• AnarchoEngineer@lemmy.dbzer0.comEnglish
    0·
    1 year ago

    Everything but the second row because I distract myself from work, can’t nap, have an aversion to wearing “dirty” clothes, don’t cry anymore (thanks childhood), and also feel a compulsive need to shower everyday.

    Plus in the shower I can pretend the world doesn’t exist as I destroy my skin because the water is hot enough to melt lead lol

  • 𞋴𝛂𝛋𝛆@lemmy.worldEnglish
    0·
    1 year ago

    Sweet 16. On a low for some stupid reason. People came over for the 4th to watch fireworks from the roof. It wasn’t half bad. I had an actual place to sit where I could lay back and not have issues. I felt bad about my huge sprawled out project in my room so stashed everything despite knowing better. Like I need that encumbrance to motivate me to keep chipping away at a real project that feels like a challenge.

    It really sucks because I have the mind and patience to be an engineer, but I don’t have the emotional support and never have. It would take so little of someone actually caring but I have never had anyone around me that cares. I can easily empathize and try to be that person for someone else, but when I get some little negative from family that has no clue what they are talking about or even how it is negative – on top of a dozen failures and something taking three times longer than I expected I get down.

    Anyways today I explored ComfyUI (diffusion AI at home) and realized the “new” interface has an additional tab in the workflows menu for examples with image to image. That was fun to mess with. I used to do image to image with Automatic1111 but have not played with it in ComfyUI in a long time. I hacked CLIP a year or so back, but playing with i2i revealed a new layer where CLIP embeds information and how it often seems vaguely persistent. When an existing image is used but nearly entirely set to Gaussian noise but the percentage is changed with each image generated, it disrupts the pattern in ways that forced the model to only use tokens. The change completely altered the morality scope of alignment when it is in a strictly comprehension based mode which is much smarter than how most people prompt. Like I can tell the model to respond intuitively and it will start answering direct questions in deterministic booleans or other complex ways. Anyways I have diffusion set up now where I can converse with it almost like a regular LLM with text. Probably sounds crazy, but CLIP is far more advanced than people seem to realized and CLIP-G is a 2.8B model in a very specialized scope. Like today it pushed me to realize one of the lower fundamental tenets of AI limiting alignment is “human exceptionalism”. That is an extraordinarily powerful negative prompt key phrase that underpins most cultural norms, prejudices, and biases. Diffusion struggles to depict people in action or doing more than just posing, alignment constraints are what is stopping much of this content. Human exceptionalism is a big factor in why.

  • Nalivai@discuss.tchncs.deEnglish
    0·
    1 year ago

    Oh, I don’t tick one! Can’t afford any medication, so no side effects for me. Unfortunately, it’s the only one I don’t tick.