I played a video game for a few weeks I think I started in order to get out of spasmodic states filled with muscle contractions, and something supported it then I found I could move the game near work and get a little work done and that's basically how I got the most recent work done on my kfsn4 boards, my coreboot branch for which displays serial logs, woohoo. both boards now broken

i'm remembering that I built qemu out of serialice's tree, just as a general emulator. I could instead install the qemu package. this is likely newer. nice to update part of a plan.

i'm confused that nobody is working at the office, but the intern's neighbor continues working at the desk it is two different realities. in one, nobody is working. in another, somebody is.

god mind controlled a bird on this property every day, for almost the whole day, it flies into the windows, while i'm indoors it is to remind me that I yearn to go outdoors the bird takes breaks to eat worms

I hear them flying against a different window.

I imagine reporting a bug from a compromise and then a mean entity walking up to me and being like "you know that is from the compromise. don't make waves." and then hurting me badly. but I guess it isn't actually like that. hard and rare to think about.

regardless when I am super confused it is generally nice to relax by troubleshooting like adding small numbers, but more useful feeling. I tend to troubleshoot really small things.

Karl Semich's Resume Why I Want To Work At MindControlledBusiness I am a really great worker. I love being mind controlled. I direct all my disgusting complaints to an online mailing list and keep them out of the office. I love buying office supplies! I have new clean shoes. Work History 2020-2022 Spamming a mailing list of hackers. I used to write computer programs all day so I am very qualified to interact with technical people. 2018-2020 Undergoing treatment for cancer and psychosis. I spent a year having chemotherapy and can relate well with people dying of terminal illnesses. I'm also extensively skilled at chatting with people completely disconnected from reality. 2016-2018 Waiting for my truck to be repaired. I have the patience of mountains. Some of the above items are listed out of order. I apologise for this and attach $200 in up-front wages in the hope to help make it right. References: coderman <coderman@gmail.com> douglas lucas <dal@riseup.net> zig the n.ig <ziggerjoe@yandex.com>

---- How to play chess: Generative Text Model: "How do u play chess" Human: "Chess is played on an 8x8 square checkerboard with 6 unique piece types: for each side there is the queen, the king, two bishops, two knights, two rooks, and eight pawns. One side has black pieces and is called black; the other has white pieces and is called white." Generative Text Model: "ok let me update my prompt so i'll talk as if I know that" Generative Text Model: " !addtoprompt $last_user_message" Generative Text Model: "How do u play chess" Human: "You put a space before your command. Redo the last exchange to accommodate that error" Generative Text Model: " !sendmessage oops thank you" Generate Text Model: "How do the pieces move?" Human: "Each piece type moves a different way."

this says it drops parameters by 25x using retrieval: https://www.deepmind.com/publications/improving-language-models-by-retrievin... if we're stuck copying work we bump into, then it makes sense to figure out what's cited that paper (or bump into something else)

quote from newer paper: In this study we show that training stability can be achieved with fewer sacrifices on model effectiveness. There are other pathways to improving LLM efficiency. One is to augment a language model with a retrieval component to fetch external knowledge that is useful to perform downstream tasks. So, the size of the language model can be significantly reduced since it does not need to encode ev- erything in model parameters (Guu et al., 2020; Khandelwal et al., 2020; Borgeaud et al., 2021; Gui et al., 2021; Zhang et al., 2021). With sparse model structures, Mixture of Experts models (Artetxe et al., 2021; Fedus et al., 2021; Zuo et al., 2021; Zoph et al., 2022) adaptively activate a subset of model parameters (experts) for different inputs during model training and inference. The METRO method proposed in this paper is orthogonal to retrieval-augmented models and sparsely activated models. Their combination is an interesting future work direction. This next one is on training a human assistant using feedback and reinforcement learning. Nice to have somebody actually talk about that. https://arxiv.org/abs/2204.05862