Testing AI generated wind turbine designs | ChatGPT | Turbine Challenge | VAWT
Vložit
- čas přidán 9. 07. 2024
- Continuing the Turbine Challenge Series. Today we'll see if ChatGPT is going to take all of our jobs by testing AI generate wind turbines.
...and then destroy the evidence. #chatgpt #experiment #diy #3dprinting #science #windturbine #nikolatesla
Windtunnel data: github.com/hellothisismax/tur...
How to submit a design for "The Turbine Challenge" (Turb Off).
• I want YOUR turbine de...
3d models and Drawings:
www.thingiverse.com/thing:634...
Accepted File Formats:
3d: .STL or .STEP
2d Profile: .jpg/.jpeg
Design submission form: forms.gle/4Tph5mnUZPTrSdER9 - Věda a technologie
Large LANGUAGE model. Got nothing to do with design and optimization. If you want to use AI to solve mechanical optimization problems you better off starting with Fusions generative design or the likes. This is akin to using a dictionary as a calculator.
I live the "could you" life, not the "should you" life.
More like using an encyclopedia as a calculator.
@@nonarKittenwell it can work, the info is in there lol.
He used GPT to design and tested in real life with 3d parts.
If someone has better solution, do it and show the video rather than criticize
@@ThiagoMarquardt When someone starts using the wrong tool for the job and then criticizing the tool for not doing the job it needs to be called out. It shows at best a lack of understanding of the subject and at worst click farming. Claiming "Have you heard, AI is going to take all our jobs, or at least thats what someone has been telling me... (0:18)" and "...some of our jobs are safe for now (9:14)..." is implying that LLM's are the only type of AI and AI is incapable of solving this problem. Engineers use non LLM generative AI every day to perform these exact optimizations. Must I accept every piece of content without criticism despite being objectively incorrect just because someone made a video and posted it to YT?
That "coin purse" is almost an exact match for a McDonald's apple pie (or its cardboard box)
You won the naming contest!
Did you specify you wanted the designs to be optimized for maximum output ? If you just asked it to design some turbine blades it fulfilled the request perfectly .
Good point, and one I left out of the video. I asked it optimize the profiles for maximum power output for a VAWT.
"I'm beginning to think Chat doesn't know what it's doing" - this is more true than you may know. It's predictive text. It has no idea what it's doing. It's the whole Chinese Room argument. General AI is on the way and that's gonna shake things up.
You may have better luck with a GPT model that is aerospace-specific.
One biggest engineering teagic in history that will happen once dumb as starts to use AI as a life support on most safety proudducts like cars,ladders so rhat puts people at death risk.We are screwed!😅
ravioli foil
😆
If you're using PETG or PLA, I've found that the best adhesives to use are hot glue if you're short on time or E6000 if you have an hour to spare. CA glue has its place, but that place is only in tiny parts or temporary fixtures. That said, if you're designing parts to be glued together, the best option is to print in ASA/ABS and use acetone to join them. If your printer can't print ASA, there's always the blowtorch/soldering iron method.
also, the turbine hungers for some Okra
Plumbers pipe solvent does the trick, ten minutes or so to grip and then 4 hours to fully cured.
After that 4 hours it's a stronger bond then the layer adhesion 😁
Loctite makes a plastic adhesive in a blue bottle that works like CA but bonds quite well to PETG and PLA.
As far as superglue, i think there is an activating spray that makes it set quickly. the interweb suggests looking up CA glue and activator
I'd call the coin purse 'the lemon' but its surprising performance proves otherwise.
Here in Portugal we call it a "vagina".
superglue is activated by moisture apparently and your finger has more moisture on it than the plastic that has no moisture on it... you have to blow on the glue... your breath contains moisture... this may be a wisetale
O, well that's an easy solution lol
What you need to do is carefully develop a persistent chat where you prompt it to be a hyper intelligent design engineer and feed it the performance data of it's designs and have it iterate on them to become better. Where these models become really "job threatening" is when complex multi agent systems are developed with large and dynamic enough memory envelopes become implemented in single board solutions; they will be able to process data and search through possibility domains with a speed and accuracy which will have no close human equivalence.
