company, but I wouldn’t want to be in that environment. A reminder that success looks different based on context, and generic advice is not very useful
An intro to Gaussian splats (link), a fascinating topic and I’m glad the experts wrote this tagline too :)
That’s it for now! I would love any feedback on if you end up reading any of these, and also how useful it is to have me echo selected items from my reading back into blogosphere. I have been toying around with ideas on how to make this easier to do, and easier to consume. Google Reader + some AI is the solution, and currently nothing else does just that. Especially because all the reading apps/ platforms feel compelled to add feeds and recommendations to make money, and then this just seems to take over…
Good list! Subscribed