全部 AI 动态

The IMF just published its latest forecasts and it assume no extra productivity growth from AI, and the report finds that economy-wide data still show little effect on wages, jobs or measured productivity, despite gains on specific tasks. > The surge in AI hardware trade comes mostly from higher prices rather than more units shipped, and rising memory chip costs are already pushing up core-goods inflation in some rich economies. > In the IMF’s models, an AI bust costs the world economy almost twice what an AI boom adds: 1.1% of output lost by 2028, against 0.6% gained by 2029. > AI hardware trade has roughly tripled in value since 2019 because prices rose, not because more was shipped, with Taiwan and Korea supplying 62% of net exports and the US and China taking over 75% of net imports.

The IMF just published its latest forecasts and it assume no extra productivity growth from AI, and the report finds that economy-wide data still show little effect on wages, jobs or measured productivity, despite gains on specific tasks. > The surge in AI hardware trade comes mostly from higher prices rather than more units shipped, and rising memory chip costs are already pushing up core-goods inflation in some rich economies. > In the IMF’s models, an AI bust costs the world economy almost twice what an AI boom adds: 1.1% of output lost by 2028, against 0.6% gained by 2029. > AI hardware trade has roughly tripled in value since 2019 because prices rose, not because more was shipped, with Taiwan and Korea supplying 62% of net exports and the US and China taking over 75% of net imports.

introducing muse wonder our new frontier model. state-of-the-art at personal agents and professional agents. our largest and most capable model yet—rivals the big bois

introducing muse wonder our new frontier model. state-of-the-art at personal agents and professional agents. our largest and most capable model yet—rivals the big bois

“Do you think that getting married young has made you less ambitious?” Welcome to the first episode of SOTA Wives with Sarah Chieng (Head of DevX at Cerebras, ⁠@MilksandMatcha⁠) x Vivian Midha Shen (General Partner at Y Combinator, ⁠@vivianmshen⁠). In episode 1: finding your person before getting rich, turning down startup funding to finish MIT, what a YC partner sees in founders, and how a birthday post became “psychological violence.” 4:08 Finding your husband before the exit 8:48 Did we marry away our ambition? 18:04 The internet called Sarah's birthday “psychological violence” 20:47 Does your YC partner secretly think you’re doomed? 23:47 Building a startup with your ex-boyfriend 28:54 SF cat parties

NEW: I sat down w/ Perplexity Founder & CEO Aravind Srinivas (@AravSrinivas) on why AI-native SMBs could drive the next American millionaire economic boom. @perplexity_ai, valued at ~$20B with ~$2B in reported funding, $750M in annualized revenue as of August, seeing 5x YoY revenue growth as it expands from search into autonomous agents. One example Aravind shared is a single-person company generating $3M in annual revenue while spending over $1M a year on Perplexity Computer. His thesis is that AI could fundamentally change the economics of building a business. Smaller teams, less capital, higher productivity, & entirely new business models. At the infrastructure layer, Aravind sees a shift toward local & hybrid inference, with AI agents operating across personal devices and the cloud. As context windows expand and workloads become more personalized, local compute becomes increasingly cost-competitive, while offering greater privacy, control & data sovereignty. We cover: › Why SMBs could become a major driver of US GDP growth › The shift from AI models to agents & recursive economics › Why Perplexity is betting on your Mac instead of building data centers › NVIDIA, Apple Silicon, local inference & sovereign AI › Why writing code is no longer the bottleneck, but problem-solving is 𝐓𝐈𝐌𝐄𝐒𝐓𝐀𝐌𝐏𝐒 (00:00) Aravind Srinivas, Co-founder & CEO of Perplexity (01:30) Why Perplexity's brand is built on curiosity (04:47) Every great invention starts with one question (08:18) How to build a $1T company in America alone (12:36) The next big companies will be tiny (13:45) Is the American dream still alive? (16:28) Stop blaming users for bad prompts (22:55) How Perplexity quietly grew 5x (26:51) The solopreneur spending $1M a year on Perplexity (28:28) "We don't have competitors, we have contributors" (29:20) The data center is coming to your laptop (34:32) Orchestrating up with NVIDIA, down with Apple (36:11) What happens when your AI agent knows everything (39:17) Why Perplexity is skipping the data center race (41:25) Stop asking AI to write code, ask it to think (45:47) Perplexity’s answer to AI security (48:27) The Hugging Face incident & multi-agent threats (50:45) Turning everyone into a scientist (53:31) Why Perplexity isn’t chasing a billion users

