Comments by "Mikko Rantalainen" (@MikkoRantalainen) on "How I use LLMs" video.

  1. 48:20 What we would really need is ability to pass the response with all the provided references to another thinking + internet access AI system with a task "Does this article content match the provided references?". I'm pretty sure that different AI models do not accidentally hallucinate badly enough to fail this kind of verification task most of the time. If you can afford enough API requests to multiple high quality models, you could automate this kind of personal assistant for you. It might take an hour per task but it should emit really high quality and trustworthy responses.
    1
  2. I'm looking forward for a personal assistant AI that can actually remember long term stuff. Something like "watch me for a year to figure out what I already know" and then it can summarize the news feeds to only contain the stuff I don't already know. The same for all the other information I'm trying to intake into my brain. Typically I have to read a lot of stuff that I already know to learn one or two new things myself. This would basically require at least post-training step to be personally customized and the step should be periodically re-processed (e.g. daily or weekly) to stay current with my current knowledge. It optimal case it could take a video such as this one and show me a modified version that only contains the stuff that I don't already know but the new video would be re-rendered in a way to make it less jarring to watch than simply trying to automatically skip forward for all the redundant-for-me parts. In practice, the video-for-me version should appear to me as if Andrej Karpathy had created a pretty short video that only contains totally new information about AIs. Now I have to watch the whole video because I'm sure it contains some valuable information but I cannot know which before I watch the whole video.
    1