Top News
Anthropic just sued the Pentagon after being labeled a "supply chain risk." Filed two federal lawsuits on Monday. Even OpenAI and Google researchers are backing them ⚖️
OpenAI's robotics lead quit over the Pentagon deal. Caitlin Kalinowski left the day the contract was announced. Her exact words on why she left 👀
Yann LeCun started AMI Labs and raised $1.03B in four months. Not building another LLM. NVIDIA and Bezos backed it at $3.5B. What he thinks the whole industry is getting wrong 🌍
Meta just acquired Moltbook. A social network run entirely by AI agents, 19,000 communities, no humans required. The backstory before the deal is wild 🤖
Product News
Karpathy open-sourced 'autoresearch'. AI agents that run and evaluate their own ML experiments with zero human input. He woke up to 100 completed experiments 🔬
Claude Code Review just launched. Multiple agents hit your PRs in parallel, substantive comments jumped from 16% to 54%. Engineers are dismissing less than 1% of findings 🛠️
……….
….
Google dropped Gemini Embedding 2. First model that natively searches across text, images, video, and audio. Worth seeing in action 👀
Figure AI robot just cleaned an entire living room in 2.5 minutes. No remote control, one neural network, human speed 🤯
……….
The Speech API We Keep Coming Back To
If you're building voice agents or processing high-volume audio, you know how much transcription accuracy matters. We've been looking at Velma Transcribe by Modulate, a speech-to-text API built for real-world audio. The kind with background noise, thick accents, and emotional nuance.
It was trained on 500M+ hours of actual human speech, allowing it to detect 10+ emotions, 20+ accents, and 70+ languages out of the box. It also handles PII detection and redaction, which is a must-have for call center monitoring or sensitive voice apps.
Despite ranking #1 for accuracy, it's significantly cheaper. Outperforming Deepgram on cost by 88%, ElevenLabs by 92.5%, and AssemblyAI by 80%
The performance on the Earnings-22 and VoxPopuli datasets speaks for itself:
Modulate: 7.8% avg. WER
ElevenLabs: 7.9% avg. WER
AssemblyAI: 9.3% avg. WER
Deepgram: 12% avg. WER
It supports batch and real-time streaming and fits into existing workflows. Try Velma Transcribe with 400 free hours.
Deep Agents Hackathon - SF
Build real AI agents with AWS, Anthropic & more.
🏆 $50k+ in Prizes!
🗓️ Friday, 27 March
9:30 AM – 7:30 PM PT
Multimodal Frontier Hackathon - SF
Build real AI agents with Google DeepMind, Lovable & more.
🏆 $47k+ in Prizes!
🗓️ Saturday, 28 March
9:30 AM – 7:30 PM PT
TinyFish Accelerator – Build the Agentic Web
YC-style virtual accelerator for founders building real web-operating agents.
💰 $2M in Seed Funding
🌍 9 Weeks · Fully Virtual
Interested in sponsoring an upcoming event or hosting your own?
reach out here
Bland AI - Full Stack Engineer (SF)
Own features end-to-end for voice AI. Build dashboards and workflows that make enterprise voice AI actually usable. Ship fast, iterate from real customer feedback.→ Apply
Flora - Staff Engineer, Product (NYC)
Be a technical leader across frontend systems. Canvas rendering, complex workflow interactions, and the media layer that powers image & video editing.→ Apply
Flora — Growth Engineer (NYC)
First growth hire. Design and ship experiments across acquisition, activation, and retention on an AI-powered creative canvas.
→ Apply
DigitalOcean - Staff Technical Evangelist, Startups (SF, Remote)
Be the technical bridge between startups and DigitalOcean. Craft architecture content for 0-to-1 founders, recruit AI-native startups to build on DO.
→ Apply
Promote Your Launch or Hiring Call 🚀
Promote a product launch, event, or hiring call to our community.
Become a Venue Partner 🏢
Do you have an office space and want to drive exposure, attract top-tier tech talent, filter VCs and builders and elevate your product brand-at no cost?
Apply below to have your space selected as our Tech Hub of the Day in collaboration with our sponsors.
Judge or Speak at Our Events 🧠
Interested in judging at one of our events? You can apply here.