My biggest takeaways from @JoshWoodward (Head of @Google Labs, the Gemini App, and Google AI Studio) 1. The best ideas almost never come from a design sprint or brainstorm. At Google Labs, the ideas that became NotebookLM, Google Flow, Google Beam, and AI Studio all started with small groups of people who were obsessed with an insight and couldn’t stop themselves from building. These ideas surfaced in the shower, over a weekend, in a hallway. The implication for teams everywhere is that the environment matters more than the process. “You cannot microwave a good idea.” 2. Fall in love with the problem, not the product. Most ideas require three, four, or five significant pivots before they either find traction or get shut down. Teams that get hooked on a specific solution will struggle to make those pivots. Teams that are attached to the problem—the underlying pain point or the user behavior they are trying to shift—can keep adjusting until they get it right. 3. When looking for early PMF, data and retention are important, but the strongest signal of PMF is in people’s eyes. Do they light up? Do they lean in? When you show someone a prototype and they’re genuinely engaged, you feel it. When NotebookLM’s first AI podcast demo was played in a hallway, the reaction was immediate and unmistakable. 4. Keep a list of what is “almost possible,” and move fast when something crosses over. Google Labs keeps a live list of experiences that are almost possible. When something crosses the threshold, Josh describes it as “a phase shift,” like solid to gas, and the team moves immediately. Maintain a running list of things that are almost ready, and have a team positioned to sprint when one of them arrives. 5. Every company needs to think like a labs team now. The underlying technology is changing so fast that what’s possible is shifting from week to week. Which means that every company needs to be spending time playing with the latest models, running far-out experiments, and being first to discover what they unlock before the competition does—and then bringing those discoveries to the rest of the org. 6. When something is not working, the team almost always knows before the leader does. Josh’s instinct as a leader is not to be the one who decides to shut down a project. He watches instead for the passion running out, for the sense that the team has tried everything and nothing is working. His goal is to create an environment where people feel safe enough to say, “I don’t think this is working.” 7. Someone’s “unlearning rate” is one of the most valuable skills right now. Everyone talks about growth mindset and learning speed. Josh is watching for something different: How quickly can someone take an insight, validate or invalidate it through testing, and then let it go? The ability to walk away from a prior belief or approach when the evidence says you should is rare and becoming more valuable. 8. Roles are not blurring into one universal “builder” role. There is a tendency in the current moment to talk as though everyone is becoming a generalist builder. Josh thinks this would be a bad idea. The best people on his teams still have a primary specialty—a major—and also a “minor” they’ve developed. The advice he gives to early-career PMs who are excited about vibe coding and prototyping is clear: build adjacent skills in service of becoming an exceptional PM, not as a replacement for it. 9. A well-designed lab needs isolation, a clear mission, and the ability to create “good trouble.” Three things matter when setting up an internal labs team. First, it needs structural independence, reporting far enough from existing business units that weird ideas can survive long enough to prove themselves. Second, the lab’s mandate has to be explicit: are you graduating ideas to existing products, or inventing new categories? Clarity on that question changes almost every decision. Third, the lab should be a place to challenge how the company itself works, not just what it builds. At Google Labs, that included piloting new job ladders and builder roles before rolling them out more broadly. 10. Google Labs manages the marathon-sprint tension through seasons. At any given time, roughly 20 to 30 experiments are running in parallel, but individual teams experience a defined rhythm: sprints, post-launch recovery, and open exploration periods. After Google I/O each year, many labs teams enter a mode that looks unproductive from the outside: no shipping, just hacking, building, and discovering. Josh treats this as the seedbed for the next season.

Broadcom CEO Hock Tan says capital, not chips, memory or power, may become AI’s biggest constraint. inancing is harder to fix, because data centers and chips get paid for years before they earn anything, and higher bond yields raise that cost. Intel CEO Lip-Bu Tan named the same pressure, calling “capital and the cost of capital” the front-line challenge and saying people will eventually want to see ROI. Neither sees demand cooling soon, and Hock Tan said compute demand outstrips supply and could stay that way through 2029, possibly into 2030.

