A large part of these anthropomorphic narratives was pushed by SV nerds to grab shareholder attention.
LLMs are transformative, but a lot of the tools around them already treat them as opaque function calls. Instead of piping text to sed, awk, or xargs, we’re just piping streams to these functions. This analogy can stretch to cover audio and video usage too. But that’s boring and doesn’t explain why you suddenly have to pay more for Google Work Suite just to get bombarded by AI slop.
This isn’t to undermine the absolutely incredible achievements of the people building this tech. It’s to point out the absurdity of the sales pitch from investors and benefactors.
But we’ve always been like this. Each new technology promises the world, and many even manage to deliver. Or is it that they succeed only because they overpromise and draw massive attention and investment in the first place? IDK, we’ll see.
That got me thinking - what if we stripped that conversational sound-like-a-human-be-safe layer and focused RLHF on being best at transforming text for API usages?
The early OpenAI Instruct models were more like that. The original GPT-3 was only trained to predict the next token, then they used RLHF to make them interpret everything as queries, so that "Explain the theory of gravity to a 6 year old." wouldn't complete to "Explain the theory of relativity to a 6 year old in a few sentences." ChatGPT was probably that, expanded to multi-turn conversation. You can see the beginnings of that ChatGPT style in those examples.
If you don’t aim for the moon, you’re gonna surely miss it. The number of orbital calculations you need to get right to land on the moon is bonkers /s.
If either of you did, you'd know it's false. Getting to the Moon (and back) by eyeballing the trajectory is a rite of passage for every KSP player.
You launch for standard equatorial low orbit, then coast until the Moon is about 90 degrees to the right / behind you, and burn ahead until you run out of fuel in the second stage. This turns your orbit into an elongated ellipse, and as you get close to its far end, the Moon will catch up and capture you in its gravity well.
When playing one of the non-sandbox aka. "progression" modes, i.e. "career" or "science", the trajectory planning tools are locked behind an expensive upgrade, effectively forcing you to do your first Moon fly-by this way.
LLMs are transformative, but a lot of the tools around them already treat them as opaque function calls. Instead of piping text to sed, awk, or xargs, we’re just piping streams to these functions. This analogy can stretch to cover audio and video usage too. But that’s boring and doesn’t explain why you suddenly have to pay more for Google Work Suite just to get bombarded by AI slop.
This isn’t to undermine the absolutely incredible achievements of the people building this tech. It’s to point out the absurdity of the sales pitch from investors and benefactors.
But we’ve always been like this. Each new technology promises the world, and many even manage to deliver. Or is it that they succeed only because they overpromise and draw massive attention and investment in the first place? IDK, we’ll see.