全部科技产业动态

META THINKS THE NEXT BIG AI BREAKTHROUGH IS MEMORY. NOT JUST SMARTER MODELS TIL: Mark Zuckerberg and Meta's chief AI officer Alexandr Wang believe one of the biggest problems holding back personal AI agents is surprisingly simple. They keep forgetting things about you. Meta's new Muse Wonder model introduces something called Active Memory, designed to remember important details from conversations weeks or even months ago and bring them back when they're actually useful. And it's not just about giving AI a bigger context window. Wang says the model is being trained to manage its own memories, decide what information matters, and retrieve it when needed. Imagine an AI that remembers your family's preferences, what you discussed last month, and what you've been planning, then uses that information to handle things like shopping or organizing activities without making you explain everything again. But there's a much bigger reason Meta is spending so much effort on memory. Zuckerberg thinks personal agents could eventually become the main way people interact with the internet. And an AI that remembers your life could have a very different relationship with you than today's chatbots... [Read the rest of the story on Antfeed] #Meta #MarkZuckerberg #AlexandrWang #MuseWonder #Muse #ActiveMemory #AIAgents #ArtificialIntelligence #PersonalAI #AIModels #Superintelligence #AIInfrastructure #SiliconValley #인공지능 #메타 #마크저커버그 #알렉산더왕 #AI에이전트 #AI메모리 #장기기억 #초지능 $META $NVDA $MSFT $GOOGL

Nvidia 576 GPUs Rubin Ultra slide!

据媒体报道，由于DRAM严重缺货、价格持续上涨，DRAM平均售价（ASP）较第二季度上涨60%，南亚科技第三季度毛利率高达85.7%： 第三季度营收较第二季度增长61.9%，达1336.5亿新台币 第三季度营业利益率80% 第三季度净利润876.3亿新台币，超过过去7年总和 每股盈余（EPS）25.4新台币 南亚将于第一季度在新厂安装设备，目标在2027年底前达到每月3万片晶圆（wpm）产能。 南亚表示将在台湾屏东兴建先进封装厂 AI需求正在挤占其他产品获取DRAM的资源，因为AI相关企业愿意为芯片支付更高价格。部分客户正在下调DRAM规格以应对缺货。 长期供货协议（LTA）正成为DRAM行业的标准。

http://MakerMaker.AI co-founders Dhaval Adjodah and Owen He with a16z's Martin Casado on recursive self-improvement, and how to tell when it's real: A lot of RSI pitches are really just people using AI to build better AI. Tech has always worked that way. You needed computers to build chips. Steam engines got better once you could get coal faster. Dhaval and Owen's bar is higher. The model has to improve itself: its weights, its harness, even the prompt that tells it what to work on next. And the gains have to hold up when you count what they cost. They track everyone's experiment spend on a "bleeder board," and one engineer, the Viceroy of Variance, has the job of breaking their results. Over the past few months, they've started seeing early signs of their system getting better and cheaper at the same time. 0:55 Why most RSI pitches miss 2:25 Optimizer vs. optimizee 5:10 The RSI equation: performance over cost 6:50 Why a closed loop does not mean takeoff 8:40 Humans are agents inside the system too 9:25 The denominator trap: progress divided by capital raised 11:40 Training GPT-2 for $100 13:40 Most RSI papers fail when replicated 16:40 Why you have to co-optimize 17:40 Why inference is the first problem 20:15 The agent rewrites its own prompt 22:20 "It's not a belief. It's happening." 25:00 Why they haven't published 27:50 Intelligence is a vector and Amdahl's law 29:45 The garden hose theory of RSI YouTube: https://youtu.be/3Q1mdn5ZGvM @makermaker_ai @_dval_ @martin_casado

http://MakerMaker.AI co-founders Dhaval Adjodah and Owen He with a16z's Martin Casado on recursive self-improvement, and how to tell when it's real: A lot of RSI pitches are really just people using AI to build better AI. Tech has always worked that way. You needed computers to build chips. Steam engines got better once you could get coal faster. Dhaval and Owen's bar is higher. The model has to improve itself: its weights, its harness, even the prompt that tells it what to work on next. And the gains have to hold up when you count what they cost. They track everyone's experiment spend on a "bleeder board," and one engineer, the Viceroy of Variance, has the job of breaking their results. Over the past few months, they've started seeing early signs of their system getting better and cheaper at the same time. 0:55 Why most RSI pitches miss 2:25 Optimizer vs. optimizee 5:10 The RSI equation: performance over cost 6:50 Why a closed loop does not mean takeoff 8:40 Humans are agents inside the system too 9:25 The denominator trap: progress divided by capital raised 11:40 Training GPT-2 for $100 13:40 Most RSI papers fail when replicated 16:40 Why you have to co-optimize 17:40 Why inference is the first problem 20:15 The agent rewrites its own prompt 22:20 "It's not a belief. It's happening." 25:00 Why they haven't published 27:50 Intelligence is a vector and Amdahl's law 29:45 The garden hose theory of RSI YouTube: https://youtu.be/3Q1mdn5ZGvM @makermaker_ai @_dval_ @martin_casado