The intelligent agent Agent Q recently released by MultiOn claims to have achieved an astonishing success rate of 95.4% in real tasks, which has attracted widespread attention in the industry. Its CEO frequently uses strawberry emoticons on Twitter, which is more reminiscent of OpenAI’s mysterious Q project, triggering a lot of speculation about the technology behind Agent Q. Agent Q combines technologies such as search, self-reflection and reinforcement learning to plan and self-heal, and significantly improves task completion rates through autonomous data collection. In the real booking task of Open Table, it improved the zero-sample success rate of LLaMa-3 from 18.6% to 81.7%, which is impressive.
What’s even more striking is that MultiOn CEO Div Garg frequently uses strawberry emoticons on Twitter, which reminds people of OpenAI’s mysterious Q project.
Netizens are full of curiosity about the technology behind Agent Q. Some people speculate that there may be support from OpenAI’s Q* project behind this. MultiOn not only opened an independent Twitter account for Agent Q, but the background image and basic information of the account are related to strawberries, which undoubtedly increased people's curiosity about the technology behind it.
Agent Q combines search, self-reflection and reinforcement learning to enable planning and self-healing. It addresses the limitations of previous LLM training techniques by introducing a new learning and inference framework, enabling autonomous web page navigation.
In the task of simulating an online store, Agent Q demonstrated its powerful search capabilities. In the real booking task of Open Table, Agent Q increased the zero-sample success rate of LLaMa-3 from 18.6% to 81.7%, with a score increase of 340%, and only after one day of independent data collection.
Although Agent Q performed well in the evaluation experiments, there is still much room for discussion and improvement in the methods currently used. For example, the design of inference algorithms, the selection of search strategies, and online security and interaction all require further research and optimization.
The emergence of Agent Q is undoubtedly a major progress in the field of AI agents, but whether it can become an upstart in the AI field or is just a clever hype remains to be tested by time. In any case, the release of Agent Q brings new possibilities and revelations to the development of AI.
References:
https://www.multion.ai/blog/introducing-agent-q-research-breakthrough-for-the-next-generation-of-ai-agents-with-planning-and-self-healing-capabilities
Agent Q’s success rate and technological innovation are impressive, but the technology behind it still needs further verification and improvement. In the future, AI agents like Agent Q will play a role in more fields, promote the continued development of artificial intelligence technology, and bring more convenience to people's lives.