I designed a model vertical axis turbine a few years ago. My criteria was it had to be adjustable for angle of attack and self starting. I have no way to test it for efficiency. But it did work as designed. My test was on a windy night I set it up at 30 mph winds it was spinning great. When I woke up it was all over in pieces. Wind peeked at 40 mph that night. Never built it full size it just wasn't needed, solar works fine.
hey max, some remarks:
1/ if a turbine wobbles so much you alter the aerodynamics and degrade performance
2/ novel airfoil... you can mount the upperside inward, and should spin faster...
3/
That kind of glue cures using the moisture in the air, so if you breathe on the parts a little before applying the glue, it will cure a lot faster.
If you want to glue quickly, put it in the humid environment.
Coin Pursie? (I am so sorry... ) I love this series!
Hi, In reality VAWT do not operate in the rev band you are using. if they did they would last minutes. Look for power curves below 200, record torque. keep going
😂 I'm subbing for the comedy alone! Seriously though, cool idea man and thanks for sharing
It could be used as a reward model for another model that learns to optimize turbine designs like that eureka paper on robot manipulation from nvidia.
I think you should make a turbine from raw carrot. I have complete faith in you.
It looks like an Egyptian hieroglyphic eye to me.
Eye of Horus
@@kbee225 that's the one
I'm not surprised by the results, considering AI doesn't really stand for "Artificial Intelligence" but rather "Averaged Information". Now proper AI in a simulator might slowly work out which design has the lowest drag and highest torque or RPMs in a shorter time than an engineer could go through the many iterations required, but the engineer can use intuition and think about flexible blades, deflector, flaps, etc, which is beyond the scope of a simulation, so I don't think that kind of job is at risk.
The problem is this: ChatGPT in it's current form is still pretty basic and flawed but it's the RATE at which it improves that is impressive...if you redo the same thing in 6 months from now you'll see it'll be even more potent.
I just asked Claude 3.5 Sonnet
LE
TE
Chord Line
Camber Line
I want to call it the knife sheaf
lets named it Pinched Oval
You can do a lot with openscad and chatgpt too. Sometimes it hallucinates code but can usually fix that easily.
also it looks kinda like a snapper
With a vertical axis wind turbine it is impossible to exceed an efficiency of 14.8% and in general we barely reach 7%.
I would call that shape "lemon" or "mouth".
Provide feedback to the gpt on the performance of each of the models and ask to improve it
Or use an AI that is actually designed to optimize and not a chatbot.
You can see the coin purse worked because the structure was twisting along the rotational axis. That probably created something of a traditional vertical turbine.
Rolling shutter effect. Smarter Everyday did a really good video on this subject :D it's why fast moving objects sometimes appear curved when taking pictures or filming them, the camera sensor fills in each frame a line at a time, and if something is moving fast enough it gets a little offset
well... that went as expected 🙂
Hahaha it did
Enterprise D deflector dish.
Whatever you call it, print it with tpu 👀
😅
and try to operate it.
Crazy glue activates with water.
Looks like Capt. Terauchi’s Marvelous Spaceship
Here’s the problem:
You are asking a blind person with no arms or legs to design an airfoil.
Language models aren’t trained on spatial data, they’re trained on language data. (Shocking)
Not sure if Max knew that or they were just trying to be trendy…
when you put it that way it makes me think chatgpt did a great job!
I would say your wind tunnel is just having imperfect airflow, which allows uneven application of forces.
It has been a long time since I’ve seen that jellyfish designed one.👍 It seems like I remember it being designed about ten years ago.
You're right, it was about 10 years ago haha
2:30 you said the glue is wet forever. Try activator spray, or spray water on it. It hardens immedeateliy
that thing with superglue tho, its catalysed by dirty hands, or rather, human hands.
btw, if you want to have it dry faster, apply it on dirty parts, it dries instantly.
dissolve some baking soda in water, and give it a spritz to make it dry instantly (the base catalyses it), rub your greasy fingers on it (skin is naturally basic)
I wonder i what would happen if you used gear ratios to up the torque required to spin it as right now it seems like you only need a small vertical slice of wind to make it work but if it was harder to spin you might be able to harvest the power more efficiently.
