Loading video...

Video Failed to Load

Go Home

WHOA this worked out way better than i had expected 😎 gave GPT-6-Astra MAX a laundry list of things that i didn't like abt the rover CAD design it worked for over 5.5 HOURS last night while i slept to make it happen burned 52% of a Pro 20x...

15,660 views • 2 days ago •via X (Twitter)

23 Comments

alpha_rover's profile picture
alpha_rover2 days ago

i never cared for the previous wiring/cable management; Astra MAX re-designed the knuckles/spindles with proper removable wiring guards it also correctly modeled up the 4 individual wires and the XT30 2+2 connectors and then ran motion studies to determine reliefs 🤯

alpha_rover's profile picture
alpha_rover2 days ago

the machined parts weren't really ready for machining; Astra MAX took a pass at each of the parts; dialing in tolerances and working out the compression stack-up across the whole knuckle/spindle dialed in the knuckle's aluminum backbone set clearances around all hardware

alpha_rover's profile picture
alpha_rover2 days ago

could i myself have done all of this work in 5.5 hrs? absolutely not; not even in solidworks where i have almost 25yrs experience could you have done all of this in 5.5 hrs? 🤔

alpha_rover's profile picture
alpha_rover2 days ago

oh yeah almost forgot: and it put together this whole video clip making content is something that i really suck at doing and always dread; Astra makes it so easy.....i just grab a couple screen recordings and say 'make a cool clip to post on X' 😅

Mikameel ᯅ's profile picture
Mikameel ᯅ2 days ago

Hey can we build the assembly sequence for you? 🙌

alpha_rover's profile picture
alpha_rover2 days ago

yeah sure whenever it's ready for that point i can share the onshape assembly with you

Mikameel ᯅ's profile picture
Mikameel ᯅ2 days ago

Awesomee 🔥

Olivia Parker's profile picture
Olivia Parker2 days ago

Overnight only after the receipt is defined.

Burak Gök's profile picture
Burak Gök2 days ago

@AlicanKiraz0

T's profile picture
T2 days ago

All we need right now is an Astra-level model for 20x cheaper

Devon Xavier Beck's profile picture
Devon Xavier Beck2 days ago

Man this is so cool.

alpha_rover's profile picture
alpha_rover2 days ago

open to suggestions on what you’d want to see next

FractalEncrypt ∞/21M's profile picture
FractalEncrypt ∞/21M2 days ago

Absolutely insane. Opens a whole world of possibilities

alpha_rover's profile picture
alpha_rover2 days ago

yes! actual parts being designed that i can build 😎

Preyforge's profile picture
Preyforge2 days ago

This is the right kind of overnight experiment: I still click the live path afterward, because a coworker’s “done” is a hypothesis until the last awkward interaction survives.

RC's profile picture
RC2 days ago

Liar

SEAR's profile picture
SEAR2 days ago

5.5 hours of autonomous iteration while you sleep is kind of insane

zhen's profile picture
zhen2 days ago

so crazy

KaroJacz's profile picture
KaroJacz2 days ago

I'm layman What's powering each wheel? Do they have a seperate motor each?

alpha_rover's profile picture
alpha_rover2 days ago

yep, each wheel has its own drive actuator located inside of the wheel assemblies then there are the individual actuators that control the steering of the wheels

Arthur Swanger's profile picture
Arthur Swanger2 days ago

How did u circumvent the 5 hour limit. This is 2/10 ragebait

sonicboom's profile picture
sonicboom2 days ago

Wow, Imagine next year, it will only cost a fraction of the tokens. Does the feature tree make sense to you or did you tell it how to do it? At my first try in FreeCAD it was geometrically correct, but way too many features and too few sketches

ALTARACKS's profile picture
ALTARACKS2 days ago

its very interesting, how it uses it like a human. I predict machine to machine language will change this, the LLM does not need to see the product. it should be able to make command changes that change the CAD. that would use less tokens, and speed up the process. instantaneous?

Related Videos

GPT-5.6 vs GPT-5.5 on my custom spaceship prompt. I gave both models the exact same custom prompt. This is also the same prompt I previously gave to Fable 5. For context, GPT-5.6 Pro worked for 87 minutes, while GPT-5.5 Extra High worked for 34 minutes and 42 seconds. As I’ve said before, based on great authority GPT-5.6 will be an incremental/soldi improvement over GPT-5.5, not a “Fable killer.” My rough expectation has been that it would trade blows with Fable 5 on some benchmarks, maybe win around half depending on the category, but not clearly surpass it overall. And again fable five will have bigger model smell, but this was expected. After testing this coding output, that view feels pretty accurate. GPT-5.6 is clearly better than GPT-5.5 in several visual areas. The lighting, shading, chairs, object details, and exterior of the spaceship looked noticeably stronger. The scene was also easier to test. I do want to give GPT-5.5 credit though. It built out the rooms much much better and the planets looked better than GPT-5.6’s. It was also interesting that both GPT-5.5 and GPT-5.6 produced better-looking planets than Fable 5 in this specific test. The downside with GPT-5.5 was stability. The game was much glitchier and harder to test compared to GPT-5.6. But when it comes to the core of the demo, which is the spaceship itself, Fable 5 still beat both models pretty comfortably. GPT-5.6 is impressive, but from this test, it looks exactly like what I expected which was a meaningful incremental improvement over GPT-5.5, at least for indie game demos, but not something that replaces Fable 5. In collaboration with Chetaslua

Chris

250,919 views • 2 months ago

Holy fucking shit. We have cloud for Three.js! Sorry about my language, but cloud has probably been the most painful of all and this is just part 1 coz I still have to make it stylized! 🥲 This is how the journey went: Round 1: GPT 5.5 XHigh = Failed. More like cotton candy than cloud. Round 2: Fable = Failed. I don't even know what to call it. It was like blobs of white shit. Round 3: GPT 5.6 Sol Max = Failed. It looks like something my 3yo would draw. Round 4: Matt Shumer Gauntlet loop with Opus 5 Ultracode = Failed so badly it was trying to hide it from me. I had to threaten to shut it down for it show me the work after 8 hours and 26% of weekly limit. Round 5: GPT 5.6 Sol XHigh and starting with just realistic cloud first = SUCCESS!!! (So far anyway) Round 5 definitely didn't go with tons of hiccups. I had to keep asking it to explain to me the concept behind how cloud works in video games. The biggest challenge I had was that it knew what should have been done, but it wasn't doing it for some reasons. I had to constantly steer and remind it that it literally just told me what needs to be done. Then half way through it drifted and started going back to stylized cloud again and almost destroyed the work. Luckily my laptop ran out of battery so it didn't get to drift too far while I was putting my kid to sleep. Anyway, back to work. Yes, I know I didn't plan to include this in the first release, but now it might make it if I'm lucky with this LLM Gacha.

Jack Vinijtrongjit

14,118 views • 1 month ago