与「LLM」相关的搜索结果

做一个下dsh和pi的对比 Pi 的目标是“最小核心 + 用户自己拼”，DSH 的目标是“几乎所有能力都插件化，官方先给你几套完整组合”。 DSH 的骨架是Cordis。 Cordis 不是普通插件加载器，它强调两件事： 1. 可逆副作用（temporal composability）：插件卸载时，注册的服务、事件、工具 schema、prompt section 全部自动撤销。 2. 依赖声明与空间组合（spatial composability）：插件通过 `inject` 声明需要什么服务，运行时按依赖挂载。 所以在 DSH 里： - 模型适配器是插件 - 工具注册表是插件 - session log 是插件 - agent loop 本身也是插件 - 甚至 UI 也是插件 没有“神圣不可动的核心”。你想换 loop、换工具策略、换上下文压缩，挂一个新插件 + 改配置就行。 启动时是 Profile + Bundle叠出来的： 空根 → dsh-base（模型、工具、沙箱、凭证…） → dsh-web-app 或 dsh-headless → 用户自己的 cordis.patch.yml → 命令行 --patch `dsh --profile web --dump-config` 能直接把当前实际挂载的树打出来。 Session 设计是硬核部分 Session 是 append-only 的事件流 模型最终看到的上下文，必须能从这条 log 完整重建出来。官方写得很死： > Model-visible means logged. 任何会进模型请求的内容，都要先变成 session event。 所以 fork、resume、回放、UI 渲染、telemetry，全部从同一条流投影。这点比大多数 coding agent 做得更彻底。 Turn / Step 有claudecode的影子，流程如下： turn/start → claim input → agent/pre-step（可拦截、可改写） → step/start → llm/stream → tool/call → tools/pre-execute → execute → post-execute → step/end → 继续 or turn/end `agent/pre-step`、`tools/*` 这些是 waterfall，监听者必须显式 `next()` 才能往下传。扩展点设计得很规整。 Pi 的实际交集 DSH 仓库里有一个包： `@deepseek-ai/dsh-llm-pi-ai` 它就是把 pi-ai当成 LLM 适配器的后端。 多 provider、协议兼容、reasoning effort 映射、catalog 覆盖，都走 pi-ai 的能力，再包一层 Cordis 插件契约。 所以： - LLM 调用层：吃了 Pi 的基础设施 - Agent 编排、工具、session、UI、沙箱：完全自己的 Cordis 体系 “LLM 适配层直接用了 pi-ai，上层重做了一套更重的可组合 Runtime”。 模式上的对应 DSH 的 Minimal 模式才最接近 Pi 的默认体验：只留 shell + 文件编辑器，专门给 benchmark 用。 官方之前在 V4 的 agent 评测里就用过这个模式。 Standard 模式和 Code（PTC）模式则是完整工具集 + 程序化工具编排，已经远超 Pi 默认的 4 工具。 如果你喜欢 Pi 那种“核心极瘦、自己动手加东西”的感觉，DSH 会显得重，配置和概念也更多。 如果你要的是可替换的 agent loop、完整事件回放、多模式预设、以及官方已经搭好的 Web 界面，DSH 的插件树和 seam 设计更系统。 一句话： Pi 是极简可扩展的 coding harness；DSH 是基于 Cordis 的完整 Agent Runtime，LLM 层复用了 pi-ai，但整体不是同一套代码。

LLMs have a limited context window. When conversations grow too long this affects output quality, performance and cost. New blog post from Earendil engineer @vegardstikbakke on how compaction addresses this and how we’ve implemented it in Pi. Read the full post below

It’s always funny how people in SF twitter bubble will say things like “Claude has unbeatable narrative momentum” and then 2 weeks later “wow did Kimi just end the race” My brother in Christ nobody outside of this little bubble even calls any AI anything other than “Chat” and they think datacenters destroy water molecules You sound dumb. Focus on things that actually impact people not the little scores on overfit evals And also people who bet against Elon are stupid. The man built real world self driving cars (non-LLM based AI) that is still the best AI I’ve ever used by far. You think he can’t come from behind and win on writing code? He’s literally catching 30 story buildings with chopsticks and that’s like one of his projects I like a lot of the tools and have friends at all the main labs. Truly amazing work being done. But the overblown commentators are actually so out of touch

LLMs still produce bugs, but those bugs are different than what they used to be. It’s less off-by-ones and more about system design, ui usability, missing broader context. Some kinds of coding has been solved, but not all. While models continue to improve, adversarial code review has been an incredibly powerful tool to catch many of these kinds of bugs. It can be as simple as a one line prompt - “use a dynamic workflow to adversarial test every edge case in an iOS simulator”, or use Claude’s built in /code-review (or /code-review low, /code-review medium, etc.)

Sam Altman just drew the entire scaling curve for AI compute in one paragraph. VCs pay analysts six figures to build charts less clear than this Six and a half years ago, the world's top token user was one OpenAI employee running 100,000 tokens a month. The worldwide per-capita average was zero. Nobody outside a lab had ever touched an LLM. Today the worldwide per-capita average is 100,000 tokens a month. The leader at OpenAI runs hundreds of billions. That's a million-fold jump at the top and a hundred-thousand-fold jump for everyone else. In 6.5 years. Altman's forecast for the next 6.5: the average person hits 500 billion tokens a month, the token leader lands somewhere in the quadrillions. "That will just become the expectation." The forecasts every enterprise IT team is using right now assume flat or 10x growth. Altman is publicly telling them to plan for a million times. Screenshot this before your CTO's next AI budget review. The math on his slides won't match the math Altman just put on stage.

