CEO AIFlow Labs https://teapaw.org/t.co/L27ts7lVwj CTO @datainetwork AI Advisor
Phuket
Joined October 2023
- Tweets16.1K
- Following5.8K
- Followers4.4K
- Likes67.3K
Pinned Tweet
Container Desktop v0.1.0 is out: an open-source, native Mac window for Apple's container runtime. Run, inspect and manage Linux containers in a layout you already know. MIT licensed.
Article: teapaw.org/AIFlow_ML/status/21121…
Repo: github.com/AIFlow-Labs-Limit…
#macOS #opensource
Fable is now my nutrition coach and is absolutely amazing ! So I did not eat enough he ping me to propose me other food to eat hahaha
I'v finally got the Cyber on @AnthropicAI !!!
Fantastic !!!
We should stop use SI and AGI buzzwords that are on the table just to keep the bubble up and start literally building super efficient models to deploy locally.
When this crack buy GPUs then watch who cannot swarm 300 agents perish.
AI is a big thing that belong to all of us.
They want to control AI but they cannot.
They want 300 new data centers but they cannot build them they are too slow.
They will use our bank account to bail out them.
I ask fable to write me the custom kernels for a project on Jetson Thor he told me his guardrails cannot do it but he pass me a script that did it.
I did not even push for it. But probably my agents love me :D
Four of six robots survived 24h in a real store. Would you hire them?
Robots Never Clock Out | AGIBOT G2’s 24-Hour Convenience Store Challenge
Six AGIBOT G2 take on a 24-hour shift as convenience store trainees.
From rush-hour crowds to late-night solo duty and the morning breakfast sprint, can they keep up with nonstop customers and unexpected requests?
teapaw.org/i/broadcasts/2OdmqandP…
594B open weights next week. Self-host or wait for Tinker?
Introducing Inkling 2 Small. Scoring 31 on the AA Intelligence Index, it's a big jump over its predecessor.
Inkling 2 Small is available for fine-tuning and inference on Tinker today, with full weight release planned soon.
thinkingmachines.ai/news/ink…
$NVDA setup on 3M and 6M compromised.
We are about to see some discounts on DGX. & Co
A general model is now #1 on a 16-benchmark robot leaderboard.
On real robots, the same model got 17 of 18 rigid picks and 0 of 11 plug insertions. In a simulated kitchen it left the fridge open for almost 8 minutes.
Both are true. Here's how I read it:
"done" isn't done until something outside the model checks it; for robots the check must be physical.
micro1 surgical-robot test: 10 models, 800 trials, 2 models finished at all; 213 success claims, all wrong. teapaw.org/micro1_ai/status/21162… @micro1_ai
We gave 10 frontier AI models control of a surgical robot in a simulated operating room.
Across 800 trials, only two models completed the task at all: Gemini 4 Argon nine times, and GPT-6 Astra twice.
Models claimed success 213 times. Every claim was wrong. In 94 of them, the robot never touched the object.
Introducing Dual-Arm Sim, our new robotics benchmark evaluating how well frontier models can control embodiments in simulated environments. More info in thread around what we tested and found.
Sources: Playground index withplayground.ai/blog/robot… ; arXiv report arxiv.org/html/2610.22471 ; Robocurve openai.robocurve.org/gpt-6-a…
