In partnership with
News of the day
1. Generalist AI unveils GEN-1.5, an AI model capable of teaching robots new tasks with just one demonstration. → Read more
2. Meta launches a new Mac app for Meta AI, enabling system-wide dictation and contextual screen analysis for business insights and task automation. → Read more
3. Anthropic is internally testing 'Model 2', an unpublished AI model that surpasses the capabilities of its publicly released Claude versions. → Read more
4. Stripe acquires OpenRouter, an AI model-routing platform, to integrate model selection and routing into its AI usage and billing services. → Read more
Our take
Hi Dotikers!
Some announcements look like just another demo, and others quietly move a line. GEN-1.5, unveiled this week by Generalist AI, belongs to the second group. The idea fits in one sentence: you show the robot a task for three to twelve seconds, the clip goes into the model's context window, and it gives it a try. No retraining, no data collection, no overnight compute run. The researchers call it a physical prompt, and the name fits.
What matters here is not really the success rate, it is the trend line. The company says adapting to a new task used to take hundreds of training steps, then dozens, then a single one, then none at all. That is exactly the slope language models followed before in-context learning became obvious to everyone. And nobody coded that ability in: it showed up on its own after eight months of pretraining on interaction data. The model even chains two demonstrations together and improvises recovery moves that appear in neither recording.
Then comes the cold shower. Fifty-nine percent average success means one jar out of two stays shut and the aperitif gets off to a rough start. The tasks are short, the numbers come from the house, and no outside lab has replayed the score. Generalist raised 400 million dollars in June at a two billion valuation, and nobody publishes this kind of post out of pure love for science.
Still, the link with yesterday is hard to miss. Cursor launched Origin to host code written by agents at a pace no human keeps up with. Generalist, for its part, talks about libraries of short, reusable gestures. After the code repository, the movement repository. Someone will sell the GitHub of gestures long before that jar opens nine times out of ten.
Alex.
Your employees are connecting AI to everything. Now what?
ChatGPT and Claude don't just answer questions anymore. Employees are connecting them directly to Notion, Linear, Jira, and the rest of your stack. The AI can read, write, and take actions on company data. Most IT and security teams have no visibility into any of it.
Harmonic Security Connectors changes that. It sits inline with every AI-to-app connection, so you see each call, control what data moves, and block destructive actions before they happen. Employees notice nothing different.
See what's actually running across your business in a live demo.
Meme of the day
