与「Wait_Me_There」相关的搜索结果

[#에이핑크#] #아는형님# 📺 [선공개] 팬들과의 추억을 담은 에이핑크의 신곡 〈Wait Me There (기억, 그 아름다움)〉♬ | 아는 형님 431회 #Apink#

[#에이핑크#] 지니 매거진에 “Apink (에이핑크)의 13주년 기념 ‘Wait Me There’ 녹음 현장 비하인드!” 가 공개되었습니다💜 지금 바로 아래 링크를 통해 녹음실 현장으로 떠나보세요! ▶ #Apink# #Apink_13th_Anniversary# #Wait_Me_There# #기억_그_아름다움#

[#에이핑크#] Apink 13th Anniversary Digital Single [Wait Me There (기억, 그 아름다움)]의 음원이 공개되었습니다. PANDA🐼들의 많은 사랑과 관심 부탁드립니다💕 🎞Music Video ▶ 🍈Melon ▶ #Apink# #Apink_13th_Anniversary# #Wait_Me_There# #기억_그_아름다움#

[#에이핑크#] Apink 13th Anniversary Digital Single ‘Wait Me There (기억, 그 아름다움)’ MV 🐼 #Apink# #Apink_13th_Anniversary# #Wait_Me_There# #기억_그_아름다움#

[#에이핑크#] Apink 13th Anniversary Digital Single ‘Wait Me There (기억, 그 아름다움)’ 🐼 2024.04.19 6PM (KST) #Apink# #Apink_13th_Anniversary# #Wait_Me_There# #기억_그_아름다움#

A few thoughts on the very near future First of all, what had previously been little more than a rumor has now been confirmed: GPT-5.6 had already been fully trained for two months and was available to selected users in early access. The obvious question is why it was not rolled out earlier. I do not think this was because OpenAI feared that the model might be overshadowed by Fable 5 or Mythos 5. Instead, OpenAI likely began working with government and regulatory authorities at a very early stage to ensure that the model could be released at all. Even after it had been previewed and announced, it still took some time before it could be rolled out publicly. That said, OpenAI clearly handled the rollout far better than Anthropic, which apparently did not have the same level of cooperation with government and regulatory authorities. Conversely, however, this also clearly means that future delays and increasingly strict model reviews will probably force us to wait longer for official releases. The next widely discussed rumor is that, within a few weeks, most likely no more than six, we will see either a preview or even the release of GPT-6. (Andrew Curran @AndrewCurran_ is one of the most reliable sources here on X, so I think that's very realistic.) The model has undergone entirely new pretraining, and the pace of releases is accelerating. The numbers are clear: Frontier labs are releasing more and better models at an increasingly rapid pace. Whereas we once had to wait months, quarters, or even half a year for major new releases, they are now arriving almost weekly. The latest frontier models may be more efficient in terms of intelligence per token, but they are also being deployed with much larger reasoning budgets. In practice, models such as Fable 5 and GPT-5.6 often consume considerably more tokens during complex or agentic tasks. This is not necessarily a sign of declining efficiency. Rather, it suggests that improvements in efficiency are being reinvested into deeper reasoning, longer trajectories and more capable agentic behavior. The result is that total compute consumption per task can continue to rise even as the underlying models become more efficient. Fable 5 and GPT 5.6 demonstrate just how intensive token usage has become. Although Sam Altman explicitly stated that GPT-5.6 is 54% more token-efficient (via CNBC), the fact remains that compute demand continues to increase, requiring more powerful and efficient computing infrastructure. Inference chips will probably become even more important as well. In summary, my initial conclusion from the latest releases is that compute demand will not merely continue to grow, but will probably exceed the available supply. This naturally means that energy demand will also increase, and, based on my initial assessment, probably more sharply than previously expected. This is likely to remain the largest bottleneck in the very near future. And this is important to me: there are bottlenecks. Not the training of the models, but besides compute, above all energy. This needs to be taken seriously! The US power grid, for example, is a major bottleneck, and the obvious question is how the necessary expansion can be achieved. Capital expenditure on data centers in the United States continues to rise sharply. This year, it exceeds 800 billion. It is not yet clear what the situation will look like in 2027, but I can hardly imagine investment declining or less CapEx being required. The reason lies precisely in the developments already mentioned: Demand is growing, particularly demand for energy. China clearly has an advantage here, a genuine moat, and I believe the West must be extremely careful not to fall behind because of the energy advantage China already possesses in practice. This could also help explain why, according to a recent Reuters report, China is considering restricting Western access to its frontier models. It may have concluded that it will win the long-term race. Unless there is a genuine breakthrough, whether in small modular nuclear reactors or fusion energy, I expect major problems to emerge over the coming years, for example by 2030. So far, I do not see any viable solutions. We can therefore clearly establish two points: Models are becoming larger, better, and increasingly useful for all users. There is no end to this development in sight. At the same time, the bottleneck appears to be growing increasingly severe, and this is already visible in practice. Regulation, energy demand, and compute demand could mean that, in the very near future, the release cadence will not accelerate as quickly as hoped or desired. This creates a clear contradiction. Thank you for coming to my TED Talk.

