• 0 Posts
  • 80 Comments
Joined 1 year ago
cake
Cake day: June 2nd, 2023

help-circle
rss
  • The problem is the hysteria behind it, leading people to confuse good sounding information with good information. At least when people generally produce information they tend to make an effort to get it right. Machine learning is just an uncaring bullshitting machine, that is rewarded on the basis of the ability to fool people (turns out the Turing test was a crappy benchmark for practice-ready AI besides writing poems), and VC money hasn’t reached the “find out” phase of that looming lesson, when we all just get collectively exhausted by how underwhelming the AI fad is.









  • Triple AAA games are usually very polished. But polish doesn’t make games fun. Polish is important with accessibility, and it’s easy to see why accessibility is important for a big studio casting a wide net.

    But fun? That comes from creativity and innovation. Big studios are averse to risk taking, and struggle to attract creative individuals, because the corporate culture seeks to stamp out individuality in the name of process and procedure.

    So yeah, more evidence of this. My money is going to Indy devs who prioritize fun over polish. (But polish is good to have too).











  • The wording of the article implies an apples to apples comparison. So 1 Google search == 1 question successfully answered by an LLM. Remember a Google Search in layspeak is not the act of clicking on the search button, rather it’s the act of going to Google to find a website that has information you want. The equivalent with ChatGPT would be to start a “conversation” and getting information you want on a particular topic.

    How many search engine queries, or LLM prompts that involves, or how broad the topic, is a level of technical detail that one assumes the source for the number x25 has already controlled for (Feel free to ask the author for the source and share with us though!)

    Anyone who’s remotely used any kind of deep learning will know right away that deep learning uses an order of magnitude or two more power (and an order of magnitude or two more performance!) compared to algorithmic and rules based software, and a number like x25 for a similar effective outcome would not at all be surprising, if the approach used is unnecessarily complex.

    For example, I could write a neural network to compute 2+2, or I could use an arithmetic calculator. One requires a 500$ GPU consuming 300 watts, the other a 2$ pocket calculator running on 5 watts, returning the answer before the neural network is even done booting.