Wednesday, December 21, 2022

Science News Dec 21



In today’s episode, we’ll talk about the recent nuclear fusion headlines, and a new result from the Webb telescope. Then we’ll have a special guest, Fraser Cain, who’ll tell us what we learned from NASA’s Artemis mission. After that, we’ll talk about remote controlled magnetic slime, why atmospheric methane levels increased during the COVID pandemic, non-fogging glasses, algae that might replace beef, the toughest material on earth, self-organised nanobots, dark photons. And of course, the telephone will ring.

Transcript, references, and discussion on Patreon.

Wednesday, December 14, 2022

Science News Dec 14



Today we’ll talk about the 50-year anniversary of Apollo’s blue marble, the Square Kilometre Array, talking to robots, the Yellowstone volcano, nanoparticles that help with carbon capture, a forest bubble – on Mars, a new method of spacecraft propulsion, tests for cancer, earthquake tracking from space, and of course, the telephone will ring.

Transcript, references, and discussion on Patreon.

Wednesday, December 07, 2022

Science News Dec 7



Today we’ll talk about zink batteries, regulations for StarLink satellites, find out how birds feel about fireworks, learn about a new way to detect sugars, thirty thousand molecules that could make better batteries, and more efficient photosynthesis. I’ll also briefly have to comment on wormholes, again. And of course, the telephone will ring.

Transcript, references, and discussion on Patreon.

Saturday, December 03, 2022

Calorie Labels: What Do They Tell You?



Transcript, references, and discussion on Patreon.

Guest Post: Collider Rap by Mike Battaglia and GPT-3

Image by rawpixel.com on Freepik

These days, it seems like Artificial Intelligence is taking over the world. But you don't have to escape its clutches - why not have a little fun with it? One of my favorite ways to lol at AI's expense is by getting them to mix together ridiculous things and see what they come up with.
This time I asked an AI model to write rap lyrics about particle accelerators pretending to be A Tribe Called Quest. Q-Tip would present the viewpoint that Sabine Hossenfelder has right: Particle accelerators are a waste of time! Then Phife Dawg would pretend he was an electron from inside a particle accelerator and argue for their importance as research tool –– Busta Rhymes does his verse from "Scenario", but changed so that it's all about particle accelerators instead.

Here are the lyrics that this AI model wrote:

Q Tip:
Sabine Hossenfelder's sayin' the accelerators a waste
Costs a lot of money, nothin' to show for the chase
They say the science is great, I say the cost is too high
Money's better spent elsewhere, that's the reason why

Phife Dawg:
Don't wanna hear none of that, I'm an electron with pride
In need of an accelerator, please don't take away my ride
From the Big Bang to the stars, I'm out here explorin' space
Don't let Sabine's perspective keep me from the human race

Q Tip:
We can't ignore the cost, it's too high for the gain
The science is important, but so is the financial strain
We gotta find a balance, between knowledge and the cost
Else we'll be stuck in a jam, and our fates will be lost.

Busta Rhymes:
Watch as I combine all the science of space
Now Sabine Hossenfelder says the accelerators a waste
Powerful impact boom from the beam,
So I'm investigating both sides of this debate of the dream
Vacuums and volts providing the power
As the electrons go faster and faster, the experts devour
Oh, my gosh! Oh, my gosh!
Time and money embark on a chase
Unh! Unh! Unh! For the cause, no time to waste
Unh, pardon me, unh! As I make this call
Should we keep pumpin resources or close the accelerator hall?
Rawr! Rawr! Like a scientific dragon
Change your little opinions, cause this knowledge is flaggin'
Try to end this debate, I'm gon' twist it like a durag
Cuz I'm caught in the middle of the debate like a purgatory
Chickity Choco, the chocolate chicken
The electrons are goin', pushin' the science mission
Rippin up the theories like Horatio
Observe the vibe and check out the scenario

---

The results are hilarious––I'm literally cracking up here––though simultaneously I worry that maybe I've enabled some kind of robot apocalypse through this escapade (what can I say, looking wistfully out into the future during lunch break). If speaking English makes an AI conscious, then what will these rap lyrics do?

Ah well, we'll find out soon enough...and when we're done playing around here let us just take a moment to marvel that an AI can even do this.

Even crazier is that it wrote this entire blog post (including - oops - this sentence!) from scratch.

---

Note from real Mike: it's true, although it wasn't exactly "from scratch." I basically gave OpenAI's new DaVinci-003 GPT-3 model an outline of the blog post. I also had to build it up in parts; first I had it do the lyrics with one prompt, and then I had it build the blog post as a separate prompt. Still, it managed to write this as the result. I have only very lightly edited the formatting by adding line breaks.

I should note that it took a bunch of playing around with the parameters and prompt quite a bit before getting this output. In particular, I found most of the struggle to be in wording the prompt correctly; I had to try a bunch of different things before I could get the model to figure out what I wanted it to do. So I guess I'll leave it to the philosophers to debate if an AI really wrote it all "on its own."

I thought the results were absolutely hilarious when I shared it on Facebook. I also think it raises some deep questions that are worth thinking about. On the one hand, I guarantee every single person reading that Busta Rhymes verse, who knows the original, will be cracking up hearing it in his voice in their heads. On the other hand, the current model is clearly not quite able to really replicate the dense multilayered lyrical wordplay and flow that real rappers are capable of. But at the rate things are moving, it probably will, possibly very soon. I don't know what to make of it.

All I know is this: as of 2022, you can tell this thing to write some rap bars about particle accelerators and Sabine Hossenfelder and it will actually do a baseline half-decent job at it. Then you can get it to write a blog post about how it wrote the lyrics and a meta-blog post about how it is capable of writing blog posts. It's really nuts. Anyone right now can go to OpenAI and play around with it and get results like this with a little effort.

GPT-4 will be available in 2023 with 500x the amount of parameters. Who knows what that will be able to do.

(And RIP to Phife Dawg, probably my all time favorite MC)


Mike Battaglia is a musician, biomedical engineer, and digital signal processing specialist. Check out some of his microtonal music on YouTube and Instagram.