Voice Ramble

Alright, so hopefully you can hear me. I’m going to try to talk out loud in an effective way. And I’m mostly doing this for testing. And I’m trying to see whether or not the Astral, um, the Astral Web app, um, platform can do this well. There’s probably going to be some weird hesitancy when I talk for the first time because I don’t know if this is the best plan. My goal here is to try to develop the most transparent, trustworthy environment for AI systems performance engineering. What this includes is probably going to be open source, is probably going to be some hiccups that are going to be heard, and there’s going to be some level of rawness that I need to convey. And, yeah, I, I think that this is going to be unique in that sense, where, in an age of, I guess, AI slop, there’s going to be a lot of thoughts, and, a lot of, how do I say this? This is a little bit awkward for me even to talk. I have a relatively difficult time, like, processing how awkward this is. Maybe, maybe this isn’t the best approach. Maybe there’s a better approach here that I can do. Either way, I think the most important thing right now is to convey to the people that I’m a human being. That I have just normal issues trying to figure out what to do. And I think the difference here is that I’m trying to convey in an effective way showing the work, the works, and in addition to that, just get things done. Particularly when we’re in the AI ecosystem, there’s a lot to get done and a lot to process and a lot to, um, to understand. And unfortunately, or fortunately, there’s a lot to really learn and deep dive. And it’s just not a matter of just learning or understanding or doing critical thinking, but you really have to move the needle forward. And in addition to that, it’s just not a matter of working on one dimension. There’s multiple dimensions here, especially if you consider this idea of hardware, uh, hardware-software co-design. And there needs to be some level of understanding on different abstractions. And it needs to be conveyed all the time. Um, an example of this is, of course, like if you do an optimization technique, let’s just say speculative decoding, for example, or continuous batching. There might be some level of regression that, unfortunately, creeps up for latency, might be for throughput, and it might be somewhere you don’t expect. So, unfortunately, when you think of a good solution, you almost have to optimize for one thing at a time. And then think about it. Think it in a way that is scientifically sound and methodical without overwhelming yourself. Because you, you do have these SLOs that you have to adhere to. But at the same time, you’re dealing with other performance engineers, and they have their own objectives. And of course, it’s supposed to be a team effort, but, you know, if there’s a scenario where you’re getting sidelined because this particular SLO is much more important than your SLO, then obviously there’s going to be some level of tension because your work might, um, cause their work to regress, which is not good. So, in that, you’ve got to figure out a nice balance of, um, teamwork while also trying to make sure your own objectives are working. So, anyway, uh, this is just me rambling. I don’t know what to call this kind of series. I’m probably gonna try to do it often. And I’m not gonna just do random live to kinda ramble about certain things, although that could be useful. I think that what’s particularly important is trying to be authentic, be transparent, and I think I’m probably gonna try to do this for learning and understanding new things, which is difficult. So, there’s a lot of different processes there, and just trying to collaborate and put things in a concise way that makes sense. Um, and also try to avoid using too much AI processing as well. So anyway, this is my first, or second post rather. I did a Hello World version, and I’m gonna see whether or not I can post this online, and to my website to kinda, um, see the voice quality, see whether or not it works, and uh, hopefully this is something I can do quite easily. So, we shall see how this works.

Voice Demo

Voice Ramble