Earth
Joined April 2010
- Tweets2.6K
- Following305
- Followers286
- Likes2.4K
You were INCREDIBLE, @kirahsapong ! People loved your talk and had so many questions for you and you answered all of them with such energy and passion.
You are such a positive force for this industry!
This is my appreciation post for @danieldafoe who always does so much and always with a smile 🥺
Such an incredible week with him at @WeAreDevs, and many firsts
(Caught the duck just as it was headed down for a nap, we also needed a nap)
Booth 10 goodies and raffle items @WeAreDevs !
Just outside hall 1+2.
Come pull up!
@supermodellabs @aquaduckai
LLM sampling errors == MASSIVE gains 💪
ALT Bullet point 1: Barbell Curl, 3 to 34 sets by 8 to 10 reps
Access your local models globally.
📍-> 🌏
One of Aquaduck’s biggest differentiators is in how it executes distributed inference across a team cluster. Almost every framework I’ve seen does it over RDMA connected cables, so participating devices have to be physically right next to each other.
We coordinate clusters over the internet, so you don’t even have to be in the same room. The tradeoff is network communication overhead. Petals did this but token speeds were unusable. We made token speeds usable.
For teams operating independent devices across time zones, it’s the best option.
Booth 10 or bust.
We land in San José tomorrow with @supermodellabs .
WeAreDevelopers World Congress North America kicks off September 23rd.
Find us at Booth 10 all three days.
wearedevelopers.us
#wearedev
Don’t miss this event! (And @kirahsapong ‘s talk on the state of local AI. 📍💻)
We have 2 tickets to give away 🎟️
Excited to speak this Friday at @WeAreDevs San Jose, CA on the state of local AI in 2026. There’s a LOT to cover.
If you’re in the Bay Area this week and want to join us at #WeAreDevs, let me know!
Sep 23-25, wearedevelopers.us
Daniel R. Dafoe retweeted
If you're interested in WebMCP, you should be following Idan Levin. He and his team have been doing extremely exciting work in this space- benchmarking, cataloguing, you name it!
Strong suggestion to follow him.
You might find this interesting -
We created the first WebMCP benchmark, comparing it with computer use and DOM-based methods across speed, cost, task success rate, and token usage. The results are incredible.
webmcp.com/benchmark
You’re getting 3-5x speed, and an order of magnitude cost improvement (up to 23x lower costs) while being 10-15% better at solving tasks compared to other methods
It's both fascinating and strange that spelling mistakes in LLM prompts force unnecessary reasoning and therefore tokens and therefore cost.
Are people working on robot Pokemon with physical AI? Imagine they were created en masse and just released into the wild so they could be "caught".
Thanks for putting this together, @rajatsuri ! Excited to be a part of this.
Anyone trying to optimize agent costs is going to love this one!
I’m looking forward to speaking at the Agentic AI Foundation #AGNTCon + #MCPCon North America this October 22-23 in San Jose, California.
The agenda is packed with conversations around what it really takes to build and scale agentic AI systems in production, from orchestration and infrastructure to evaluation, resilience, observability, and debugging.
I hope to see you there!
Register now + join me 👉 bit.ly/44MI7LM
🔥 Explore the schedule: bit.ly/4f6Pp35
Welcome to local AI and then some! Connect as many of your devices as you want into your own on-prem inference network and access it from anywhere.
Come try it out! aquaduck.ai/sign-up
💻💻🌎💻💻
This quoted post is unavailable.