Automating the search for artificial life with foundation models

by hardmaru • 2 years ago

170 36 2 years ago

Congrats David & the whole team! Really enjoy everything Sakana AI produces and always look forward to your research results.

hamburga• 2 years ago

I actually found artificial life. Crocs. They keep on reproducing effectively and walking around (symbiotically with humans), with some mutation though the polysexual recombination process of Product Manager design reviews.

vintermann• 2 years ago

I think that it's a bit silly to call something life just because it resembles stuff you see under a microscope.

But I can't deny that it's beautiful. Unlike crocs.

mvkel2• 2 years ago

Fun fact: Sakana AI is founded by some of the authors of the original transformer paper, "Attention Is All You Need"

sourcepluck4• 2 years ago

Where's the fun part? I can't exactly imagine throwing this out as an anecdote to entertain a few friends during a sophisticated little soiree.

vintagedave2• 2 years ago

No need to poke fun. I found it interesting. Among friends who are interested in AI it’s the kind of random fact you’d throw into conversation.

jazzyjackson2• 2 years ago

Some people have a low bar for fun, for example, learning something new that connects to something they already knew, and saying to themselves, "Neat!"

diggan3• 2 years ago

to entertain a few friends during a sophisticated little soiree

Isn't this basically what (we'd like to think) HN is?

fedeb953• 2 years ago

try harder

exe34• 2 years ago

you have friends?

mvkel2• 2 years ago

Did you get lost looking for the fantasy football subreddit or something?

jeroenvlek• 2 years ago

In this context I can also highly recommend the Sara Walker episodes on Lex Fridman:

https://youtu.be/-tDQ74I3Ovs?si=1m0JV8gZEl4WFedG

https://youtu.be/SFxIazwNP_0?si=R7yZroSNbw5Jjc0H

https://youtu.be/wwhTfyX9J34?si=ceXh_aehsjQPklUT

nextworddev• 2 years ago

Curious - what’s the intended product direction of Sakana AI? Is it mainly a research lab or is it doing commercialization?

diggan3• 2 years ago

It's a incorporated for-profit company with VC investments, so somewhere/somehow there needs to be commercialization.

rbanffy3• 2 years ago

Can always be a pure-play IP house.

ribadeo• 2 years ago

The name of this company has real meaning in Português which I reckon is unintended.

jazzyjackson2• 2 years ago

I mean it's Japanese for Fish, but yeah, perhaps we need a database of false cognates sorted by number-of-languages-that-consider-it-vulgar

As for Portuguese, GPTo3 tells me "depending on context it can mean “bastard,” “scumbag,” “dirty-minded jerk,” or imply that someone is a lecherous creep. It’s essentially an insult calling someone sleazy or untrustworthy."

Would you say that's about right?

diggan3• 2 years ago

perhaps we need a database of false cognates sorted by number-of-languages-that-consider-it-vulgar

Or, like most people, we can assume the intent from the context and if someone says "Use git", we know they're not telling us to use a bum/rat/scum/whatever but the SCM :)

theGnuMe• 2 years ago

Are these cellular automata or Something more?

readyplayernull• 2 years ago

2025 prediction: Wolfram declares agentic cellular automaton supremacy.

fedeb953• 2 years ago

he's slowly building his army that will conquer all the computers in the world.

danwills• 2 years ago

Looks like it has both CA and particle-based. Sure it's cool but personally I'm a huge fan of a related type of simulation: 'Reaction Diffusion' both in terms of the less-aliasy (more continuous, and usually band-limited) visual results that it can produce, and also the range of primitive operations is more continuous and I would say wider, or at least meaningful in a different way (Laplacian, curvature, curl, gradient, complex numbers etc rather than discrete operations). There's some fun experiments on my YT from a while ago, here's an example:

https://youtube.com/shorts/WpY6eYZmmok?si=pxDoMWrsPf4idwNj

upghost2• 2 years ago

Wow, it is really interesting the difference in comments between ALife and AI stories on HN.

For some of you out there, there's a great book that really hasn't gotten enough attention called "The Self-Assembling Brain" [1] that explores intelligence (artificial or otherwise) from the perspectives of AI, ALife, robotics, genetics, and neuroscience.

I hadn't realized the divide was a sharp as it is until I saw the difference in comments. i.e. this one[2] about GPT-5 has over 1000 comments of emotional intensity while comments on OP story are significantly less "intense".

The thing is, if you compare the fields, you would quickly realize that which we call AI has very little in common which intelligence. It can't even habituate to stimuli. A little more cross disciplinary study would help is get better AI sooner.

Happy this story made it to the front page.

[1] https://a.co/d/hF2UJKF

[2] https://news.ycombinator.com/item?id=42485938

fedeb953• 2 years ago

thanks for your resources. I am myself concerned with the question of artificial life, and I wonder if it is even possible to search for it, or rather it will emerge on its own. Perhaps, in a sense, it is already emerging, and we humans are its substrate...

upghost2• 2 years ago

I'm not even sure that the goal of Artificial Life is actually "life", although that may be the AGI equivalent of ALife -- AGL or "Artificial General Life"?. In practice I think the discipline is much closer to the current LLM hype around "Agentic AI", but with more of a focus around the environment in which the agents are situated and the interactions between communities of agents.

Much like the term "Artificial Intelligence", the term ALife is somewhat misleading in terms of the actual discipline.

The overlap between "agentic AI" and ALife is so strong it's amazing to me that there is so little discussion between the fields. In fact it's closer to borderline disdain!

Y_Y2• 2 years ago

Apart from the obvious distinction that many of us on HN are making (or trying to make) money on LLMs I think you've also hit a broader point.

There appears to be a class of article that have a relatively ratio of votes to comments, and concerns such topics as, e.g. Programming Language Theory or high-level physics. These are of broad interest and probably are widely read, but are difficult to make a substantial comment on. I don't think there are knee-jerk responses to be made on Quantum Loop Gravity, so even asking an intelligent question requires background and thought and reading the fine article. (Unless you're complaining about the website design.)

The opposite is the sort of topic that generates bikeshedding and "political" discussion, along with genuine worthwhile contributions. AI safety, libertarian economics, and Californian infrastructure fall into this bucket.

This is all based on vibes from decades of reading HN and its forerunners like /. but I would be surprised if someone hasn't done some statical analyses that support the broad point. In fact I half remember dang saying that the comments-to-votes ratio is used as an indicator of topics getting too noisy and veering away from the site's goals.

rbanffy3• 2 years ago

many of us on HN are making (or trying to make) money on LLMs

I’d also highlight the misalignment between creating better AI and working towards AGI and extracting value right now from LLMs (and money investors).

j0hnb2• 2 years ago

Before I read the article all I could think about was what if AI was used with SETI's data, would we find something there?

rrr_oh_man• 2 years ago

What has prompted you to comment, after almost 10 years?

j0hnb2• 2 years ago

No specific reason, i'm here multiple times a day but I rarely comment.

Automating the search for artificial life with foundation models

Related Stories