今日已更新 269 条资讯 | 累计 41043 条内容
关于我们

标签：#Rust

找到 502 篇相关文章

AI 资讯

The 4-layer voice-agent latency stack, traced with OTel spans

** How I instrument ASR, LLM, TTS, and the client with OpenTelemetry, and which number in each layer I actually look at ** TL;DR. A voice agent is four moving parts stuck together: speech to text, the model that writes the reply, text to speech, and the client that plays the audio back. End to end latency hides which of those four is slow on any given turn, so I stopped tracking it as one number and started tracing each stage as its own OTel span with a shared session id. The number I watch hardest is barge-in: when the user starts talking over the agent, how many milliseconds until the agent actually stops sending audio. In our setup we want that under 200ms, and when p95 barge-in creeps past that, the agent feels like it is talking at you instead of with you. Everything below is how I wire the spans, what attributes go on each one, and the p95 I page on per layer. The thing I keep saying, and the thing that keeps being true: voice agents fail in production not because of raw latency but because nobody simulated the audio and LLM pipeline together. You can have a fast ASR, a fast model, a fast TTS, and a voice agent that still feels broken, because the failure lives in the seams between them and in the parts (barge-in, jitter) that no single-stage benchmark touches. Tracing is how I get the seams to show up. A note before the layers. This is just the setup we run, the spans we emit, and the mistakes that made us add each attribute. Some of it is probably specific to our stack and will not transfer. I will flag that where I can. The shape of a turn, and why one span is not enough One turn is: user says a thing, agent says a thing back. Underneath that is roughly: audio frames come in, ASR turns them into text (streaming partials as it goes); the text plus history goes to the LLM, which streams tokens back; as text comes out, TTS turns it into audio, also streaming; the client receives audio frames and plays them, with some buffering to smooth out jitter. If you wrap

2026-06-09 原文 →
AI 资讯

ExtendDB: Open Source Amazon DynamoDB Compatible Adapter with Pluggable Storage Backends

AWS recently announced ExtendDB, a DynamoDB-compatible adapter that lets developers use the DynamoDB API with different storage backends, starting with PostgreSQL. The project supports existing SDKs and tools without modification, giving teams greater flexibility to run DynamoDB-style workloads outside of native DynamoDB while maintaining compatibility with current applications and workflows. By Renato Losio

2026-06-07 原文 →
AI 资讯

Building a SQL Lexer in Rust: Why I Replaced `Vec ` with `&str` and `Ident(String)` with Spans

I've been building a database engine from scratch in Rust, and I recently finished the lexer. The lexer itself wasn't the most interesting part. What I found more valuable was how my design evolved as I learned more about Rust and how compilers and database systems are typically implemented. My First Approach When I started, I stored the input as a Vec<char> . It felt straightforward because I could access characters directly without worrying about UTF-8 boundaries. I also represented identifiers like this: Ident ( String ) At first glance, this seems perfectly reasonable. Every identifier token carries its own text, making it easy for the parser to consume. The Problem As the lexer grew, I started asking myself a simple question: The identifier already exists in the original SQL query. Why am I allocating another string and copying the same data into every token? For a query like: SELECT username , email FROM users ; the source text already contains: username email users Creating separate String allocations for each identifier means duplicating data that already exists. I also learned an important detail about Rust enums. The size of an enum is influenced by its largest variant. Once variants start carrying additional data, every token instance becomes larger than it otherwise needs to be. Moving to a Span-Based Design Instead of storing identifier text directly inside tokens, I switched to storing only the token kind: Ident along with source location information: Span { start , end , line , column , } Now the token only answers two questions: What is this token? Where did it come from? If the parser needs the actual identifier text, it can recover it directly from the original SQL source using the stored byte range. Replacing Vec<char> with &str The second design change was moving away from: Vec < char > and operating directly on: & str using lifetimes. Instead of creating another collection containing the entire input, the lexer now walks over borrowed source tex

2026-06-06 原文 →
AI 资讯

What if weather observations could participate in blockchain security?

We are exploring an experimental blockchain mechanism called "Proof of Weather" In the world of blockchain, various methods are used to achieve network consensus. The most well-known is Bitcoin’s Proof of Work (PoW). While PoW is an excellent mechanism, it has one major drawback. It consumes an enormous amount of electricity. At one point, I found myself wondering: Does blockchain really require such vast computational resources? Isn’t there something else that’s needed? This led to the creation of Dawn, the experimental cryptocurrency project I am developing, and an experimental blockchain mechanism called Proof of Weather. In this article, I will discuss: Why I decided to use weather How Proof of Weather works Security considerations Implementation in Rust How Does Proof of Work Work? Proof of Work is often explained as a mechanism where computers compete against each other in computational tasks. However, one important property of PoW is that it produces outcomes that are difficult to predict in advance. Miners repeatedly perform massive amounts of hash calculations, and only those who happen to meet the conditions can generate a block. This unpredictability plays a role in determining who can produce the next block. However, this process consumes enormous amounts of electricity worldwide. So I wondered: Aren’t there already phenomena in nature that are difficult to predict? Why Weather? Proof of Weather utilizes weather data as that unpredictable element. Of course, weather forecasts exist. However, Temperatures several days in the future Atmospheric pressure at specific locations Precipitation Wind speed and other factors cannot be predicted with absolute certainty. In particular, when combining observations from multiple locations, it becomes even more difficult to accurately calculate future values in advance. In other words, meteorological observations have the potential to be used as A real-world information source where future values cannot be fully predic

2026-06-06 原文 →
AI 资讯

Analysis of Mo Gawdat and Marina Mogilko’s Conversation About the Future of AI, Startups, Education, and the Labor Market

AI Does Not Cancel Reality I watched the conversation between Mo Gawdat and Marina Mogilko about the future of AI. The conversation is strong. It contains important ideas, but it also contains many claims that sound large in scale, although on closer inspection they rely on very broad generalizations. AI is indeed changing the labor market, education, startups, content, hiring, and ways of thinking. But it does not cancel money, connections, trust, the human vector, creativity, necessity, morality, or people’s ability to adapt. Video on YouTube AI in hiring: automation amplifies chaos Many people have entered the job market. Companies receive huge volumes of resumes. HR departments cannot handle the volume. It is natural that part of the selection process is moving to AI. But there is a serious problem here. Candidates are also starting to play against AI. Resumes are adjusted to vacancies. Cover letters are assembled around keywords. Profiles become optimized for the filter, not for real work. In such a system, the best specialist does not necessarily pass. Often, the person who understood the selection mechanism better passes. The result: the picture becomes cleaner, while the quality of the decision becomes lower. The company gets not the strongest candidate, but the candidate who matched the algorithm best. This leads to lower hiring quality, lower productivity, and slower development. “I built a startup in six weeks”: a product is not a startup The conversation includes the idea that an AI startup would once have taken years and hundreds of engineers, and now it can be built in weeks. Technically, this is true. Prototypes are now built faster. Small teams have powerful tools. One person can now do more than a group could do before. But two different things are mixed here. Building a product faster has become real. Building a startup faster has become real only when resources are present. A startup is not only code. A startup is money, connections, trust, reputa

2026-06-06 原文 →
← 上一页 下一页 →