与「M・A・R」相关的搜索结果

Dans le manifeste "techno-optimiste" de Marc Andreessen, il y a une phrase qui m'a marqué : "Our enemies are not bad people – but rather bad ideas." Nos ennemis ne sont pas des mauvaises personnes. Ce sont des mauvaises idées. Prenons Jancovici. L'homme est brillant, sincère, travailleur. Il ne se lève pas le matin en se disant qu'il va nuire à l'humanité. Mais l'idée qu'il porte la décroissance, le rationnement, la frugalité érigée en horizon civilisationnel est une idée profondément destructrice. Elle prend des esprits brillants et les transforme en commissaires politiques d'un futur appauvri. Et le plus fascinant, c'est ce que cette idée fait aux gens qui l'adoptent. Dans mon entourage, une grosse partie de mes amis est sur cette ligne décroissantiste, avec tout le package qui va avec. L'argent c'est mal mais ils en veulent. Il faut moins prendre l'avion mais ils rêvent de voyager partout. Il faut consommer moins mais ils ne renoncent à rien de ce qu'ils aiment vraiment. Et tous ont un point commun : ils sont déprimés. L'un d'eux m'a même confié qu'il était sous antidépresseurs. Ce n'est pas un hasard. C'est mécanique. Quand tu crois que ton désir de vivre, de créer, de t'élever est moralement suspect tu te détruis de l'intérieur. Tu passes ta vie à t'excuser d'exister. Tu vis dans la dissonance permanente entre ce que ton corps veut (plus, mieux, plus loin) et ce que ton idéologie t'ordonne (moins, sobre, immobile). D'où ma théorie : Quand on pense quelque chose de fondamentalement faux décroissance, communisme, extrémisme religieux (de tout ordre) ce n'est qu'une question de temps avant que ça devienne vraiment destructeur. D'abord pour soi. Puis pour les autres. Les mauvaises idées tuent. Lentement chez ceux qui y croient, brutalement chez ceux qui les subissent. C'est pour ça que la bataille des idées n'est pas un luxe d'intellectuel. C'est la bataille la plus importante de notre époque.

Drake disses LeBron James and Kendrick Lamar on “1AM in Albany” “I shouldn't even be shocked to see you in that arena, because you always made your career off of switching teams up” “Please stop asking what’s going on with 23 & me, I’m a real ni—a, and he’s not, it’s in my DNA” “Muggsy Bogues dunked for once, even I'm a bit amazed, someone give the kid a raise”

Nicole: Clamor Within The Silent "Mage," the Voiceless "Angel" #GenshinImpact# #Nicole# Many of the stories Nicole envisions begin with a phrase as old as stories themselves: "Long, long ago..." Perhaps, for some authors, "long ago" is less the threshold of a tale than a refuge from all further questioning. For once a story is consigned to unreachable distance, all the people and events bound up within it may as well have faded into obscurity, leaving no path by which the present might truly follow them back. And yet, in Nicole's thoughts, those stories of "long ago" endure still. Long ago, the lights of many nations filigreed the earth in gleaming gold. Long ago, the daughters of heaven wandered in carefree grace betwixt divine courts and mortal cities. Long ago, the sovereign of ages past, who had descended into deepest darkness, had not yet brought disaster back to his homeland. Long ago, the three radiant moons that hung aloft in the night heavens were still three... The years brook no pause, the wheel spinning onward all the same. Within a cycle emptied of hope, the spark of paradise perished alongside the lives annihilated. Servitors once loyal to their ruler turned traitor for the sake of the lives that had been fashioned. The black dragonlord returned from beyond the stars. The moonlit chariot of the heavens shattered like crystal... The story had not reached its end. Rather, it was those who had once read it, and those who had once listened to its telling, who had long since taken their leave. The angel who once longed to write laughter into the world lost all that had once made laughter possible. In the end, she even lost the voice with which she laughed, and could speak no more. Thus, it seemed that the story Nicole kept within her would never properly find its beginning. She could only go on cradling her many "long agos," roaming the world and watching as all things passed, changed, and were remade by time. That remained so until the day she met a tiny mage in an enormous hat, its extravagant brim tipped up with such pride it seemed almost ready to brush the sky. Compared with the hat she wears now, however, it was still a fair few sizes smaller. The little witch tucked herself down at Nicole's side and listened intently as one story after another unfolded, not stirring until she had finished her drink and emptied her bag of biscuits. "So you really are an angel? And all those stories were real, too?" "Then yay, that works! I'm a mage, and you're an angel, so we match perfectly." The merry little mage stretched out a hand to the angel who had come so far, all the way from those distant "long agos," and warmly invited her along. "Want to join my Hexenzirkel? I only just came up with the name, but it has a real ring to it, don't you think?" "It's going to be the funnest, awesomest, best club in all of Teyvat — I promise!" "..." No answer came from the angel. None could. But the little mage did not wait for one. She caught the angel's hand in both of her own and slipped the last piece of candy into her palm. "There. Now our mages' pact is made." Nicole had no thought of joining then, nor could she have guessed what this little gathering called the "Hexenzirkel" would one day grow into. But at the very least... It did sound like a place with happiness enough to spare.

