I spent some time watching this Machine Learning Street Talk (MLST) interview with Yannic Kilcher this morning.
As a brief disclaimer, you can learn from somebody without having to agree with them or support everything they do along the way. People are complex and sometimes they can be chaotic. One of the early topics they covered was the complexity and reliability of the content being produced. I’m actually totally ok with MLST going as deep as they want during discussion and episodes. Not everything has to be written or shared for a general audience. Sometimes it is ok to have a highly technical conversation and dialogue with people who are capable of tackling those sorts of higher order challenges. A consequence of that type of effort will be that the audience size will be smaller. I generally believe that around 10,000 people worldwide are really hyper focused on consuming highly technical machine learning content. Audience sizes beyond that are picking up a different type of community based on purpose, interest, or circumstance creating an intersection with the general machine learning community of interest.
I’m going to try to utilize Twitter going forward to share links to the YouTube videos I consume each week. That should help allow me to provide some context and it will give you a sense of the content creators that take up several hours of my week every week. To be fair every week it will be me sharing the same links and commentary for the most part to the creators in my top 5. It turns out that I cannot really have more than about 5 podcasts in my weekly rotation. That is where I max out in terms of content consumption.
My Top 5 podcasts right now (when this post was written):
All-In with Chamath, Jason, Sacks & Friedberg
The Vergecast & Decoder (both are Nilay productions)
Hard Fork with Kevin Roose and Casey Newton
Machine Learning Street Talk
Lex Fridman Podcast (I’m 30/70 on listening to these based on topic)
My honorable mention would be the New Heights podcast from Kelce brothers.
Earlier this morning I was able to really work ahead of the backlog by one block of writing and I’m considering just scheduling the content to go live without recording audio for that post. I’ll probably spend some more time refining that block of writing and end up recording the podcast audio during some evening this week. We will see how that goes.
This morning I dug into the content block that I had assigned to this weekend with reckless abandon. You may recall that recently I shifted my entire writing backlog to the right by about a year and added a new set of content to the queue instead. I’m currently working on those new blocks of content creation and it happens to be going pretty well so far. Within my current workflow I’m targeting some audio recording every Saturday Morning and working on content creation during both Saturday and Sunday. My goal of course is to complete one block of the backlog each week while considering the next 5 in the backlog. That functionally gets me ready to work on the next blocks of content by considering them before ever trying to write or research. That is an important part of the process. You sometimes have to think deeply about something before you try to write about it.
A few general updates…
Yesterday, I did update the weblog menu to remove a couple of the social network items from the top to streamline content delivery.
I may at some point just remove the GitHub link as well. That one remained, but it is possible that it may get removed here in the next 30 days. That is something that is being considered.
The homepage is back to the about content instead of showing current weblog posts as well. I’m not sure if that is going to be a permanent solution to the homepage. It’s pretty easy to get to the posts section and nothing would stop somebody from reading them in order if that old school blog experience was desired.
I’m considering converting my entire writing collection into a corpus for model training again. It has been a couple years since the content was converted into a corpus like file for model training. The last time was for use with GPT2 about 3 years ago.
Things have been moving along and I have content prepared to go live until the end of May. It’s good to have gotten back to where I have a solid backlog of content ready to publish. Part of that gives me the ability to read and edit it 30 days later which is the key point where it’s like reading it fresh and my editing improves. You are probably aware that it is much harder to edit recently written things. Time is the best friend of editing.
One of the things that I’m trying to accomplish is putting a bit more effort into writing short little missives instead of just dropping storms of tweets. I’m sure it is a function of the wildness that has been occurring recently on that platform. Things are starting to get a lot weirder in general outside in the macro-economy. Seriously, global economic interactions are getting a bit weirder than they used to be in the last 5 years. You would have to zoom way in from that thought to get to the reality of my whining about sending a few tweets. However, that is exactly where you are right now as you read these sentences on the old weblog. Instead of waiting to have larger missives I’m just going ahead and hitting publish on shorter writing sessions. That might increase the volume of blog posts going out, but that is fine. By fine I mean that I find it acceptable and thus the path is set on a go forward basis.
One alternative I considered for a bit was to just write my tweets as the title of my weblog posts and let the integration push them over to Twitter. That would be an interesting way to go about sharing content on that social network. I do have the WordPress application on my phone and could make that plan a reality. It’s a lot more effort to build out a blog post. It requires categories, keywords, and both a title and body set of content. Those are not blocking factors of course they are just elements that require a little more time to make the plan work on an ongoing basis. It might be a good excuse to set more featured images on short posts. Overall, it is probably a bad plan to try to execute. I’m going to give it a little bit more consideration and you might see a few test posts to see how the practice of actually doing that works for me day to day.
Most of the time the writing process is about using a keyboard and typing at my desk. Writing on my smartphone just does not work for me based on the tactile interaction and process overall. I did generate a quote for a System76 laptop this weekend. It’s printed out and sitting right next to me as I consider making the move away from my Google Pixelbook to a different type of laptop. The total number of laptops that I have owned in the last two decades is actually not a very large number. I tend to have them until they experience catastrophic failure and are replaced. That is probably the same sort of way other users interact with laptop technology. Nobody really just needs a stack of spare laptops sitting around.
Over the years some of the posts from this weblog get some traffic, but the vast majority of them do not really surface out on the internet. This is for all practical purposes this weblog is just my secret corner of the internet. This evening I’m sitting and wondering about what to do after I deleted 198,364 emails from my gmail account. Cleaning up that many emails was truly liberating. I did not need the emails and now they are gone. While I was busy cleaning up my email archive the entire universe of Twitter seemed to catch fire yet again. It appears that Substack got restricted on Twitter today. My weblog system still drops links over on Twitter. Nothing about that appears to be restricted. It’s really the only social media advertising that I keep providing. To that end I keep writing longer and longer titles for my weblog posts.