We open-sourced OranSim, the market simulator we built to test a campaign before it ships. Run it on your own campaign or just play around: It runs your campaign past a million virtual consumers before you spend a cent. Each is an LLM persona with its own taste and memory. They react to your creatives and influence each other. Real crowd dynamics, not a single model's opinion.

Claude's watermark probably doesn't work how you think. As the CTO of GPTZero, I'll explain how Anthropic, Google and OpenAI are building text watermarking in this brief explainer and whether it can be defeated. Almost all forms of watermarking that are fast and cheap enough for a frontier lab have the same formula, following the KGW method: In generation: 1. Let's say you've generated n tokens so far. Take those n tokens + a secret key to generate a random hash 2. Use that hash to randomly reweight the probabilities for the n+1 token, and then sample from that new distribution. In the simple case, you could split 50% of all English words into a green or red set based on your hash, and boost the probability of words in the green set. For watermark detection: 1. For each token, see if it was in the green or red set. 2. To do this, recreate the hash based on the secret key and the text preceding the current token. Then, recreate the green and red set of words. 3. Once you've checked all the words in the text, if the next token is selected disproportionally from the green set more than 50% of the time, you claim the text has the watermark. I can tell you want to ask the following: 1) Isn't it easy to mess up the hash if you paraphrase the text? The answer is mostly yes, however, you can use a statistical model to get your hash instead of a deterministic function (SIR, Adaptive Watermark). Since the entire watermark is probabilistic, this is fine. 2) Doesn't this make the text much worse? The answer is yes, it does - Yes, it does – but for most people, it's imperceptible (Google claims in human feedback study with 20,000 texts), since there are exponentially many ways to write the same paragraph. DiPmark does something more sophisticated to avoid shifting the text distribution on average. Of course, watermarks fail on short text or highly predictable texts like "2+2=4". 3) Shouldn't it be easy to figure out the green and red sets? The answer is no. You would need an exponentially large number of samples from the watermarker to reconstruct those sets exactly, but it's a risk if the detector is open to the wild (Watermark Stealing) Still, there are couple challenges that a frontier lab needs to overcome: 1. Their watermark needs to work token-by-token because they are streaming their text to users. Many watermark methods plan sentences or paragraphs at a time, or change the text after its entirely written, in order to make their watermark robust to paraphrasers, and a frontier lab cannot afford to do this yet (SemStamp, PostMark) 2. If the secret key leaks, the watermark is busted. To avoid a large blast damage from this, you need to have a couple secret keys in rotation. 3. There are some texts, like code, that cannot be arbitrarily changed, otherwise the code will break. In those cases, the watermark needs to selectively change words in parts of the text that can tolerate synonyms (i.e. like variable naming) - see SWEET, EWD, Invisible Entropy. 4. They will need to educate their users on how to deal with false positives and false negatives of a detector, which is a big challenge (one we put a lot of effort into) So, how do I see this playing out in the next 6 months? 1. If Anthropic releases the watermark detector publically, I think they defeat their own watermark. People find reliable watermark removal strategies by testing against Anthropic (AI detectors like GPTZero have an advantage here because they can train against these adversaries once they become popular). 2. If they keep the detector private to the government, like Google has done, it's "safer". However, there are some papers showing trained approaches that work robustly to zero-shot break watermarks without any data, simply because they try to write the text just like a human (Zhang et al. 2024, Watermarks in the Sand). Also, making your detector makes it battle-tested and stronger long-term (my experience). 3. In my testing, the watermarks don't survive intense paraphrasing (especially if you combine word choice and syntax attacks), or human text substitution (rewrite your AI text by plagiarizing human authors). The free paraphrasers I've tried have quickly bypassed Google Deepmind's SynthId for what it's worth. 4. All-in-all, frontier labs are likely okay with this because they expect most users to not attack the watermark, and also because they + European regulators likely don't care past a certain point - its good enough. 5. Overall, I think users of frontier LLMs will not really care about this, because 1) they don't realize watermarks are there, 2) EU will force everyone to conform, 3) this seems more like regulatory hoop-jumping than an earnest effort from frontier labs to expose LLM use Lastly, people's first concern shouldn't be watermarking, it should be AI detectors! If you're posting, "its not X, its Y!!", I don't think the watermark is going to make a difference :)

卧槽，真的被这个东西惊到了。 挖到一个叫 KrillinAI 的本地工具，免费，视频翻译、精准字幕、超自然配音、声音克隆，全流程一条龙搞定。 中英互翻效果特别稳， Whisper识别准确率高得离谱， LLM按段翻译上下文不乱， CosyVoice的TTS听着几乎像真人在说话。 支持Win和Mac，下载即用，还能直接用yt-dlp扒油管或者本地上传。 横屏竖屏自动适配，出海做内容的真的可以直接拿去用了。 以前为了翻一个视频又是剪映又是各种插件， 现在一个工具全搞定，字幕对齐干净，配音自然度超出预期。 免费、本地跑、无水印。

冯诺依曼结构对计算的精确性要求太高了，一套 fp16 算下来又慢又费电，llm 完全可以牺牲点精度和稳定，用类似模拟量栅极替代 ReLu 函数，人脑细胞的膜电位变化精度稳定性本来也不高，如此可立省 99% 能耗。