스크립트
00:00:00I came across a post on X where the post itself is interesting.
00:00:05I'll link it below, but where it was the last sentence that caught my attention.
00:00:10I'm getting really sick of having to re-evaluate this stuff.
00:00:15And this stuff is simply figuring out how to work with a model,
00:00:21how to properly prompt it and which model to use for which task.
00:00:25So I'm getting sick of having to re-evaluate this stuff every two months,
00:00:30which I would say is quite generous.
00:00:32You could put every two weeks here, I would say.
00:00:35To be honest, a pause might be nice for that reason alone.
00:00:40And let's talk about that.
00:00:41We're, of course, in an era at a point in time where we got new releases,
00:00:48new AI model releases by the top AI labs almost every week.
00:00:54That is how it feels like.
00:00:56It definitely ramped up in speed again,
00:00:58but it's not just the frequency of model releases,
00:01:02it's also the hype that's always built up when a new model is released.
00:01:06I talked about this in my last video about the new GPT-6 Astra model
00:01:11and how that was hyped up.
00:01:13And you know all of that.
00:01:14If you're on X, you see all those hype posts whenever a model is released
00:01:18and then the disappointment shortly thereafter.
00:01:22So the frequency is higher and the intensity, I would say, is higher than a year ago.
00:01:28And combine that with the fact, and I also talked about that in the other video,
00:01:33that of course we're talking about a development that has a huge impact on our future.
00:01:38I mean, we all don't know what exactly we'll be doing in two or three years.
00:01:44We don't know where this technology is heading and how good it will be in two or three years.
00:01:50It is definitely getting better step by step almost every day.
00:01:56And if you compare models from today to the models a year ago, there is a large gap.
00:02:03And of course there was this very special point in time in late November with Opus 4.5 being released
00:02:09where we had this especially big step up.
00:02:12And since then, the steps have been smaller.
00:02:16It's not that big game-changing step anymore.
00:02:19But still, even since November, we got models and tooling around the models, of course, AI agent harnesses
00:02:27that ensure that these models can work on tasks for longer on their own
00:02:33and that if you give them a way of measuring their outcomes, they can achieve really great results.
00:02:40At the same time, I don't know about you, but this is very exhausting for me, for Kenton here, who posted this post.
00:02:49It's of course exhausting because it feels like you're constantly missing out on stuff.
00:02:56You don't know, as I mentioned, where we're heading and what our future will be.
00:03:00And we just never get a break.
00:03:03Last week, it was GPT-6 Astra.
00:03:06Then we had Fable 5.1.
00:03:08Now, it has been pretty much confirmed that we'll soon have Grok 4.7 and so on.
00:03:15We'll keep on going.
00:03:16We'll get new GPT models.
00:03:17We'll get new Fable models or Anthropic models in general.
00:03:22It doesn't stop.
00:03:24And that is clearly exhausting.
00:03:27It's also exhausting that the tooling, of course, is always evolving.
00:03:31The models are one thing.
00:03:33But the agent harnesses and how we use them, that's also changing.
00:03:38Constantly, you can read that you should be building loops.
00:03:42Then you should be building graphs.
00:03:44Then you should be building software factories.
00:03:48I don't even know what I should be building right now.
00:03:51I just know that I'm not really doing any of that in the way that it's sometimes advertised.
00:03:58Of course, I am building applications and tools and things, software of all kinds with AI in ways where the AI can work on its own on tasks,
00:04:10where it can measure its outcomes, and where it can keep going on its own.
00:04:15Not on the entire software in one go, but on the individual features, often also in parallel.
00:04:22I'm, of course, doing that.
00:04:23And you could be calling that loop engineering.
00:04:26I guess you could call it a software factory if I build a process where a pull request is automatically analyzed and handled.
00:04:34And I've been doing that too, but it feels like there is a new way of working with AI you have to learn every week.
00:04:42And then we have these new models where you also have to adjust how you work with a model for that new model
00:04:49and for the different models of the different providers if you have access to multiple models.
00:04:54So it essentially feels like every week what I did before doesn't matter anymore and needs to change.
00:05:03Clearly, that is, of course, wrong.
00:05:06Now, for one, it also means you can easily take off a week or two or a month because you're not missing much.
00:05:12What you're learning today may simply not matter in a month, at least if you follow that narrative of always being on the cutting edge
00:05:23and missing out and not going to make it if you're not building software factories, for example, right now.
00:05:29If you instead take a step back and you really spend time figuring out a process that works in general for you when it comes to working with these models
00:05:39and building software in a reliable way, that works much better, at least for me.
00:05:44That does not mean that you should stay asleep and never evolve how you work with these models, but it does mean that you don't have to jump on every new hype.
00:05:56Instead, automatically, you will evolve how you work with these models, but you shouldn't feel pushed to do it.
00:06:04It's a bit like in 2019 when we had the JavaScript framework wars, which I often mention, where it could also feel like a new JavaScript framework is released every week
00:06:15and you have to learn it in order to be on top.
00:06:20And if you are missing out on this new framework, you're not going to get a job.
00:06:25You're not going to make it.
00:06:26You're still using, I don't know, Angular.
00:06:31You're not going to make it.
00:06:32It feels a bit like back then.
00:06:33And back then, the important truth was that you also didn't have to switch or learn a new framework every week.
00:06:45It was a good strategy back then to just stick to your tech stack, get really good at it.
00:06:51And then, of course, evolve as a developer and generally as a human.
00:06:56You always keep on learning.
00:06:58It never stops.
00:06:59But it also isn't like twice a day where you need to change everything you know.
00:07:05Now, of course, granted, the pace picked up a bit with AI.
00:07:09Change is more rapid now than it was six or seven years ago.
00:07:14But still, calm down, relax, really.
00:07:17And it's hard for me too.
00:07:19I'll be very honest here.
00:07:20But I've been on vacation last week and that helps.
00:07:23Take a step back.
00:07:25Find your way of working with these models.
00:07:27When a new model launches, play around with it.
00:07:30See how it fits into your established workflow.
00:07:32See if you then maybe want to adjust your workflow.
00:07:35You'll do it automatically if results don't live up to what you want.
00:07:38Or you switch back to the old model, of course.
00:07:41But don't feel like you need to change everything all the time.
00:07:46It's easy to feel that way for all of us.
00:07:48But that doesn't lead anywhere.
00:07:50And, of course, I guess we would all or many of us would be hoping for a break.
00:07:55At least sometimes.
00:07:56But that's not going to happen.
00:07:57We'll have that high frequency of model releases for the time being.
00:08:01And we're still all figuring out how to work with these models.
00:08:04We're still in the Wild West era here.
00:08:07So I'm pretty sure in a year we'll work quite differently with AI than we do today.
00:08:13But that does not mean that you need to change how you work with it every day.
00:08:19It's a slow evolution nonetheless.
00:08:21It should be for you at least.
00:08:23These are my two cents on that.
커뮤니티 글
아직 글이 없습니다. 이 영상에 대한 첫 번째 글을 작성해 보세요!
이 영상에 대해 글쓰기