Pinned Tweet
Agora v0, released in January. It's taken such different shapes for clients, almost ready to make v1, after these last few deployments.
Huge respect for platform engineers, wow is it hard to prune FDE work, have to be meticulous about what becomes part of the core product.
1
1
1,886
If your AI agent isn't running at 2000 tokens/second while ordering goods with 1-2 day delivery, sorry ngmi.
Apparently not everyone wants a Mac mini, Whoop, or Reachy robot for Christmas. Learning a lot about the people in my life...
Kimi K3.1 Pro by @Kimi_Moonshot running at up to 2,000 tokens/sec on @cerebras 🎄
’Tis the season of open models. Tell us your current favorite, and we’ll send you our annual Cerebras swag for your stocking :)
3
41
Cursor sub was great because you could use any model, but now it's just composer and grok. I want to use Gemini models but where?!?
58
This is very bad.
Looks like @PrimeIntellect was compromised in a very strange way. No public comms from them yet, but disclosed by OpenRouter to affected folks (such as myself)
46
Andrew Ross retweeted
Wow what an amazing coincidence, not a single one of your opinions poses a risk to your personal social status
Fly brain neural nets opened the argument against brutalizing generated beings; without knowing how GTA6 NPCs "think", this feels wrong (personally).
Extrapolate a few years / decades, I think simulation will need better governance.
xitter.gq/JerichoGTA/status/2123…
144
The leather jacket is chefs kiss 🤌
Travis Kalanick says he’ll pull up to Nvidia HQ with employees ready to walk out for Jensen: “We don't have control of the outcome, but let's fucking fight."
"I obviously experienced this. I went through six months of ops before I went down. And I held on as tight as I could, but eventually I just couldn't."
"But any wisdom I can provide about that experience to Jensen or the team of people around him that I think are ultimately going to lean in to fight to get him back, because it matters so much, that's what I'm bringing to the party."
"I sent Jensen the screenshot of the stories when there was a petition for me. I'm like, 'petition good, walk out better'. I didn't get a response."
"But I will show up to NVIDIA HQ with a wafer encrusted megaphone with tons of NVIDIA employees who want to walk out, tons of CEOs, founders who believe in what I just said, who are believers and don't want an AI overlord. And let's fight for what's right and give it a best shot. We don't have control of the outcome, but let's fucking fight."
1
85
The dot com bubble was really like that huh. Wish I could've experienced as a sentient entity
$AKAM
3
130
Pivot to RL envs for Saudi's skidding Toyota SUVs
Researchers have used reinforcement learning to teach a 5-tonne John Deere skid-steer loader to perform complex acrobatic manoeuvres autonomously. Trained in simulation and transferred to a real machine, the system can balance on two wheels, navigate rough terrain, spin, and control its bucket before landing safely.
The research demonstrates the potential of AI-driven automation for heavy construction and industrial machinery.
Video Credit: @leggedrobotics / @ETH
Paper: wevlv.co/4kltYys
#engineering #technology
1
66
1) AI native is a moving target with a spectrum of definitions depending on the company (example: Jev / decision models open new set of solutions to deploy)
2) It's common to be severely kneecapped by the client's existing tech stack / data policy (FDEs work within a constrained environment)
3) Agents are only like 20% of the problem. Data ingestion/sterilization, decision trees, ontology, context management, etc. are both up and downstream of Agents, and require many more human hours unpacking firm knowledge than designing Agents.
4) Software is not solvable, AI Implementation is not solvable. 100% AI / agent native is not possible. (Ex. Once great system prompts are technically overfitting after a few months)
I have yet seen a single successful AI FDE that turns a legacy company into true agent-native operation. Prove me wrong
1
2
115
Codex agent's aren't allowed to operate a terminal via computer use. AI safety community should probably focus on securing the airports / NSA, instead of blocking local dev workflows
1
73


