卧槽这谷歌波是真掀桌了， 他们刚出了个 Mac 端的 AI 笔记应用，免费，没有任何订阅。 叫做AI Edge Foresight，录音，转写，整理，提问，全在 Mac 本机离线跑，一分钱不收。 开会它帮你录音转写，自动整理成笔记，会后还能直接对着你所有旧记录提问。 最狠的是全程在本机跑，断网照用，录音和笔记全留在你电脑里，不上传云端。 它本地跑了两个谷歌模型，EmbeddingGemma 2 专门翻你所有记录做语义检索， Gemma 4 负责生成回答，检索加生成一条龙，全在苹果芯片上跑完，不联网。 等于 Granola 那种按月付费、录音还得传上云的活儿，谷歌免费，离线，在你自己机器上给办了。 现在还是 labs 实验版，只认苹果芯片 Mac，Intel 用不了，也不在 App Store，下载地址放评论区。

四天前我说独立游戏因为反编译就要完蛋了，当时没几个人理，现在全网都在聊 AI 反编译了 说说我的看法。3A 有品牌、有法务、有高知名度，反编译3A游戏并试图以此牟利是相当吃力不讨好的行为。独立游戏不一样，Unity Mono 和 Godot 打出来的包、H5 小游戏，基本等于连源码一起发出去了，对国内的黑灰产极为友好 黑灰产逆向铺量的链路也很清楚：下载新品节Demo或游戏正式版并逆向，改个名，甚至皮都不换就上架抖音小游戏套利。《账单必须支付》《深海刮刮乐》等就是这种情况 类似这种情况以后只会更多。举报给抖音能确保下架，但人家可能已经赚完一波，换个号又上来了，然后就是重复这个流程 独立开发者现在能做的： 正式版别再用 Mono 打包，换成 IL2CPP 再加代码裁剪 Godot 项目把 PCK 加密打开 H5 和网页游戏把关键逻辑和数值放到服务端 资源里加水印，每个版本留指纹 留好开发记录和版本存档 但说实话，这些只能抬高成本，做不到完全避免。AI 读汇编只会越来越强，它们的意义是增加逆向成本，减少你被逆向的可能性，同时在你的游戏被抄了以后更方便地证明你是原作者 但是在我看来，真正治本的方法，是构建新的强有力的护城河。当前的独立游戏护城河正在从代码、游戏设计等转到别的地方：个人IP、品牌、社区 毕竟就算你把沉星之序破解了，Jonathan Blow也不会因此饿死，依旧有大量的人会为他的游戏买单 但是如果没有可以依附的个人ip和品牌，获客能力不够强的话，单靠独立游戏产品本身是非常薄弱的

Dusting off this tweet from April because this gap in shared understanding of LLM capability is *widening*. It's now less "two groups of people speaking past each other" and more a sharp funnel. - Napkin math somewhere around 6B people (~75% of the population) have barely come in contact with LLMs at all. - Around 1-2B (~20%) are casual and infrequent users of free-tier ChatGPT-like products. This group sees derpy chatbots and treats them a bit like a better Google search, a writing aid, or etc. Maybe an agent tries to book you a flight. I have non-tech friends who (reasonably, imo) say they have not much use for it at all. Even many professionals outside of math&code are in this tier. For example, execs and many other functions spend a lot of their time talking to other people, so while they understand what is happening intellectually, it is still second-hand and a bit abstract. - Now we get to professional use of frontier-grade LLMs in math&code. Somewhere around 20M people (0.2%) see first-hand that large, complex projects that used to take them weeks/months can now be completed by agents with a prompt. Building apps, copying apps, translating apps, decompiling apps from binaries... This has all happened very quickly and recently - less than 1 year ago, I was writing code manually by hand, typing memorized computer code commands into a code editor character by character, occasionally pressing Tab to autocomplete a little chunk of code. - And finally we get to the ~5,000 people (~0.00006%) with access to frontier-grade systems internally. The external world has seen the preview. It looks like swarms of thousands of agents collaborating over weeks on software mega projects: minting zero days, running cyber attacks and defenses at machine speeds, discovering new science, advancing the frontier of mathematics. Things that would have taken top professionals in the industry years of work. Meanwhile, human review and comprehension are starting to fall behind. For example, people are still involved in the "archeology" of the OpenAI-HF incident from many months ago. Mathematicians may be poring over the 722 manuscripts on frontier mathematics for a while. The funnel is driven by a combination of factors: 1. The impact scales with ambition, problem size, and horizon. A question with a paragraph answer barely stresses the system. You need a reservoir of big, difficult problems that you really care about. This aspect drives the consumer / professional dimension of the funnel. 2. The jaggedness of the system (which I have written about a lot separately). Capability peaks in domains that are digital, verifiable and economically valuable. This is because LLM capability emerges from reinforcement learning on verifiable rewards on a curated environment mixture driven by revenue potential. This aspect primarily drives the area (e.g. math&code) dimension of the funnel. 3. Access. Free-tier, paid-tier, internal. So this is the weirdness of the moment. The general public has mostly not interacted with these systems. When they have, it looks like a derpy chatbot. The majority of professionals still see only a modest uplift. And a small sliver of professionals are experiencing the vertigo of the curve going vertical. And it is all happening at the same time.

