Tweets

🌕 @gru_space is building durable space habitats so humans can one day live on the Moon and Mars.

Its first missions will mine lunar regolith to construct a long-term pressurized habitat on the Moon for commercial space tourism — a hotel on the Moon.

Congrats on the launch @skyler_chan_!
ycombinator.com/launches/P9g-g…U

Stoke Space is racing to build the world's first fully reusable rockets that can launch, survive reentry, and fly again almost right away.

In this episode of Hard Tech, @aaron_epstein sits down with @stoke_space co-founders @AndyLapsa and @Rkt_Da to find out why they chose to take on one of the hardest problems in rocket science, how being a smaller startup gives them an advantage, and what full reusability could unlock for the future of spaceflight.

00:00 — Intro
01:16 — Stoke Space’s mission: Rapid reusability
02:18 — Why Second Stage capsules fail reentry
03:34 — Stoke Space’s stage 2 solution
05:30 — Reusability-First Design Philosophy
07:25 — Early Engine Development & Test Strategy
10:48 — Vertical Integration & Manufacturing
11:21 — Iteration Speed as a Competitive Advantage
12:29 — Software as Core Infrastructure
14:00 — Path to Orbit & Launch Operations
15:04 — How This Could Change The World

OpenAI for Healthcare:

ElevenLabs just solved speech transcription.

Their new model, Scribe v2, is SOTA in word error rate. It handles pauses, laughs, complex words with ease.

It's so good at segmenting speakers and made ZERO errors on the ultimate test: identical twin voices. 

How it works 👇

We’ve made lots of improvements with Veo 3.1! It’s more expressive, supports portrait mode and SOTA video upscaling to 1080p and 4K. Rolling out starting today - try it in the @GeminiApp (for plus/pro/ultra users) and @FlowbyGoogle - enjoy!

Ben Horowitz on foundation models as infrastructure:

“People believed the big foundation models would be giant brains that could do anything better than anybody.”

“It has not played out quite like that.”

“Maybe the application behavior is actually more important than having the biggest model trained with the most GPUs.”

@bhorowitz @jkhamehl

ChatGPT Health keeps your health chats, files, and memories in a separate dedicated space.  

Health conversations appear in your history, but their info never flows into your regular chats.  

View or delete Health memories anytime in Health or Settings > Personalization.

Context engineering isn't just about throwing more data at your LLM - it's about giving it the right context at the right time.

In this talk for @OReillyMedia, our Developer Relations Engineer @tuanacelik walks through how memory blocks help you build agents that maintain structured context for complex tasks. She demonstrates artifact memory blocks using a restaurant order tracking bot - showing how to distill entire conversations down to just the essential structured information (pizza type, toppings, address) rather than processing the full chat history.

Key concepts covered:
· Different types of memory blocks (static, fact extraction, vector, artifact)
· Context ratio management - balancing chat history vs system prompts vs memory
· Using agent workflows to construct and optimize context step-by-step

The example shows how artifact memory blocks can transform a meandering conversation into a clean, structured order - exactly what you need for production agents handling real-world tasks.
Watch the full talk: youtube.com/watch?v=POO3ck…

Because the system harnesses internal activations already happening within a model, and reserves heavier computation only for potentially harmful exchanges, it adds only ~1% compute overhead.

It’s also more accurate, with an 87% drop in refusal rates on harmless requests.

Your AI retrieves based on semantics but highlights based on keywords. See the problem?

You search for "iPhone performance" and it only marks those exact words—missing all the good stuff like "A15 Bionic chip, zero lag." The answer's right there, just invisible.

This is especially frustrating in RAG systems where retrieval gets the semantics right, but users still have to hunt through long documents.

We tackled this head-on and 𝐨𝐩𝐞𝐧-𝐬𝐨𝐮𝐫𝐜𝐞𝐝  the solution: 𝐚 𝐛𝐢𝐥𝐢𝐧𝐠𝐮𝐚𝐥 𝐬𝐞𝐦𝐚𝐧𝐭𝐢𝐜 𝐡𝐢𝐠𝐡𝐥𝐢𝐠𝐡𝐭𝐢𝐧𝐠 𝐦𝐨𝐝𝐞𝐥 that actually understands context. It outperforms existing options (Provence, OpenSearch, etc.) with an 8K token window and works in both English and Chinese.

It's MIT-licensed and ready to use. Thought folkmilvus.io/blog/zilliz-tr…might find it ushuggingface.co/zilliz/semanti…JwwmZUW

Hugging face: https://t.co/dEBPDs1kXH

———
👉 Follow Milvus, created by Zilliz, for everything related to unstructured data!