Thomas H.

Web · Automation · IA

[jack @jack]

we're launching BUZZ!

a new groupchat platform for teams of people and agents of all sizes, built to reduce our dependency on slack and github. model-agnostic, decentralized, self-sovereign, and open source. 🐝

https://t.co/8IaMVeTQNo

[Marktechpost AI @Marktechpost]

Everyone's chasing the same tradeoff in voice AI:

Fast response → shallow answers.
Deep answers → painful latency.

Sakana AI Introduces KAME: A Tandem Speech-to-Speech Architecture That Injects LLM Knowledge in Real Time

Here's the core idea:

Instead of making the speech model "smarter" (expensive, slow to train, hard to scale), they kept a lightweight S2S model on the front end doing what it does best — responding immediately.

Then they ran a full back-end LLM completely asynchronously in parallel.

As you speak, a streaming STT component builds your transcript in real time and continuously fires it to the back-end LLM. The LLM sends back progressively refined "oracle" signals that get injected directly into the front-end's generation stream — mid-sentence, in real time.

The front-end doesn't wait. It starts talking. Then it corrects itself as better oracle signals arrive.

That's "speak while thinking." Not a metaphor. That's literally what the architecture does.

The numbers:

→ Moshi (baseline S2S): MT-Bench score 2.05, near-zero latency
→ KAME (S2S + gpt-4.1 back-end): MT-Bench score 6.43, near-zero latency
→ Unmute (cascaded system): MT-Bench score 7.70, 2.1 second latency

3x quality jump. Zero latency cost.

Full analysis: https://t.co/gsN2F3PxAh

Paper: https://t.co/JURr5luiem

Model weights: https://t.co/XwiGYNQEBW

Inference code: https://t.co/hqvHlF4KjS

Technical details: https://t.co/EPg5LfYsAj

@SakanaAILabs #audio #voiceai #ai #data #speechai

[Vercel @vercel]

Every chat platform has its own event model, threading system, and streaming quirks.

We felt this pain internally when we challenged every team to build agents to multiply their output, and the agents were easier than the chat plumbing.

So we built Chat SDK to remove that bottleneck.

Get started directly or via your coding agents with:

▲ ~/ 𝚗𝚙𝚖 𝚒 𝚌𝚑𝚊𝚝
▲ ~/ 𝚗𝚙𝚡 𝚜𝚔𝚒𝚕𝚕𝚜 𝚊𝚍𝚍 𝚟𝚎𝚛𝚌𝚎𝚕/𝚌𝚑𝚊𝚝

Read more ↓
https://t.co/aynoQlBMIs

[Harry Stebbings @HarryStebbings]

Why Giving Your Product Away for Free Should Be Marketing's Biggest Budget Line:

"Giving your product away for free should be the biggest portion of your marketing budget.

That can be freemium, discount codes or free offers for people to try the product.

But your free giveaways should be bigger than your paid marketing spend." @ElenaVerna

Will inference be the biggest line on the marketing budget of the future @mmurph @jasonlk @dharmesh @maorshlomo

[Firebase @Firebase]

Go beyond generic chatbots and create immersive audio interactions within your apps 🤖→🗣

Watch the demo to learn how to leverage Gemini's native audio models via Firebase AI Logic to provide a unique vocal experience.

Chapters:
0:27 - Streamed conversational experiences with Gemini
1:03 - Gemini Flash Native Audio response modality
1:37 - System instructions
2:07 - Real-time metrics
2:30 - Selecting custom voices
2:50 - Starting a live session
3:19 - App demo

[Thomas H. @ThomasHazeraPRO]

Bon sang mais c’est génial 😂

[Noah Zweben @noahzweben]

Announcing a new Claude Code feature: Remote Control. It's rolling out now to Max users in research preview. Try it with /remote-control

Start local sessions from the terminal, then continue them from your phone. Take a walk, see the sun, walk your dog without losing your flow.

[Stitch by Google @stitchbygoogle]

🚨 Our #1 most requested feature is here:

You can now export designs from any Stitch agent directly to Figma as editable layers.

Vibe Design is perfect for exploring many ideas in minutes. But sometimes, you need that final layer of polish.

Now you can move seamlessly between vibe design and polish.

[Thomas H. @ThomasHazeraPRO]

@RayaneRachid_ Malin merci 😉

[Vercel @vercel]

We're experimenting with ways to keep AI agents in sync with the exact framework versions in your projects. Skills, 𝙲𝙻𝙰𝚄𝙳𝙴.𝚖𝚍, and more.

But one approach scored 100% on our Next.js evals:
https://t.co/8ACw9BgudB

[Mistral AI @MistralAI]

Introducing Voxtral Transcribe 2, next-gen speech-to-text models by @MistralAI.
State-of-the-art transcription, speaker diarization, sub-200ms real-time latency.
Details in 🧵

[DeepLearning.AI @DeepLearningAI]

Google introduced Universal Commerce Protocol (UCP), a proposed open-source standard that enables AI agents to handle online purchases end to end, from discovery and ordering to payment and returns.

The protocol was developed with major retailers including Etsy, Shopify, Target, and Walmart, and payment providers such as American Express, Mastercard, Stripe, and Visa.

Learn more in The Batch: https://t.co/UujYsELwtq

[Thomas H. @ThomasHazeraPRO]

@elsanimy @karpathy 🤣🤣

[Thomas H. @ThomasHazeraPRO]

Sacré travail de composition et d’animation 👀👏
Assets midjourney?

[Anton Kolisnyk @AntonKolisnyk]

I'm so excited you all can see our new Edition website https://t.co/01w8cFdd0a This one is most fun and interactive we made. Team crushed it from every aspect of it, creative concept, design, video, product presentation and of course all #webgl renaissance magic.

[Thomas H. @ThomasHazeraPRO]

Prochain test 🤗

[Zeno Rocha @zenorocha]

Today, we have an open-source launch for you.

Announcing React Email 5.0.

1. Dark Mode Switcher
2. Tailwind 4 support
3. Resend Integration
4. 8 New Components

[Thomas H. @ThomasHazeraPRO]

@MArch_Emir61 @GoogleLabs Good question 😤

[Thomas H. @ThomasHazeraPRO]

L’alternative a CapCut’s avec le projet open source de @opencutapp 👀?

A lire aussi, leur petit manifeste 🍿
https://t.co/ymZ3GFwZAW

[Maze @mazeincoding]

for anyone who missed it, here it is

this is the most insane part about capcut's new policy

[Thomas H. @ThomasHazeraPRO]

🍿

[Wasteland Capital @ecommerceshares]

Do NOT install any agentic browsers like OpenAI Atlas that just launched.

Prompt injection attacks (malicious hidden prompts on websites) can easily hijack your computer, all your files and even log into your brokerage or banking using your credentials.

Don’t be a guinea pig.

[Thomas H. @ThomasHazeraPRO]

@privetavdey Love it

[Thomas H. @ThomasHazeraPRO]

Excellent 👍, même si ça ressemble plus ou moins a « firebase Studio » de Google

[Dani Grant @thedanigrant]

Ok so...imagine your browser let you edit the web like a figma

Live from @OpenAI DevDay

[Thomas H. @ThomasHazeraPRO]

Ou comment financer la bulle de l’ia… Quand une simple « conversation » se transformera en pure conversion 🫤 @grok Ton avis?

[tobi lutke @tobi]

Shopify merchants will be able to sell directly in ChatGPT.

We’ve been working with @OpenAI for quite some time so people can search and buy products in chat, and it’s something we’ve had a hard time keeping quiet.

Rollout is coming very very soon.