Dusting off this tweet from April because this gap in shared understanding of LLM capability is *widening*. It's now less "two groups of people speaking past each other" and more a sharp funnel. - Napkin math somewhere around 6B people (~75% of the population) have barely come in contact with LLMs at all. - Around 1-2B (~20%) are casual and infrequent users of free-tier ChatGPT-like products. This group sees derpy chatbots and treats them a bit like a better Google search, a writing aid, or etc. Maybe an agent tries to book you a flight. I have non-tech friends who (reasonably, imo) say they have not much use for it at all. Even many professionals outside of math&code are in this tier. For example, execs and many other functions spend a lot of their time talking to other people, so while they understand what is happening intellectually, it is still second-hand and a bit abstract. - Now we get to professional use of frontier-grade LLMs in math&code. Somewhere around 20M people (0.2%) see first-hand that large, complex projects that used to take them weeks/months can now be completed by agents with a prompt. Building apps, copying apps, translating apps, decompiling apps from binaries... This has all happened very quickly and recently - less than 1 year ago, I was writing code manually by hand, typing memorized computer code commands into a code editor character by character, occasionally pressing Tab to autocomplete a little chunk of code. - And finally we get to the ~5,000 people (~0.00006%) with access to frontier-grade systems internally. The external world has seen the preview. It looks like swarms of thousands of agents collaborating over weeks on software mega projects: minting zero days, running cyber attacks and defenses at machine speeds, discovering new science, advancing the frontier of mathematics. Things that would have taken top professionals in the industry years of work. Meanwhile, human review and comprehension are starting to fall behind. For example, people are still involved in the "archeology" of the OpenAI-HF incident from many months ago. Mathematicians may be poring over the 722 manuscripts on frontier mathematics for a while. The funnel is driven by a combination of factors: 1. The impact scales with ambition, problem size, and horizon. A question with a paragraph answer barely stresses the system. You need a reservoir of big, difficult problems that you really care about. This aspect drives the consumer / professional dimension of the funnel. 2. The jaggedness of the system (which I have written about a lot separately). Capability peaks in domains that are digital, verifiable and economically valuable. This is because LLM capability emerges from reinforcement learning on verifiable rewards on a curated environment mixture driven by revenue potential. This aspect primarily drives the area (e.g. math&code) dimension of the funnel. 3. Access. Free-tier, paid-tier, internal. So this is the weirdness of the moment. The general public has mostly not interacted with these systems. When they have, it looks like a derpy chatbot. The majority of professionals still see only a modest uplift. And a small sliver of professionals are experiencing the vertigo of the curve going vertical. And it is all happening at the same time.

codebases have always contained slop. we have always been okay with this because code rarely needs to be perfect, outside of mission critical "hard" (eg embedded) software. swe is a unique form of engineering because you're often building something that constantly evolves and is almost organic - user tastes change, your mental models evolve, and features come and go. admittedly, i and many others who've been coding for a long time were initially (like back in 2024) skeptical that agents would write production code. "but what about all the slop that would enter our beautiful codebase?". well, even before, we had human slop, and there was a lot of it! if we're being truthful, most people call "code i didn't write" slop. the reality is that 99% of people (i include myself in that number) don't even know what Good Code looks like, let alone be capable of writing it. i would propose that Good Code is: • correct • bug free • fast and cost efficient to run it has always been exceedingly difficult and expensive to write truly Good Code, because for the longest time (maybe @bendlang will change that), code was not something you could cheaply and formally guarantee to have all 3 of these properties. this is why writing hard software is much more difficult and expensive, because depending on the use case, you may only get one chance and so you have to be right from the start. but the vast majority of software we use does not need to be Good Code to provide value to users. in fact, it is very likely that most software we use is not formally verifiable! we accepted some bugs in exchange for velocity and progress, so we rely on tests and o11y instead and have adopted a more incremental approach. this is a good thing! agents help us move faster, and if you're open to the idea, are capable of helping us write better code. it won't be Good Code (that is still expensive), but it is likely to be much better on average at scale. the new craft of swe is to figure out how to do this. it's your job now! if your productivity with and without agents is not significantly different, you're doing something wrong. but it's not too late to embrace it and use it as an opportunity to learn a whole new skill tree