I was given early access to Grok 3 earlier today, making me I think one of the first few who could run a quick vibe check. Thinking ✅ First, Grok 3 clearly has an around state of the art thinking model ("Think" button) and did great out of the box on my Settler's of Catan question: "Create a board game webpage showing a hex grid, just like in the game Settlers of Catan. Each hex grid is numbered from 1..N, where N is the total number of hex tiles. Make it generic, so one can change the number of "rings" using a slider. For example in Catan the radius is 3 hexes. Single html page please." Few models get this right reliably. The top OpenAI thinking models (e.g. o1-pro, at $200/month) get it too, but all of DeepSeek-R1, Gemini 2.0 Flash Thinking, and Claude do not. ❌ It did not solve my "Emoji mystery" question where I give a smiling face with an attached message hidden inside Unicode variation selectors, even when I give a strong hint on how to decode it in the form of Rust code. The most progress I've seen is from DeepSeek-R1 which once partially decoded the message. ❓ It solved a few tic tac toe boards I gave it with a pretty nice/clean chain of thought (many SOTA models often fail these!). So I upped the difficulty and asked it to generate 3 "tricky" tic tac toe boards, which it failed on (generating nonsense boards / text), but then so did o1 pro. ✅ I uploaded GPT-2 paper. I asked a bunch of simple lookup questions, all worked great. Then asked to estimate the number of training flops it took to train GPT-2, with no searching. This is tricky because the number of tokens is not spelled out so it has to be partially estimated and partially calculated, stressing all of lookup, knowledge, and math. One example is 40GB of text ~= 40B characters ~= 40B bytes (assume ASCII) ~= 10B tokens (assume ~4 bytes/tok), at ~10 epochs ~= 100B token training run, at 1.5B params and with 2+4=6 flops/param/token, this is 100e9 X 1.5e9 X 6 ~= 1e21 FLOPs. Both Grok 3 and 4o fail this task, but Grok 3 with Thinking solves it great, while o1 pro (GPT thinking model) fails. I like that the model *will* attempt to solve the Riemann hypothesis when asked to, similar to DeepSeek-R1 but unlike many other models that give up instantly (o1-pro, Claude, Gemini 2.0 Flash Thinking) and simply say that it is a great unsolved problem. I had to stop it eventually because I felt a bit bad for it, but it showed courage and who knows, maybe one day... The impression overall I got here is that this is somewhere around o1-pro capability, and ahead of DeepSeek-R1, though of course we need actual, real evaluations to look at. DeepSearch Very neat offering that seems to combine something along the lines of what OpenAI / Perplexity call "Deep Research", together with thinking. Except instead of "Deep Research" it is "Deep Search" (sigh). Can produce high quality responses to various researchy / lookupy questions you could imagine have answers in article on the internet, e.g. a few I tried, which I stole from my recent search history on Perplexity, along with how it went: - ✅ "What's up with the upcoming Apple Launch? Any rumors?" - ✅ "Why is Palantir stock surging recently?" - ✅ "White Lotus 3 where was it filmed and is it the same team as Seasons 1 and 2?" - ✅ "What toothpaste does Bryan Johnson use?" - ❌ "Singles Inferno Season 4 cast where are they now?" - ❌ "What speech to text program has Simon Willison mentioned he's using?" ❌ I did find some sharp edges here. E.g. the model doesn't seem to like to reference X as a source by default, though you can explicitly ask it to. A few times I caught it hallucinating URLs that don't exist. A few times it said factual things that I think are incorrect and it didn't provide a citation for it (it probably doesn't exist). E.g. it told me that "Kim Jeong-su is still dating Kim Min-seol" of Singles Inferno Season 4, which surely is totally off, right? And when I asked it to create a report on the major LLM labs and their amount of total funding and estimate of employee count, it listed 12 major labs but not itself (xAI). The impression I get of DeepSearch is that it's approximately around Perplexity DeepResearch offering (which is great!), but not yet at the level of OpenAI's recently released "Deep Research", which still feels more thorough and reliable (though still nowhere perfect, e.g. it, too, quite incorrectly excludes xAI as a "major LLM labs" when I tried with it...). Random LLM "gotcha"s I tried a few more fun / random LLM gotcha queries I like to try now and then. Gotchas are queries that specifically on the easy side for humans but on the hard side for LLMs, so I was curious which of them Grok 3 makes progress on. ✅ Grok 3 knows there are 3 "r" in "strawberry", but then it also told me there are only 3 "L" in LOLLAPALOOZA. Turning on Thinking solves this. ✅ Grok 3 told me 9.11 > 9.9. (common with other LLMs too), but again, turning on Thinking solves it. ✅ Few simple puzzles worked ok even without thinking, e.g. *"Sally (a girl) has 3 brothers. Each brother has 2 sisters. How many sisters does Sally have?"*. E.g. GPT4o says 2 (incorrectly). ❌ Sadly the model's sense of humor does not appear to be obviously improved. This is a common LLM issue with humor capability and general mode collapse, famously, e.g. 90% of 1,008 outputs asking ChatGPT for joke were repetitions of the same 25 jokes​. Even when prompted in more detail away from simple pun territory (e.g. give me a standup), I'm not sure that it is state of the art humor. Example generated joke: "*Why did the chicken join a band? Because it had the drumsticks and wanted to be a cluck-star!*". In quick testing, thinking did not help, possibly it made it a bit worse. ❌ Model still appears to be just a bit too overly sensitive to "complex ethical issues", e.g. generated a 1 page essay basically refusing to answer whether it might be ethically justifiable to misgender someone if it meant saving 1 million people from dying. ❌ Simon Willison's "*Generate an SVG of a pelican riding a bicycle*". It stresses the LLMs ability to lay out many elements on a 2D grid, which is very difficult because the LLMs can't "see" like people do, so it's arranging things in the dark, in text. Marking as fail because these pelicans are qutie good but, but still a bit broken (see image and comparisons). Claude's are best, but imo I suspect they specifically targeted SVG capability during training. Summary. As far as a quick vibe check over ~2 hours this morning, Grok 3 + Thinking feels somewhere around the state of the art territory of OpenAI's strongest models (o1-pro, $200/month), and slightly better than DeepSeek-R1 and Gemini 2.0 Flash Thinking. Which is quite incredible considering that the team started from scratch ~1 year ago, this timescale to state of the art territory is unprecedented. Do also keep in mind the caveats - the models are stochastic and may give slightly different answers each time, and it is very early, so we'll have to wait for a lot more evaluations over a period of the next few days/weeks. The early LM arena results look quite encouraging indeed. For now, big congrats to the xAI team, they clearly have huge velocity and momentum and I am excited to add Grok 3 to my "LLM council" and hear what it thinks going forward.