When coming to Hangzhou we shall leave as Bai Suzhen🐍 The legendary white snake! What? You tell me because I'm a rabbit tank I can only be "white anaconda"? Ja vielen Dank auch... 🙄

I was so touched reading comments under the video clip of Xiao Zhan eating ice cream on Douyin that I decided to translate some of them. Just like the way #XiaoZhan# sends “postcards” to XFXs on his every trip. XFXs also confides in him like a brother or a family member. (Part 1. I put a few more comments under part 2) 💌 Zhan Zhan, I want to share with you a bit about my life. My recent situation is not very good. Maybe that's why when I listen to the background music, tears suddenly flow. Family pressure, and a small child... I don't know if my choice a year ago was right? I used to have work and colleagues. But now I've given up my job and become a full-time mother. Maybe I am not a qualified mother to say this. When I was a child, I was carefree and wanted to grow up quickly. Now I long to live those carefree days. Okay, no more complaining. Let's move forward together, you also. 🌹 💌 Zhan gege, retaking the exam is really tiring 😭. I will definitely pass in 2025, right gege? 💌 Zhan Zhan, I have been trying to get pregnant for many years without success. Next month I will try IVF. I am a bit scared, but I also hope to have a baby smoothly. 💌 Gege, should I take the postgraduate entrance exam? I can't make up my mind. 💌 I live alone in a foreign country. Watching Xiao Zhan's vlog and reading XFXs’ sincere messages moved me to tears ❤️ 💌 Zhan gege, today is my birthday. Today I am 21 years old 😆. I will dedicate my birthday wish to you. Guess what I wished for, I will tell you quietly. I wish that my Zhan Zhan will have a smooth and healthy life. 💌 Gege, I just submitted my graduation thesis. I'm looking for a job now, and feeling so much pressure. I don't know where my next journey in life is going. My college years are almost over, and I feel sad. But when I see you abroad, it really feels healing. It's like I’m breathing a moment of free time with you, and seeing my carefree self in a parallel world 😭 💌 Good evening, Zhan Zhan. I just got back from Beijing three days ago. I'm a little anxious and confused about what kind of job to look for next. 💌 Gege, let me show you today’s sunset at my place. 💌 Zhan gege, I've been so busy lately. Every day I'm overwhelmed with work. I feel like I'm about to collapse. Watching the video and listening to this bgm, in this moment, I suddenly miss you so much. 😭 💌 Zhan gege, I am about to get married, but I am still so confused. I don’t know if I can be a good wife, or a qualified mother in the future... I always have an inexplicable fear of the unknown future... 💌 Zhan Zhan, actually I have been very tired lately. Last week I accompanied my dad to Guangzhou for surgery. Last week my mom was hospitalized and I also had to accompany my dad to do dialysis. Today I had to take the blame for someone at work. Tonight after helping my boy with his homework, I watched your video as soon as he went to bed. The vlog felt so warm and beautiful. The feeling of helplessness recently made me want to cry but couldn't. I've long considered you a relative, a very close friend in my heart. So when I saw you, I naturally shed tears unconsciously. Let me confide my feelings with you in this comment section. Thank you for your presence and comfort. I will try to cheer up and continue to work hard to live my life well. I also wish you all the best, health, safety, happiness and worry-free. 💌 I want to eat that ice cream. Marriage is bitter. I don't want to try it in my next life. 💌 Zhan Zhan, I will retire in 5 years. I hope you can hold a concert so I can go see it. I've known you since 2019. I really want to see you at least once. 💌 Rewatching this clip made me want to cry, even though I was so happy watching the vlog earlier at noon. Seeing you eating like a kitten, you must be happy. You were so self-disciplined that you only ate one glutinous rice ball. But you bought your favorite ice cream and ate it slowly.