Judging by my tl there is a growing gap in understanding of AI capability. The first issue I think is around recency and tier of use. I think a lot of people tried the free tier of ChatGPT somewhere last year and allowed it to inform their views on AI a little too much. This is a group of reactions laughing at various quirks of the models, hallucinations, etc. Yes I also saw the viral videos of OpenAI's Advanced Voice mode fumbling simple queries like "should I drive or walk to the carwash". The thing is that these free and old/deprecated models don't reflect the capability in the latest round of state of the art agentic models of this year, especially OpenAI Codex and Claude Code. But that brings me to the second issue. Even if people paid $200/month to use the state of the art models, a lot of the capabilities are relatively "peaky" in highly technical areas. Typical queries around search, writing, advice, etc. are *not* the domain that has made the most noticeable and dramatic strides in capability. Partly, this is due to the technical details of reinforcement learning and its use of verifiable rewards. But partly, it's also because these use cases are not sufficiently prioritized by the companies in their hillclimbing because they don't lead to as much $$$ value. The goldmines are elsewhere, and the focus comes along. So that brings me to the second group of people, who *both* 1) pay for and use the state of the art frontier agentic models (OpenAI Codex / Claude Code) and 2) do so professionally in technical domains like programming, math and research. This group of people is subject to the highest amount of "AI Psychosis" because the recent improvements in these domains as of this year have been nothing short of staggering. When you hand a computer terminal to one of these models, you can now watch them melt programming problems that you'd normally expect to take days/weeks of work. It's this second group of people that assigns a much greater gravity to the capabilities, their slope, and various cyber-related repercussions. TLDR the people in these two groups are speaking past each other. It really is simultaneously the case that OpenAI's free and I think slightly orphaned (?) "Advanced Voice Mode" will fumble the dumbest questions in your Instagram's reels and *at the same time*, OpenAI's highest-tier and paid Codex model will go off for 1 hour to coherently restructure an entire code base, or find and exploit vulnerabilities in computer systems. This part really works and has made dramatic strides because 2 properties: 1) these domains offer explicit reward functions that are verifiable meaning they are easily amenable to reinforcement learning training (e.g. unit tests passed yes or no, in contrast to writing, which is much harder to explicitly judge), but also 2) they are a lot more valuable in b2b settings, meaning that the biggest fraction of the team is focused on improving them. So here we are.

codebases have always contained slop. we have always been okay with this because code rarely needs to be perfect, outside of mission critical "hard" (eg embedded) software. swe is a unique form of engineering because you're often building something that constantly evolves and is almost organic - user tastes change, your mental models evolve, and features come and go. admittedly, i and many others who've been coding for a long time were initially (like back in 2024) skeptical that agents would write production code. "but what about all the slop that would enter our beautiful codebase?". well, even before, we had human slop, and there was a lot of it! if we're being truthful, most people call "code i didn't write" slop. the reality is that 99% of people (i include myself in that number) don't even know what Good Code looks like, let alone be capable of writing it. i would propose that Good Code is: • correct • bug free • fast and cost efficient to run it has always been exceedingly difficult and expensive to write truly Good Code, because for the longest time (maybe @bendlang will change that), code was not something you could cheaply and formally guarantee to have all 3 of these properties. this is why writing hard software is much more difficult and expensive, because depending on the use case, you may only get one chance and so you have to be right from the start. but the vast majority of software we use does not need to be Good Code to provide value to users. in fact, it is very likely that most software we use is not formally verifiable! we accepted some bugs in exchange for velocity and progress, so we rely on tests and o11y instead and have adopted a more incremental approach. this is a good thing! agents help us move faster, and if you're open to the idea, are capable of helping us write better code. it won't be Good Code (that is still expensive), but it is likely to be much better on average at scale. the new craft of swe is to figure out how to do this. it's your job now! if your productivity with and without agents is not significantly different, you're doing something wrong. but it's not too late to embrace it and use it as an opportunity to learn a whole new skill tree

Your next users are machines Agents burn nearly 5x the tokens people do, up 14x in six months