maudnals (@maudnals) header
maudnals (@maudnals) profile picture

Building tools and communities. Web x AI x Security/Privacy. Developer Relations Engineer @ Google, Chrome. Tweets my own.

Paris
Joined November 2014
Photo by maudnals (@maudnals)Photo by maudnals (@maudnals)Photo by maudnals (@maudnals)Photo by maudnals (@maudnals)Photo by maudnals (@maudnals)Photo by maudnals (@maudnals)Photo by maudnals (@maudnals)Photo by maudnals (@maudnals)Photo by maudnals (@maudnals)Photo by maudnals (@maudnals)
Cheatsheet: How to make your WebMCP tools safer
4
3
1
15
2,501
This is just the TLDR. See all these hints in action in our demo ⤵️ youtube.com/watch?v=YCOy5uGJ… with @kaskulikowski and I
4
88
maudnals retweeted
I’ve been experimenting with Jev and WebMCP 😁 - Jev selects the next tool and fills enumerable inputs. If the same call needs free text or a number, it falls back to LLM. If that same call also needs free text, a number, or another non-enumerable value, an LLM fills only those remaining fields. What's also interesting is that based on the confidence score given by Jev we can make decisions in harness - if Jev is 97% sure we might trigger the tool - but if the confidence is, let's say, 60% we might want to do additional checks! Evals next? 🤔 1/2 🧵
5
5
3
36
7,138
Spots are limited - so if you're in Paris and are building with/curious about OpenAI APIs and have a specific use case in mind for your startup, apply now!
🇫🇷We are hosting a special OpenAI Builder Lab: Paris edition next week with our friends at @joinstationf in Paris on Tuesday. Join me and @romainhuet, @kagigz, @stevenheidel and other @OpenAI tech leaders for a hands on building day. Apply here events.openai.com/openaibuil…
2
1
1
13
2,111
I'm giving #Bluesky a try, find me there with the same handle! 🦋
1
332
Do you own an Android app and a website that share the same account system? Signal it to password managers, so your users can sign in seamlessly using autofill 🪄 Also useful if you have multiple websites that share the same account system! developer.chrome.com/blog/se…
4
16
24
6,492
Enhance your prompt engineering with a practical evaluation approach, LLM-as-a-judge ⤵️
Struggling with manual work to evaluate different prompts and models? Learn how to efficiently test and compare different LLMs and prompts using the "LLM as a judge" technique. This approach can save you time and provide more accurate results. web.dev/articles/test-llm-ca… #ai #webdev
2
294
XSS vulnerabilities keeping you up at night? 😱 Google's new "Commitment to Secure by Design" whitepaper has answers! Safe Coding and web platform improvements are key. Read more (page 7): static.googleusercontent.com…
2
1
24
1,202
Inspired to build something by @addyosmani 's demo? Here's an additional push - The Google Chrome Built-in AI Challenge is running until Dec 3rd, with several categories and cool prizes! Join the Hackathon, try the API, give feedback, win prizes! developer.chrome.com/blog/ai…
"What would browser APIs for AI look like?" Here's a preview of @GoogleChrome's latest experiments integrating AI directly in the browser using Gemini Nano. The team has been working hard to bring the power of on-device AI models to developers via (experimental) new Web APIs. These include: ✍️ Writing new content (Writer API) ♻️ Rewriting existing text (Rewriter API) 📚 Summarizing information (Summarizer API) 🗣️ Translation (Translation API - coming soon) What sets this apart is the speed and privacy - you'll notice remarkably quick responses, all happening right in your browser thanks to the small language model being local. The initial exploration revolved around the Prompt API, allowing users to broadly interact with an AI model similarly to chat interfaces you might be familiar with. In the demo video I've recorded, you can see these features in action - from generating a post about a trip to rewriting content in different styles and lengths, and even translating text into various languages. Many of these experiments can be enabled through Chrome flags. We're excited to see what you build with the experimental APIs. Do feel free to pass along feedback to @KenjiBaheux @domenic and @andreban
6
13
12,456
This is a BIG change to Chrome DevTools Perf Panel to make it easier to see and reproduce issues BEFORE doing a full trace. So you can be more confident you’re tracing the right thing.
🔥 Just released in Chrome 129 today, a new Performance panel landing page! 🔹 monitor your local Core Web Vitals in real time 🔹 compare them to real-user data from CrUX 🔹 and more Learn all about it in today's blog post: developer.chrome.com/blog/de…
17
63
4,524
PSA: creating a passkey for a website does not imply that you can no longer log in with a password
2
2
1
23
4,070
maudnals retweeted
To run a model on WebGPU, all you need to do is specify { 𝚍𝚎𝚟𝚒𝚌𝚎: '𝚠𝚎𝚋𝚐𝚙𝚞' } when loading the model/pipeline. It's really that simple! 😍 For example, here's how to compute text embeddings locally in your browser with WebGPU acceleration: v2.scrimba.com/s0lmm0qh1q
2
4
1
52
3,724
maudnals retweeted
I'm excited to announce that Transformers.js V3 is finally available on NPM! 🔥 State-of-the-art Machine Learning for the web, now with WebGPU support! 🤯⚡️ Install it from NPM with: 𝚗𝚙𝚖 𝚒 @𝚑𝚞𝚐𝚐𝚒𝚗𝚐𝚏𝚊𝚌𝚎/𝚝𝚛𝚊𝚗𝚜𝚏𝚘𝚛𝚖𝚎𝚛𝚜 or via CDN (example below) 👇
14
202
15
1,111
87,423
💡What if you could answer the most common question when working with video content: Does it contain something you want? I made an MVP, with #WebAI of course, that watches with you to grab images of key moments for you -meet doesVideoContain.js! ⭐Github: github.com/jasonmayes/doesVi…
12
14
6
60
17,015
Great first look into the Prompt API in Chrome by @christianliebel!
1
2
186
maudnals retweeted
🟡#dotJS is coming this Thursday 🥳 🌟Join us on June 27 to listen to our fantastic speakers @LeaVerou @markdalgleish @BenLesh @_jessicasachs @mgechev @cramforce @bendhalpern @jamesqquick @feyeleanor @marcba @rawkode @stoyanstefanov @maudnals @Lunakepio @rubyjane88 and Bassel Ghandour! Book your ticket ➡️ dotjs.io/register #JavaScript #React #NodeJS #AngularJS #VueJS #typescript #JS #webdeveloper #TechConf
6
1
22
4,912
Always forget that `nvm use --lts` is correct (not `nvm use lts`)
253
Prompt engineering for smaller LLMs is a different game! Check out our tips for getting the most out of smaller models like Gemma 2B for on-device use cases: web.dev/articles/practical-p… #AI #MachineLearning #PromptEngineering #webdev #webAI
2
8
387
Amazing example of on-device AI! #webai
Hey everyone, get ready for another cool #WebAI sideproject 🎉 In the last weeks I spent a lot of time building a #RAG solution directly in the browser without dependencies to any AI cloud provider. And today I am proud to share it with you😊
1
2
222