# RLHF is just barely RL Reinforcement Learning from Human Feedback (RLHF) is the third (and last) major stage of training an LLM, after pretraining and supervised finetuning (SFT). My rant on RLHF is that it is just barely RL, in a way that I think is not too widely appreciated. RL is powerful. RLHF is not. Let's take a look at the example of AlphaGo. AlphaGo was trained with actual RL. The computer played games of Go and trained on rollouts that maximized the reward function (winning the game), eventually surpassing the best human players at Go. AlphaGo was not trained with RLHF. If it were, it would not have worked nearly as well. What would it look like to train AlphaGo with RLHF? Well first, you'd give human labelers two board states from Go, and ask them which one they like better: Then you'd collect say 100,000 comparisons like this, and you'd train a "Reward Model" (RM) neural network to imitate this human "vibe check" of the board state. You'd train it to agree with the human judgement on average. Once we have a Reward Model vibe check, you run RL with respect to it, learning to play the moves that lead to good vibes. Clearly, this would not have led anywhere too interesting in Go. There are two fundamental, separate reasons for this: 1. The vibes could be misleading - this is not the actual reward (winning the game). This is a crappy proxy objective. But much worse, 2. You'd find that your RL optimization goes off rails as it quickly discovers board states that are adversarial examples to the Reward Model. Remember the RM is a massive neural net with billions of parameters imitating the vibe. There are board states are "out of distribution" to its training data, which are not actually good states, yet by chance they get a very high reward from the RM. For the exact same reasons, sometimes I'm a bit surprised RLHF works for LLMs at all. The RM we train for LLMs is just a vibe check in the exact same way. It gives high scores to the kinds of assistant responses that human raters statistically seem to like. It's not the "actual" objective of correctly solving problems, it's a proxy objective of what looks good to humans. Second, you can't even run RLHF for too long because your model quickly learns to respond in ways that game the reward model. These predictions can look really weird, e.g. you'll see that your LLM Assistant starts to respond with something non-sensical like "The the the the the the" to many prompts. Which looks ridiculous to you but then you look at the RM vibe check and see that for some reason the RM thinks these look excellent. Your LLM found an adversarial example. It's out of domain w.r.t. the RM's training data, in an undefined territory. Yes you can mitigate this by repeatedly adding these specific examples into the training set, but you'll find other adversarial examples next time around. For this reason, you can't even run RLHF for too many steps of optimization. You do a few hundred/thousand steps and then you have to call it because your optimization will start to game the RM. This is not RL like AlphaGo was. And yet, RLHF is a net helpful step of building an LLM Assistant. I think there's a few subtle reasons but my favorite one to point to is that through it, the LLM Assistant benefits from the generator-discriminator gap. That is, for many problem types, it is a significantly easier task for a human labeler to select the best of few candidate answers, instead of writing the ideal answer from scratch. A good example is a prompt like "Generate a poem about paperclips" or something like that. An average human labeler will struggle to write a good poem from scratch as an SFT example, but they could select a good looking poem given a few candidates. So RLHF is a kind of way to benefit from this gap of "easiness" of human supervision. There's a few other reasons, e.g. RLHF is also helpful in mitigating hallucinations because if the RM is a strong enough model to catch the LLM making stuff up during training, it can learn to penalize this with a low reward, teaching the model an aversion to risking factual knowledge when it's not sure. But a satisfying treatment of hallucinations and their mitigations is a whole different post so I digress. All to say that RLHF *is* net useful, but it's not RL. No production-grade *actual* RL on an LLM has so far been convincingly achieved and demonstrated in an open domain, at scale. And intuitively, this is because getting actual rewards (i.e. the equivalent of win the game) is really difficult in the open-ended problem solving tasks. It's all fun and games in a closed, game-like environment like Go where the dynamics are constrained and the reward function is cheap to evaluate and impossible to game. But how do you give an objective reward for summarizing an article? Or answering a slightly ambiguous question about some pip install issue? Or telling a joke? Or re-writing some Java code to Python? Going towards this is not in principle impossible but it's also not trivial and it requires some creative thinking. But whoever convincingly cracks this problem will be able to run actual RL. The kind of RL that led to AlphaGo beating humans in Go. Except this LLM would have a real shot of beating humans in open-domain problem solving.

𝙕𝙚𝙣𝙡𝙚𝙨𝙨 𝙕𝙤𝙣𝙚 𝙕𝙚𝙧𝙤 𝙉𝙞𝙘𝙤𝙡𝙚 𝘿𝙚𝙢𝙖𝙧𝙖🐰💗 #绝区零   ##ゼンレスゾーンゼロ# #ZZZ# #zzzero   ##NicoleDemara#

𝙕𝙚𝙣𝙡𝙚𝙨𝙨 𝙕𝙤𝙣𝙚 𝙕𝙚𝙧𝙤 𝙉𝙞𝙘𝙤𝙡𝙚 𝘿𝙚𝙢𝙖𝙧𝙖🐰💗 #绝区零  ##ゼンレスゾーンゼロ# #ZZZ# #zzzero  ##NicoleDemara#

𝙕𝙚𝙣𝙡𝙚𝙨𝙨 𝙕𝙤𝙣𝙚 𝙕𝙚𝙧𝙤 𝙉𝙞𝙘𝙤𝙡𝙚 𝘿𝙚𝙢𝙖𝙧𝙖🐰💗 #绝区零 ##ゼンレスゾーンゼロ# #ZZZ# #zzzero ##NicoleDemara#