Morpheus (@MorpheusAIs) header
Morpheus (@MorpheusAIs) profile picture

Morpheus is the infrastructure that powers the internet of tomorrow and its universe of agents.

Joined September 2023
Photo by Morpheus (@MorpheusAIs)Photo by Morpheus (@MorpheusAIs)Photo by Morpheus (@MorpheusAIs)Photo by Morpheus (@MorpheusAIs)Photo by Morpheus (@MorpheusAIs)Photo by Morpheus (@MorpheusAIs)Photo by Morpheus (@MorpheusAIs)Photo by Morpheus (@MorpheusAIs)Photo by Morpheus (@MorpheusAIs)Photo by Morpheus (@MorpheusAIs)
OpenRouter's DeepSeek traffic fits on one rack. The second rack shows up when the apps and agents do.
ok to put what i'm saying about demand in context openrouter is the most successful model router, you can almost serve 100% of their deepseek traffic yesterday on a single GB300 rack this is worth 6-7M...even if there's 10x this demand secretly that's still < 100M of compute
1
3
1
16
1,265
Your agent. Your keys. Your data... and now your paycheck. Eric is taking what's been started with personal AI and pointing it at the whole machine economy. Proven work gets paid without fear of a government or tech overlord mashing a kill-switch.
Machines will create trillions. Who gets paid? I call it the Autonomous Mesh. No one owns the mesh. Thank you @DJohnstonEC for the idea that started it, and @MorpheusAIs for showing the way. #AutonomousMesh #AI View article
1
7
27
1,163
Mining Bitcoin in the shower? Not possible. Right? find out more in this week's newsletter (link below)
1
4
1
14
1,088
Morpheus retweeted
AERO WILL BE THE UNIFIED LIQUIDITY LAYER FOR ALL OF ETHEREUM.
46
74
12
706
35,299
Regulatory capture in action...
DHS just told every civilian federal agency to rip out DeepSeek.... Not just the app... "Any third-party services developed using DeepSeek implementations" too They think data on servers in China. Prompts sent to China. Malware. Open weights on my spark don't phone anyone, No server in China. No server anywhere. Just my little gold box A hosted app and open weights you run yourself are not the same risk. Please don't lump them together and you push builders off open models..
1
5
1
13
1,417
This is a very cool (and totally weird!) model... all normal LLMs create text through token-by-token inference, printing out each word sequentially. Diffusion text models like Mercury create the entire output simultaneously and then it appears at once (this is normally how image models work) I don't understand it but it works FAST
Mercury 3 is live on Venice. Inception’s diffusion model runs 4x faster than Haiku 5.5 and 8x faster than GPT-6 Luna, beats both on agentic search and coding. Available anonymously.
41
22
1
430
25,735
"Just remove open source from the civilian agencies. That's it. We're not coming for you. We'd never restrict what you use." "What's that? That's a nice project. Built on Chinese compute? Interesting. Interesting." "What? You think we launched that lawsuit? No way. Why would we do that? We definitely don't want to shut anyone down who's using AI we can't control." "Told to use American AI or else? That's so unreasonable. But you know, we can't let China win. America first. It's not very American to steal everyone's data? What? We're not stealing. We don't have control of that. Your data is SAFE." "A leak? Some competitor came along and is using your very same ideas and system? It's a coincidence. Or maybe someone spied on you at a coffee shop. That's it. Privacy glass can't hide everything." "You're bankrupt? What? They're threatening to take everything? You can't keep up with the compute costs and the legal fees and now your family's out on the streets because you can't afford rent? Don't worry! We've got a job for you. It's in the permanent underclass fitting shapes into those toddler toys to train AI's spatial awareness." "You don't like that idea? You wish we hadn't taken that Chinese AI away from you? How ungrateful! How anti-American! Guards!" "Should've supported us. You won't like what happens now." Or adopt permissionless compute before the restrictions hit. Link in reply.
The Department of Homeland Security is requiring all civilian federal agencies to remove DeepSeek products and 'any third-party services developed using DeepSeek implementations,' and to stop buying them.
1
5
15
1,576
Rejoice 16 GB M6 Mac mini owners 🍎 Ternary Bonsai 2 27B now runs on the Neural Engine: - 52 tok/s on code (GPU: 36) - 300 tok/s prefill (GPU: 185) - 12 W (GPU: 30 W) - Works with Grok Build Reproduced on a second M6 mini. github.com/cowboycoderhq/Ter…
4
5
4
43
11,099
For anyone using OpenCode: Stop. Switch to Morpheus. mordiem.com
🚨 Confirmed. OpenCode hacked. Keys stolen. Telling customers to revoke.
8
1
21
1,972
Apologies for the late Week in Morpheus, I was out of town for a funeral and couldn't get the hotel Wifi to load properly. Should've brought the Starlink! Anyway, here it is, talking about the power of open source models punching above their weight class... (Link below)
3
5
20
1,686
Meta’s Muse wonder is, “much clearer and smarter output than Astra” when working on physics
Got to try this model early over the past few days. Tried it out on some really hardcore physics tasks and it gave me a much clearer and smarter output than Astra. The vibes were absolutely on point, so these benchmarks make sense! Super exciting to see Meta push the frontier
5
12
1,242
Solid talk on the future of private money and the network of intelligence NEAR. Congrats to our friends at Venice. PS. Hope you caught the flight @TrustlessState. * borrowed from Haseeb's presentation
Back in Austin post Singapore. TLDR: ZCASH, NEAR & all things quantum are hot. @balajis landed the first Nation State (Kazakhstan) to go all in on being a “digital state” complete with competition for providing traditional government services by private companies.
6
19
911
Back in Austin post Singapore. TLDR: ZCASH, NEAR & all things quantum are hot. @balajis landed the first Nation State (Kazakhstan) to go all in on being a “digital state” complete with competition for providing traditional government services by private companies.
6
8
1
44
1,975
Devs are making the Mac mini's Neural Engine sing. 27B model, 52 tok/s on code, under 12 watts, on a 16 GB box. Local AI keeps getting better. This is why we're bullish on owning your own AI. A big congrats to @CowboyCoderHQ turning the new mac mini into a finely tuned machine.
Rejoice 16 GB M6 Mac mini owners 🍎 Ternary Bonsai 2 27B now runs on the Neural Engine: - 52 tok/s on code (GPU: 36) - 300 tok/s prefill (GPU: 185) - 12 W (GPU: 30 W) - Works with Grok Build Reproduced on a second M6 mini. github.com/cowboycoderhq/Ter…
3
3
19
1,284