I think I'm going to build a brake so I can measure torque. I've been told(by a few comments lol) that's the proper way to test turbines.
the main problem with the cirrent crop of llms is that theg really ablr to operate on a things rhatbhas beendescribed in text. rheybhave no real experience and no really no idea about the reality we move around in.. its like having a blinf person describing the colour blue.
turb-eye-n
in spirit of the coin purse, how bout the "WhirlyWallet"?
Are you applying a increasing load to generate a torque curve?
Yep! Exactly right.
Hmmm airfoil was doing over 1000 rpm, more than double the other two, the butterfly was the easiest to start, and with some improvements may even have a decent amount of torque (for its speed). That is not a failure to my thinking. Depending on your usage some sort of hybrid design would give you a good mix of rpm, torque, and the ability to start easily. Hybrid design being something GPT used correctly is really good at. I know you are just messing around, but you could literally come up with a ground breaking design if you use GPT correctly here.
It sounds like maybe I should try again :)
Comeon... you have a wind-tunnel but didn't know how to accellerate Superglue (aka. Poly Urethan)... try water aerosle (it works with your breath, but don't do it, se bellow!!), may buy some accelerator spray (aka a spray can with water)
And when you ask why it is in your fingers without touching it... in the reaction parts of the PU evaporates and condense on whatever is near, take kare that it is not your eyes!
you should get activator for the super glue
use sodium bicarbonate to instantly harden superglue
Ca with skin sweat and amino acids cause it to set instantly
How are you getting 0.04 Watts? Just by visually looking at it, it should be more than that.
It probably because of losses in the test setup. Friction, electrical resistance, other things (part of the wind tunnel is cardboard 😂). I don't adjust for these things. I'm measuring the actual power at the load. Next Season I think I may build a brake and measure torque instead of power output.
@@hellothisismax-just thinking, but haven't tried it. It seems to me that you already have a very good brake and a good way to control it. A turntable under the motor/alternator with an arm against a load cell would give you a torque value. VAWTs power comes in pulses but a flywheel should smooth things out.
The all seeing eye
Try Claude it is much better at that!
hmm , this is like the fan showdown ...
Lemon.
Diamond
The Pineal Gland
You have to let him (AI) finish and not
Make print mid-iteration
You did not ask ChatGPT to make an "efficient"wind turbine
I did, but I forgot to mention it in the video.
whats your testing setup?
I'm using a 150kv brushless motor as a generator and a variable load controlled by some software I wrote.
instructions unclear, it still worked?
This is quite interesting lol I used the same method to create sex toys to enter a competition
Chat Jippity
ask it to write some openscad
It’s The Kissy Lips‼️💋
Buy an activator, you will never stop using superglue after that .. i use it for everything from wood filling to gluing stuff in the lathe :P activator/accelerator also makes the glue "filling" more =)
You do know you asked a language model....?
Its not generated anything, its just designes its scrapped off the internet.
All the models are oscillating. The support struts are too thin.
great video, but please dont do cell shading in your hands
aeropurse
lol @ Gipitee
The karen
such an underrated channel
Was it gpt 4? (no 4o)
It was 3.5
@@hellothisismax you should really try 4 one (not sure about 4o compared to 4, but still 4o free and should be significantly smarter then 3.5)
Still gpt can't make good something completely new, its only can learn from given information, and try to find some relations in words. But it should be more focused with finding form for you from what he trained on with 4 version
It would be fun to see if 4 can do a better job than 3.5, I might give that a try.
@@hellothisismax try 4o too, it seems very smart
Aerovulva™ how about a boiled egg
A hard boiled egg?!?! That's not a vegetable 🤣
@@hellothisismax . . . soft boiled???
You know what I'll do a hard boiled egg if you win lol
@@hellothisismax hahaha, to be fair, maybe we can compromise with an eggplant. ;)
I'm currently growing an eggplant plant, this might turnout great hahaha
Ha! I beat GPT. Not by much, but still.
Ai creations cant realy be used at all. That gpt thing gives alwqys wro g answer when going deep math
Chat gpt shit this 😂😂
Vulva
Don't use super glue on plastic
We learned what we already knew, transformer networks like GPT can barely form a coherent sentence, they're like an idiot interpreting a textbook they don't understand.
Indeed, you cannot answer "how do you make dinner" by merging 1000 recepies.
AI hype will not age well. playstation 1 graphics will age better.