the reason i'm so chronically online on X is that i owe my entire career to it. every job i've had has been through relationships i've made here, and as i hit 50k - a number i never thought i'd see on my profile - i can't help but feel a little emotional. you see, i started building up my profile when @wifelette and @tomdale took a chance on me many years ago - a nobody in melbourne who was still learning how to code in js - and invited me to speak at @EmberConf. i had never been to a conference at that point in my life, let alone speak at one. they found me through my tweets sharing my blogposts about learning the framework. i'm not exaggerating when i say that my life changed after doing that talk. public speaking was my greatest fear and i was visibly shaking for the first 5 minutes of my talk, but somehow i managed to remember everything i had practiced. that talk was how i got my first real job (and moved to the US) at @DockYard, thanks to @bcardarella also taking a chance on me. throughout this time i continued posting on twitter and building up my profile little by little. after you've done one talk, the next one is a little easier and so i kept doing more of it - seeing the world i had never been exposed to before. i got my job @netflix the same way, through doing a talk, and people taking a chance on me. some of you may be surprised to know that i never majored in CS and that i'm a self taught programmer. so going through a FAANG interview was inconceivable to me at that point in my life. i actually tried to back out of the interview, but they convinced me to do it anyway, and im glad that i did. i then got my job at facebook through Dan Abramov (i miss him here!!), who i met at a conference talk i did in London. we stayed in touch on twitter for a while before he reached out to me asking if i would be interested in working on the react team (uh, hell yes?! but wait, me??). massive imposter syndrome has followed me everywhere. now at @cursor_ai and @SpaceXAI, i feel so lucky that i get to work alongside such crazy talented, driven, and ambitious people who want to build beautiful products for everyone. and i'm also still in disbelief that this many people care about what i have to say here. thank you everyone for taking a chance on me, and thanks @X for being there for me this whole time 🤍

