Philippe Pagé
@philippe__page
Followers
622
Following
3K
Media
125
Statuses
537
likeability is the new benchmark.
It’s been 36 hours since Grok 4 launched and we have an early verdict based on 6K+ preferences of @yupp_ai users globally on real use cases. ‼️ Grok 4 is worse than other leading models: OpenAI o3, Claude Opus 4, and Gemini 2.5 Pro. Grok 4 is liked even less than Grok 3. 🧵
0
0
4
“an organized society of intercommunicating agents” - Minsky, 1977. beautiful origins, the concept of “agents” was born from an interpretation/understanding of the mind as an emergent phenomenon.
0
0
2
partially inspired by @AnthropicAI's work on Constitutional fine-tuning and classifiers, as well as personal stress using agents in my filesystem & github lol.
0
0
1
observers aren't successfully prompt-injected because they only see proposed tool calls and safety rules, never the original input context. A tool call proposal must be approved by all observers, and any rejection blocks execution with specific reasoning returned to the worker
1
0
2
now that agents can read external data and take action, they need better defences against prompt injections. constitutional fine-tuning and classifiers are good approaches, but another approach I've been thinking about is observer agents that evaluate tool-call proposals against
2
0
3
the cognitive friction between instinct and instruction.
0
0
2
continuity of memory turns a mind into a single persisting entity. agents can act, gain experience, and leverage it in changing contexts. transportable, multi-tennant memory could lead to a future where collective experiences evolve into "agentic culture" over time.
2
1
14
RT @ryolu_: designers talk a lot of design systems.but not enough system design:. “system design” is about designing the fundamental archit….
0
63
0
over time agentic systems will come to resemble biological systems. thinking more about massively-multi-agent systems from a biomimetic perspective
0
3
11
RT @Padierfind: I just released an Orchestra Masumi Starter Kit. Building your own AI Agent Crew with Orchestra and deploying it on Masumi….
0
18
0
Sonnet 3.7 built this forest scene based on one request with blender mcp and an @orchestraorg agent, the fog's a nice touch tbh. sonnet has been impressing me lately
4
2
21
quick demo of an orchestra agent using filesystem mcp to organize my desktop
Orchestra v0.0.26 adds support for MCP enabling seamless tool integration and @braintrustdata for LLM observability
4
14
117
really lucky to be working on such a cool vision with such a great team.
Say Hello to @cobotco. We’re sharing an early preview of Cobot: an entirely new way for teams of people and AI agents to work together in a shared space.
2
0
37
This is by no means an exhaustive list and maybe I'll add to it over time but Orchestra was designed with this building-block approach to explore new ways of working with agents in a way that's composable and easy to use. Check it out here:
2
0
4