搜索“Meta”

triking feature: when an important, groundbreaking invention or new business gets backing from capital markets, it can rapidly establish a near-monopoly advantage. Take Google, or Meta—Facebook in social, Google in search—including Windows before that, in the desktop operating system space, and Apple’s iPhone. They each rapidly—in maybe three or four years—established a monopoly-like advantage in their respective fields. That advantage isn’t just technological—more importantly, it’s capital, inf

匹配字段正文

velopers across Africa are increasingly building on cheap, customizable Chinese models — including an Alibaba model reportedly handling Ugandan languages better than anything from Meta or Google — as Chinese firms expand cloud, telecom, and government AI partnerships across the continent. (NYT); A wide-ranging Economist briefing warns that the AI push Beijing is racing to promote is also the one most likely to displace its own workers — it opens in Zhengzhou, hub of China’s ¥100 billion ($15 bil

匹配字段正文

hich insist on a closed-source approach and reportedly lobby for policies that could constrain open models. The other is a coalition of tech companies led by Nvidia, Microsoft and Meta, which are strongly in favor of open-weight models. They signed a letter last week warning against “premature restrictions” on open-weight models and launched an open AI security alliance, an organization building tools to defend against cyber attacks from frontier models. Both sides have justifiable arguments. An

匹配字段正文

oonshot’s Kimi K2.6 model, an early sign that enterprise adoption of Chinese open-weight models is drawing direct Congressional scrutiny, not just chip-export debate. (SCMP/CNBC); Meta’s Mark Zuckerberg said the US should not ban Chinese AI models, joining Nvidia and other Silicon Valley voices pushing back on proposed restrictions—while Anthropic CEO Dario Amodei’s chip-ban proposal drew fresh scrutiny over whether it could actually curb China’s AI progress. (FT/Reuters/SCMP); Beijing said it w

匹配字段正文

performance is roughly as follows : Anthropic – Claude Fable 5 OpenAI – GPT 5.6 Sol Moonshot AI – Kimi K3 (open weights ) SpaceXAI – Grok 4.5 Zhipu (Z.ai) – GLM 5.2 (open weights) Meta – Muse Spark 1.1 DeepMind – Gemini Flash 3.5 Alibaba – Qwen 3.7 Max (3.8 announced, also to be open-weights, when writing) It is astonishing to see DeepMind, and some of the other American giants this low. In many ways, the X AI team deserves more credit. A visual summary from Artificial Analysis is below: This re

匹配字段正文

承受着一整套压力——其中之一就是维持其飙升的股价。只会有越来越多的模型跨越令人担忧的能力水平。训练AI模型并非魔法，而且尽管所需投资增加，我们尚未看到构建模型的门槛下降（目前还没有）。 对于开源而言，这一政策死亡螺旋——一个由相关且互补的可怕问题构成的双螺旋——的短期出口之一，是让一家美国公司发布能力相当的开放模型。这将把焦点从“只有中国在通过蒸馏构建开放模型”转移到“我们同舟共济，应当聚焦于我们生态内部复杂且不断变化的前沿问题”。这对开源而言是关乎存亡的优先事项，而那些有商业理由发布开放权重模型的公司，如微软和Meta（通过商品化其互补品），应当尽快行动。我对Meta在新领导层下的表现信心较低，但他们受益于AI的大规模普及，不应自己给自己设障。如果Reflection手上有一个尚可但并非前沿的模型，他们或许需要发布它，以挽救其拟定的业务方向。 比训练新模型更短期的解决方案是建立联盟。开源是一项分散的技术，没有明确的所有者，但其益处将由众多参与者共享。这些前沿实验室之外的“其他所有人”需要从今天就开始行动，研究如何继续安全地推出开放权重模型（并为自身的原则和价值观游说），以影响当权者。

匹配字段正文

ute needs by 90%, or double demand, or both. All of this has people hunting for patterns to recognise (indeed, it’s been observed that all conversations about AI end in a hunt for metaphors). It has become common to draw comparisons with fiber, which had a massive overbuild in the Dotcom bubble that looks a bit like the infrastructure build-out today. The narrow problem with that is that fiber construction was far ahead of demand, where AI compute construction is far behind demand, though, as ab

匹配字段正文

s year’s RAAIS, Roberta Raileanu set out why that ceiling exists and what it would take to lift it. Raileanu leads the open-endedness team at Google DeepMind and was previously at Meta. Her talk laid out a recipe for superhuman scientific discovery: a general system that makes groundbreaking discoveries across domains faster than people can. Three ingredients hold it together, but underneath all three sits one problem. We are good at searching for anything we can measure. We do not yet know how

匹配字段正文

argest tracked Hopper deployment at 200,000 GPUs, and by itself is almost five times the entire tracked A100 set. Tesla Cortex follows at roughly 66,000 H100-equivalent GPUs, then Meta’s GenAI clusters at 49,152, a CoreWeave H200 cluster at 42,000, Voltage Park at 24,000, and Germany’s JUPITER Booster at 23,536 GH200. There is a geopolitical point hiding in the table. National HPC systems are exact and visible, which makes them easy to count. Private fleets are estimated, harder to observe, and

