How much AI did you use to write up this article? It tripped up my "fake AI-written article" detector a few times despite being interesting enough to read to the end
2,639 karma · joined August 11, 2010
florian dot laurent at gmail
I picture entities playing with our universe, "it starts slow but check it out at the 13.8B mark"
What would be more useful would be an automated list of places where the post has been discussed (and maybe pull the top comments from there through API?)
I’d love to hear a bit about the ML side of things: what was your experience with various models? Do you see a clear cost vs quality tradeoff with current state of the art models? How do open vs closed models compare?
When we launched 3 years ago our differentiator was that we could train both cheaper and faster by running on TPUs, these days GPUs have mostly caught up, and open source models are not as competitive as they once were.
It’s making ~5k/month these days, not bad as we’re no longer actively working on it, but a fraction of what we were doing a year ago.
The main challenge for us was the non-technical part. We built an API-first product because we love the tech and felt it’d allow us to focus on that part. But we still had to do marketing, sales support etc which we didn’t enjoy or excel at.
Now we’re both back in larger companies where we can focus on doing ML. It was satisfying to build a working business from scratch, no regrets, but I’m definitely happier now.
I've tried various headphones with "Spatial Audio", I have Airpods which do spatialization, yet none of this comes close to that barber shop.
Demo: https://notes.dt.in.th/HDRQRCode
Interestingly that one worked on iPhone, while the new emojis one doesn't
The challenge is getting them to run efficiently, which typically involves learning JAX.
- Original: https://www.youtube.com/watch?v=FYcMU3_xT-w&t=5s
- AI: https://www.openai.fm/#8e9915b0-771d-4123-8474-78cc39978d33
A good TikTok video gets "injected into your brain". You have zero effort to provide and suddenly this stuff is in your mind. I'm not saying it's a good thing, I realize the danger, but that's the core mechanism.
A friend in marketing describes this in terms of "brain calories". Eg if people have to think in order to understand your landing page, you failed to communicate your ideas efficiently, as it "requires too many brain calories". TikTok content requires zero brain calories.
One could say that only very shallow information could be spread this way (eg people dancing, video game clips) but I'm not sure that's true. The real challenge would be to turn an arbitrary source of information (wikipedia, hn) and make it immediately graspable. I suspect modern AI models could already go quite far in this direction.
Veritasium is a good example of interesting yet very graspable content: https://www.tiktok.com/@veritasium/video/7329576935317622058
> One prompt? Fair. 10? Still ok. 100? You're pushing it. 10M - get help.
Assuming you could do 10M+ LLM calls for this task at trivial cost and time, would you do it? i.e. is the only thing keeping you away from LLM the fact they're currently too cumbersome to use?
> LLMs are trained on much more than the whole Internet -- they also consume handcrafted answers produced by armies of highly qualified data annotators (often domain experts). Today approximately 20,000 people are employed full-time to produce training data for LLMs.
In a way this is a "simulated game engine", trained from actual game engine data. But I would argue a working simulated game engine becomes a game engine of its own, as it is then able to "propell the game" as you say. The way it achieves this becomes irrelevant, in one case the content was crafted by humans, in the other case it mimics existing game content, the player really doesn't care!
> An engine would also work offroad.
Here you could imagine that such a "generative game engine" could also go offroad, extrapolating what would happen if you go to unseen places. I'd even say extrapolation capabilities of such a model could be better than a traditional game engine, as it can make things up as it goes, while if you accidentally cross a wall in a typical game engine the screen goes blank.
Recently a travel agency used our platform to generate images of people in the destinations they were advertising.