When I woke up, Sentient @ SentientAGI also released its own ranking list https://yaps.kaito.ai/sentient Fortunately, I have written some articles about Sentient before, currently ranked 15th in 30D As the most capable AI project at present, Sentient is undoubtedly a very important object worth paying close attention to Recently, they have launched a meta agent framework ROMA, which means that AI is no longer just a single brain, it has become a team that can command. Let's explore it further ROMA test data Seal-0: 45.6% - This score is not high, but it indicates that ROMA is already capable of handling some complex reasoning problems in the early stages - a single large model often performs lower on such tasks FRAMES: 81.7% -81.7% accuracy indicates that ROMA performs well in integrating information from different sources and formats SimpleQA: 93.9% -93.9% is close to the perfect score, indicating that ROMA is stable and reliable in handling single, direct tasks The key components/processes of ROMA are roughly as follows ① Atomizer: Determine whether a task is "atomic", that is, whether it can be executed directly; If not, further decomposition is needed ② Planner: Plan/decompose non atomic tasks and generate subtasks ③ Executors: Execute "atomic" tasks or subtasks, which can be calling LLM, API, or other agents ④ Aggregator: Merge the results of subtasks to form the final output of the parent task What problems has ROMA solved ① Task decomposition+parallel processing - breaking down complex tasks into subtasks for parallel processing. This improves efficiency and makes the system more flexible when facing large tasks ② Multi agent/multi provider+tool integration - supports different LLM providers, APIs/tools, which means high flexibility ③ Open source+community oriented - very developer friendly. Can quickly test and prototype develop ◇ ROMA application scenarios Even if a framework is built completely, it still needs to have application scenarios. Currently, ROMA has a wide range of application scenarios: ① Research tasks: such as academic research, market analysis, and integration of competitive intelligence ② Content creation+long document generation: such as reports, white papers, guides, etc ③ Complex problem-solving tasks: such as product design, project planning, strategic planning, etc ④ Tool based Agent System The launch of ROMA undoubtedly makes decentralized AI a reality. When OpenAI is no longer open, sentient can only explore a path that belongs to Open+AI on its own!