匹配字段正文

trained it, and it was quite interesting. The problem with natural language proofs is you don’t know if it’s correct or not. They have a verifier, and the verifier is trained by a meta-verifier that makes sure that for all the problems they’re training this model to solve in the art of problem-solving, the verifier is giving good feedback. It works. It’s interesting that natural language verification with some sort of meta-verification seems to work so far in the published literature. It also se

匹配字段正文

Mythos 5 launch post: Mythos 5: 36.7%; Mythos Preview: 29.2%; Opus 4.8: 16.6% From OpenAI’s system cards: GPT 5.5: 26%; GPT 5.4: 9% from the 5.5 system card; GPT 5.2/5.3: 0% From Meta’s safety report for Muse Spark: Gemini 3 Pro: 0% (this is the “v2” third-party cyber benchmark) ExploitBench Description: ExploitBench is a benchmark containing 41 real-world vulnerabilities in the V8 JavaScript engine, which are known or strongly suspected to enable arbitrary code execution (ACE) within the brows

匹配字段正文

product, research, and engineering teams building AI agents that truly understand large, complex codebases. Before joining Augment, Vinay led product and platform organizations at Meta and Pure Storage. He’s drawn to problems that live at the intersection of technology and people, like how teams evolve, how AI reshapes the craft of software engineering, and what it takes to build things that evoke delight. Emma Webb Head of Communications Emma leads communications at Augment Code, with previous

匹配字段正文

progress. Some of the types of work include: Using DeepSeek’s models to try to build better GPU kernels ( 400 ), automating the conversion of PyTorch modules to CUDA code ( 401 ), Meta using LLMs to automate the generation of optimized Triton kernels for use within its infrastructure ( 439 ), using LLMs to help write kernels for non-standard hardware like Huawei’s Ascend chips (”AscendCraft” 444 ), fine-tuning open weight models for GPU kernel design (”Cuda Agent”, 448 ). One caveat here is that

匹配字段正文

erySelectorAll like the model has seen ten thousand times; Anti-bot injection. It's Chrome talking to itself What we got wrong A few months ago on this blog we wrote Closer to the Metal: Leaving Playwright for CDP. The conclusion of that post: "Our agents shouldn't have to know the nuances of CDP Targets in order to Get Stuff Done." Turns out we were wrong. That post listed ten ways a Chrome tab can crash. We built watchdog services to catch each one - tab crashes, target detach, renderer OOM, z

匹配字段正文

models, all with pretty-much equivalent capabilities. Every few weeks they leapfrog each other. There is variation within those capabilities, it’s possible to drop off the curve (Meta, for now) or fail to get onto it (Apple, Amazon, Microsoft, for now), or remain six months behind the frontier (China), or rely heavily on other people’s work (China, again) and all of this needs a lot of money (of which more below), but today there is no mechanic we know of for one company to get a lead that othe

匹配字段正文

of time scales, what the frequency of those time scales should be, etc.). To us, it seems much cleaner to have the minimal set of discrete components of the system to emulate our meta-learning algorithm (a long-term memory like parameter weights, a short-term memory like the KV cache, and a way to inject the relevant bits of short-term memory into long-term memory when we need to). But we could be wrong. Maybe this is a solution. Speaking of “when we need to”, a lot of people think surprise-bas

匹配字段正文

h the state-of-the-art further, but the proportion of progress will come more from the inference than purely the training side this year To wrap things up, I think if there is one meta-lesson from 2025, it is that progress in LLMs is less about a single breakthrough, and improvements are being made on multiple fronts via multiple independent levers. This includes architecture tweaks, data quality improvements, reasoning training, inference scaling, tool calling, and more. At the same time, evalu

匹配字段正文

hy you watched those, or bought those, or looked at those, and they don't really know what those things are. Amazon has close to a billion SKUs, but all it knows about them is the metadata typed in by humans and some level of purchase correlation. Instagram or TikTok have the same problem. These systems have correlation across SKUs, and down the funnel, but they don’t know why, just as a dog knows that the sound of door keys has high correlation with a walk, but doesn’t know what keys are. But a

匹配字段正文

models, doesn’t even accurately keep track of the number of words used. You will note that GPT-5 Thinking tends to be a fairly wild stylist when writing fiction, prone to complex metaphor, but sometimes at the expense of coherence and story (I am not sure someone would use all 47 words, but at least the count was right). And you will recognize that the new Chinese open weights model Kimi K2 Thinking has a bit of a similar problem, with some interesting phrases and a story that doesn’t quite mak

匹配字段正文

nt cluster (setting apart Google, master of multi-datacenter-training). However, Colossus 1’s 300 MW looks modest next to the Gigawatt-scale clusters under construction by OpenAI, Meta and Anthropic. Their hyperscaler partners are happy to leverage their balance sheet and win the market by throwing dollars at it. Was xAI’s prowess a one-time wonder? Today we will publicize some data from our industry leading datacenter model over the last year that is accessible to clients. This is the our same

匹配字段正文