MAGIC MAN WORLD TOUR 2023 SOUTH AMERICA . 📍Brazil Such a crazy moment in Brazil 🇧🇷 Still remember a big part of my 20s was“ Hi Brazil !” And I couldn’t help but get chills when I say it standing in front of u all 🫠 The passion u all had & how deep u were in the moment with me meant everything to me Hope u enjoyed the show and took some thoughts about urselves home with ya I love u all and i had the greatest time💋 Also thanks to the legend @alokoficial for being there for me as a friend , family and my senior. Im honored to perform with u and thank u so much for hosting me and my team allowing us to experience #BRAZIL# 🇧🇷❤️ I can’t wait to say “HI BRAZIL” in front of u all again 🥹 Obrigado 🇧🇷 📸📹🎬 🇧🇷Brazil rafaelstrabelli ; alissondmphoto #MAGICMANWorldTour# #JacksonWangWorldTour# #TEAMWANGrecords# @teamwangofcl

Grok 4.6 is my daily driver now. This is why: When I’m using interactive agents in a CLI/terminal (or even Slack), what matters most to me is speed & token throughput. Intelligence and differences in performance on benchmarks is negligible to me when speed is sacrificed. If I am there to steer & guide the agent, or make some quick fixes, then I don’t want to wait 5+ min per turn. Opus 5 (fast) is a good solution to this, but it is WAY too expensive. Grok 4.6 is perfect though, incredibly fast, almost as smart as Fable, and relatively much cheaper than anything else right now. On the flip side, for cloud agents working in Factories & background tasks triggered on crons, I don’t mind using intelligent, slow models like Fable / Sol. I’d even opt for a model router for cloud tasks that biases to heavy, smart models. I guess the meta point I’m making here is that model choice is largely a function of whether there’s a human waiting in the loop.