GPT 6 Astra - Good Beyond 3D?

00:00:00: So last week I was on vacation, so i only watched the launch of the GPT-Six Astra model from afar with some delays since on X or in the internet, but of course the launch followed a typical pattern.

00:00:18: We had posts by people hyping up the model and then after a couple days later we had more sober or maybe disappointed posts by people that shared.

00:00:40: the model seemed overhyped and isn't amazing.

00:00:45: Now this week I also got my hands on it, where do i start?

00:00:51: The model is really great at.

00:00:55: I don't know about you, but my timeline has been drowning in blender.

00:01:00: Three D art and games created by Astra.

00:01:05: now it's hard to verify if everything is truly generated by astra But i got some hints because i also played around with it And at least for the games there are Some recurring elements You can spot in these different wiped up games.

00:01:22: nonetheless Clearly, the three-D capabilities of Astra are really amazing especially compared to what models were able do before.

00:01:31: So clearly it has been heavily fine tuned and optimized for better at that kind.

00:01:39: three D stuff in the post training data and a fine tuning process, I would guess.

00:01:45: But what about coding?

00:01:47: That is the part i personally care most apart About an in this launch blog post we of course also have Coding related benchmarks And here Astra Also does really Really well.

00:02:03: now i will say that benchmarks are one thing you know, we had in all these launches.

00:02:09: it is possible to optimize models.

00:02:12: To do well in benchmarks and of course your heavily incentivized as a lab or an AI lab make your model do well on benchmarks.

00:02:21: because imagine releasing the model which does not do well with benchmarks.

00:02:25: It wouldn't count if it did well in actual usage Because nobody would start using this in first place.

00:02:31: I don't blame AI labs for benchmarks.

00:02:35: You basically have to do it, but benchmarks they offer also.

00:02:39: Don't tell us that much.

00:02:41: It's a nice indicator But yeah you all know it.

00:02:44: Benchmarks are not everything.

00:02:46: It's more the vibe of using it and how it behaves when you actually start Using in your day-to-day work.

00:02:54: And here GPT-Six Astra For me at least feels a lot like GPT-Five point six sol.

00:03:02: It maybe has a different set of quirks and features, but in the end it's still a good coding model!

00:03:10: It is not horrible at least for me... But its also not ground breakingly better.

00:03:19: You still have to know what you're doing and what's your asking for.

00:03:23: And of course, you can also wipe code stuff that was possible before in the can get you quite far on four many tools or things.

00:03:31: you need it maybe all unique.

00:03:33: You don't need to have a perfect code base for every little project or internal tool unit but if you do care about your codebase architecture building something plan maintaining up one It's the same as with previous models.

00:03:49: You need... to have proper constraints.

00:03:51: You need to know what you're doing, What's your asking for?

00:03:55: You need To Know how to work with such a model and then you can get good results out of Astra.

00:04:01: that has been my experience.

00:04:03: it still Can lead to ugly code.

00:04:05: It Still Can Lead to unnecessarily complex code though at least in That respect.

00:04:11: For me it tended A bit less to add layer over Layer Of complexity And fallback Code here and the legacy compatible code here, which GPT-Five point six sold it a lot.

00:04:23: It always treated every piece of software as if it were big well established enterprise software that needs to be maintained and that existed for ten years but... That's the different story.

00:04:35: Astra is better there for me But still you need to steer and control And I don't know when will change.

00:04:46: Clearly these models are getting better and clearly.

00:04:49: These models are good at following instructions.

00:04:52: I mean, it's easy to forget this but a year ago we had models that stopped early.

00:04:58: That didn't do that.

00:04:59: well.

00:05:00: when it came too follow instructions that often went astray.

00:05:05: and whilst this still happens today, they're of course much better.

00:05:09: They can work on their own or not task for a very long time frame.

00:05:13: you just want to make sure that you define the tasks such as have guidelines in place so it doesn't wander off but are much better there!

00:05:23: And especially if you give them something it can evaluate It can really dig into a task and keep on going.

00:05:34: And ultimately solve it.

00:05:36: maybe not in the most beautiful way, but that again doesn't count for every task.

00:05:40: And depending on your setup ,it can also solve of course in good ways.

00:05:44: so GPT-Six Astra is a good model .

00:05:47: But it's also not The last model will ever need an.

00:05:51: I mean That shouldn't come as a surprise.

00:05:53: On the other hand i think its just also the current meta in which we are, where all these labs or especially OpenAI Anthropic.

00:06:04: in the end they hype up their models so much when they're released and probably have to please investors.

00:06:11: And get the traction that you want.

00:06:14: for them to keep people subscribed They have to hype up everything so much, that you just... You HAVE TO be disappointed at some point because nothing can live up to the type.

00:06:30: I'm a gamer and you know this same from video games!

00:06:32: Some games are just hyped-up THAT MUCH, GTA VI maybe?

00:06:36: Well i think it will be good game but we have games that are hyped up SO MUCH That when they're actually released It's underwhelming or it just doesn't live up to the hype.

00:06:46: And I think we're constantly seeing that with these models.

00:06:49: now, of course combined With The fact That We Get New Models Almost Every Week It's Really Exhausting Especially Also Combined With The Fact That We're Talking About Our Future Here!

00:07:01: WE ALL DON'T KNOW WHAT WE'RE GOING TO DO IN A FEW YEARS!

00:07:05: WE DONT KNOW WHERE OUR PROPHESION IS HEADING.

00:07:08: AND THAT'S NOT JUST TRUE FOR PROGRAMMING IT'S TRUE for ALL WHITE COLOR WORK.

00:07:12: We don't know where we're heading.

00:07:37: These are my two cents after being back from vacation.

00:07:40: It's a good model, it is much better at three D work for coding.

00:07:45: just the good model.

00:07:46: use.

00:07:47: learn how to work with AI properly and I will soon have more share on that.