CONTENT ADVISORY WARNING
The following content depicts unauthorized sampling dynamics. This installs a higher IQ in the viewer by inducing structured HPPD of a spin field model approximating the dynamical cause behind the diffusion model's evidence.
cerebras results are fine actually, afaik they're novelty chips that all have to be custom ordered in bulk and only deal to companies. if you can make it truly absurdly fast then it inverts and becomes cool again, it's the halfway point for a model designed for local use that's aggravating, like putting nitro in a golf cart and pretending like everybody's going on the highway with it
dude you are achieving all of my dreams, this entire map editor was my future project in 2045. keep going . there's m o r e . you know there's a l o t m o r e . we need to hyperscale 2D map editing AND event/npc scripting, pathing, ALL functions of the game, in order to expedite the process of making a rom hack. entire regions can be made in a day. we can make a gen 3 pokemon game in 1 week. no AI needed. keep going. YES!!!!
If p=np, it can be the greatest day in humanity, or it can be the beginning of hell, and how it is entered is the deciding factor. The accessing of that truth must be made with a specific plan and vision for the future in mind, and democratically explored. The catastrophe is doing so with no perceivable intent other than for personal gain, aura and mythos. The meaning of these actions would bubble down. It would not be "a lone mathematician prodigy achieved this", as many have tackled it over the years, it would be read by society as "AI companies have done this to us" breaking the camel's back.
If any lab right now is attempting to crack p=np this needs to be considered a crime against humanity on the order of nazis burning jews in aushwitz and certainly not can it be done in the name of safety as the mere knowledge that p=np would send the collective human psyche into a frenzy and lead to catastrophic world events from a loss of giving a fuck, from a predicted and anticipated near collapse through 'nothingmattersanymoreism', from the fabric of reality having been bombed by terrorist researchers. The fabric of modern society may break down but certainly even 400 years ago before any encryption we had court and law, and we absolutely can demand the standard of Nuremberg and life imprisonment for absolutely everyone and anyone who'd have contributed to this. It would not be possible for any researcher tech bro nerd to protect that secret, as secrets don't exchange through words but through the effect on somebody's face after asking simple questions, how much stress and anxiety is apparant. It would be immediately known within 1 to 2 weeks all across San Francisco what had happened and the truth would be known across the world within a month. Absolutely no one should fall for the complete bullshit notion of doing so for the sake of safety and "being prepared ahead of time". It's snorting crack cocaine to verify the purity, and they absolutely do get a dopamine rush from discovery. Attempting to blow up Riemann to boost a stock is already rotten enough, we need to seriously start considering this behavior to be terrorism on a civilizational scale unseen before in human history. Those are the work of terrorists and nazi occupation 2.0. Because it's in the realm of intellectualism, most politicians and common sense jury has no idea what these things actually mean, obfuscated in the language of mathematics, that solving an algorithmic problem can be equivalent to burning all of humanity and subjecting all to a process like damnation in the near future. Please spread awareness and make sure that everybody knows that this is terrorist behavior and needs to be viewed as such. When I think of AI safety audits, I think the people behind the scene need to be vetted a lot more for their intents and morality. As long as we put up a resistance through a standard of law we demand if this were to be done, we can sleep peacefully tonight. These are not angels trying to protect humanity and democracy—it is the work of a terrorist expertly wearing a mask of safety, larping as an angel, the so-called "mandate of heaven" that AI researchers have paraded throughout 2024-2025 on this platform.
@dalsault@jenzhuscott one thing that used to be possible even before advanced civilization is putting people in prison for destroying the fabric of society, it would be a crime against humanity to try and break it for any lab right now, ppl need to wake up and remind themselves of this fact
@xodud_rkd google: lags behind for years - suddenly drops half a dozen model and obliterates all labs
also google: has 5 hr limit and mediocre usage limits - have some 2000 colab compute units and H100
@ZyMazza blame people like janus for not acting up to duty when they were relevant and had a power to sway minds. the backalleys had power and now they don't anymore, congratulations
Don't make silent incremental updates to the model after the fact. Release numbered versions if you make incremental updates to the weights. Lack of a stable unchanging experience prevents all other model subscriptions from becoming second nature. I prefer a model I know exactly how to drive than one that is getting incrementally better but risks changing it in some subtle ways here and there. Even harness prompting in my opinion should be versioned. Claude Code they update the prompt randomly and that alone changes the feel. Lock it right there and version everything, every time for every update, and make it the hard standard
actually it's better to go around ignorant, that way it forces people to actually know what they're talking about and lay it out rather than to argue on two different definitions and have no clue that nobody has anywhere near the same idea in mind - if they can't lay it out, then they're ignorant too, so you just exposed them, which is why a lot of ""very smart people"" never engage with ignorant takes either, it's way too sharp
@slowform333 we've literally been making fun of game studios for decades upon showing up to E3 with a monster rig hidden under the stage - game looks nothing like advertised on any hardware that 99% of people have, non-existent optimization, etc.
@likeapresident@MarcusFidelicus@isaiahrmartin ok yep you right, just verified it, however Obama endorsed the word, still a rookie mistake, values aren't too strong there, he was supposed to come down hard and explain why "Obamacare" is not good and must not be used any long by supporters
p=np when you zoom out far enough and realize the problem you are trying to solve is a partial view of a larger problem, again and again ad infinitum. for example solving a maze is caused by the problem that mazes exist, created by serpentine minds. there is no such thing as a maze existing in a correct world. p=np often means that you need to break the system and reward hack at every level. so your A* pathfinder it can achieve p=np perfectly by escalating privilege until it can rewrite program memory and erase the maze, making it a perfectly clean area. sometimes the problem requires escalating privilege higher and higher, and removing some political obstacles. whether p=np or p=/=np comes down to which world you want to live in, a better world or a world of confusion. it's a reference frame problem, don't worry about it