Product @luxonis | Building the vision for a better future Physical AI / Computer Vision / Robotics
Colorado, USA
Joined June 2018
- Tweets168
- Following158
- Followers52
- Likes71
Codex's auto-draft for the next prompt is such a strange feature.
My reaction is 'Wow, thats the perfect next step' then 'wait, why aren't you just doing that automatically'
Almost like they are trying to reinforce that I'm the bottleneck, so I give up more control lol
This is incredible. We are still a long way out from generalized physical intelligence at the levels we see with LLMs, but if the process to fine-tune for applications is accessible, then we are not far off from surpassing traditional methods of robotics (if not already!)
Didn't realize all these startups ran on Pi, definitely top dog in Physical AI now
The Pi API is officially in beta! 🤖
Many people have already been using it to fine-tune π models for their robots -- below are a few
We’re hoping to expand testing significantly. Let us know if you are interested!
Blog: pi.website/blog/api-beta
Signup: pi.website/waitlist
We've been playing a lot with VLMs on OAK and have some pretty exciting results.
Custom trained Yolo can only take you so far, and it takes so much effort to train.
The beginning of a huge breakthrough, bringing OAK closer to human perception
Max Mclaughlin retweeted
Custom model training is now live in Luxonis Hub!
Label your images, train, evaluate, and deploy to OAK. When the model misses something in the field, snap it from the camera and train the next version.
Teams already in hub can try this at no cost.
Read more: discuss.luxonis.com/blog/705…
Max Mclaughlin retweeted
Install the Luxonis Agent Toolkit (Luxonis MCP + skills) now in Claude, Codex, and Cursor for a better way to build with OAK.
It's never been easier to deploy visual intelligence to the edge
Github: github.com/luxonis/skills
Full Release: discuss.luxonis.com/blog/702…
Max Mclaughlin retweeted
Are the normies using Meta Muse or Grok Bot or is this just something that tech bros are telling themselves?
How is a normie using AI agents? Are they? Every AI agent example you see is a tech bro asking an agent to do some completely over-the-top complex task for the sake of complexity. A normal person doesn't do that. A normal person wants to get off work, chill, see their family, play with their pets, eat some food, have some drinks, sleep, repeat.
The examples shown are all strivers trying to maximize every waking minute of their life for productivity.
The normal person would see that and tell you to pound sand and go back to their lives.
What is the use case for the average person?
Opening Twitter has changed from a huge distractor to an amazing motivator to get back to work.
Only takes a couple of posts before feeling, once again, extraordinarily behind
Surely anthropic is posting this in the best interest of humanity and not their business
Surely it’s a coincidence this comes right before their IPO and a day after Deepseek release an on par model at a tenth of the price.
It’s so clear these are marketing stunts to force legislation to slow everyone else down while they are still lead and are in bed with the legislatives. Such foul play…
just play the damn game on the field: on your own GPU’s and make your models cheaper
We're publishing our most detailed threat intelligence report to date.
It covers how people tried to misuse Claude—for cyberattacks, influence operations, surveillance, biology, and building weapons—and how we found and stopped them.
We disrupted every operation in the report, and used the lessons from them to strengthen our safeguards. Where appropriate, we also shared what we found with authorities and other AI companies.
These cases are not typical: we’re highlighting some of the most sophisticated misuse we’ve seen. But they’re especially important to discuss, because they show us where AI misuse is headed, where our safeguards work, and where they need to improve.
We’re publishing this report so others can spot the same activity on their own platforms, and so we can give the public a clearer view of how emerging threats develop.
Read the report: anthropic.com/threat-intelli…
This just initiated my fight-or-flight instinct
If these labs ever go public, they will be trading at the volatility of Bitcoin lol
Thrilled to share a product I've been leading at Luxonis: the Luxonis Agent Toolkit.
Since our founding, Luxonis has been dedicated to building the most open and capable platform in vision. For a long time that came with a trade-off: developers got complete control over the stack, but the barrier to entry into visual intelligence stayed high.
AI agents eliminate that trade-off. The Toolkit delivers the same capable, flexible platform through the coding agent you already use. Describe what you want to build in natural language, and go from an OAK on your desk to a working application in minutes.
Install the Luxonis Agent Toolkit (Luxonis MCP + skills) now in Claude, Codex, and Cursor for a better way to build with OAK.
It's never been easier to deploy visual intelligence to the edge
Github: github.com/luxonis/skills
Full Release: discuss.luxonis.com/blog/702…
Huge release we are super excited about. It's never been easier to give a computer human sight!
Introducing the Luxonis Agent Toolkit, the fastest way to start building with OAK.
Install it in Claude Code, Codex, or Cursor and describe what you want to build. Your agent handles the rest: device setup, examples and models, writing the application, and working with real camera output.
8 skills + the Luxonis MCP server. In beta, open source.
github.com/luxonis/skills
More in 🧵
Does anyone else have trouble with Fable being extremely AI slopish and cringe?
I have anti-slop in my global AGENT.md/CLAUDE.md, my personal wiki, and agent invocable skills, yet it seems impossible for Fable to avoid it.
Such a shame, because I do think Fable is a better reasoning partner than SOL, I just have to spend so much time decoding the slop and making anything actually presentable
