Tech Updates
74 subscribers
18 photos
257 links
Tech news, cool tools, projects, blogs, articles and more.
Download Telegram
https://huggingface.co/nvidia/parakeet-tdt-0.6b-v2

parakeet-tdt-0.6b-v2 is a 600-million-parameter automatic speech recognition (ASR) model by nvidia designed for high-quality English transcription, featuring support for punctuation, capitalization, and accurate timestamp prediction.

It seems better than whisper-3 large (https://huggingface.co/spaces/hf-audio/open_asr_leaderboard) with 1B less parameters.
👍1
https://www.reuters.com/business/openai-agrees-buy-windsurf-about-3-billion-bloomberg-news-reports-2025-05-06/

OpenAI has agreed to buy Windsurf, an AI coding tool, for about $3 billion, marking the ChatGPT maker’s largest acquisition to date.
https://github.com/astral-sh/ty

An extremely fast Python type checker and language server, written in Rust.

In early testing, it's 10x, 50x, even 100x faster than existing type checkers. ( >600x speed-ups over Mypy in some real-world projects.)
https://appwrite.io/products/sites

Appwrite, an Open Source Vercel alternative, publishes Appwrite Sites:

"With Sites, Appwrite offers a streamlined solution for easily deploying static and server-rendered applications. Everything is designed to simplify your workflow, from creating a site and connecting a domain to leveraging our templates“
Microsoft and Google released their coding agents - probably as an answer to Codex from OpenAI:

Coding agent from Google:
https://jules.google/

Coding agent from Microsoft:
https://github.blog/changelog/2025-05-19-github-copilot-coding-agent-in-public-preview/

Jules has a couple of free queries per day, Copilot Agent is included in Pro+.
https://www.anthropic.com/news/claude-4

Anthropic released Claude Opus 4: "… the world’s best coding model, with sustained performance on complex, long-running tasks and agent workflows".

And Sonnet 4:
"…is a significant upgrade to Claude Sonnet 3.7, delivering superior coding and reasoning while responding more precisely to your instructions."
❤‍🔥1
https://www.perplexity.ai/de/hub/blog/introducing-perplexity-labs

Perplexity Labs can craft everything from reports and spreadsheets to dashboards and simple web apps — all backed by extensive research and analysis. Often performing 10 minutes or more of self-supervised work, Perplexity Labs use a suite of tools like deep web browsing, code execution, and chart and image creation to turn your ideas and to-do’s into work that’s been done.
https://github.com/apple/container

macOS 26 is getting native support for Linux containers.
Apple open sourced a tool for creating and running Linux containers using lightweight virtual machines on a Mac. It's written in Swift, and optimized for Apple silicon.
4
https://www.docker.com/blog/docker-desktop-4-42-native-ipv6-built-in-mcp-and-better-model-packaging/

Docker desktop adds native IPv6, built-in MCP, and better model packaging.

One of the important parts is:
"Unlike typical setups that run MCP servers via npx or uvx processes with broad access to the host system, Docker Desktop runs these servers inside isolated containers with well-defined security boundaries. All container images are cryptographically signed, with proper isolation of secrets and configuration data."
Two researchers released an early version of their new LM architecture.
The advantage is that it does not use any variant of multi-head attention or recurrence, and it works well with long context lengths. They are calling it "Avey". Everything is open-sourced under a Apache-2.0 license.

The main advantage would be to massively decrease the memory usage for the context window, that means even a context of 100k+ tokens would still be compute and not memory bound. It’s quite early and slower than current tech though.

Paper: https://arxiv.org/abs/2506.11305
GitHub: https://github.com/rimads/avey-dpa