Cloud

Manus: Better than OpenAI thanks to AI autonomy?


Besser als OpenAI dank KI-Autonomie?
The new OpenAI challenger Manus is designed to create personalized solutions and provide detailed evaluations without any human intervention.

shutterstock – QINQIE99

The next would-be “OpenAI killer” is coming from China, after DeepSeek set out to dethrone the industry leader in 2024. The new artificial intelligence goes by the name of Manus and was developed by the startup Butterfly Effect. According to the company, its product is a “general agent” designed to improve the tools offered by Western companies.

Manus is best compared to OpenAI’s Deep Research, a service that searches online services to find information. What is found is then compiled into documents which, according to OpenAI, should produce a comprehensive report at the level of a research analyst within half an hour.

Other points of reference are tools such as Anthropic’s Computer Use API and OpenAI’s Operator Agents. Both tools can be used in a web browser to perform basic tasks such as filling out forms and using e-commerce sites.

Manus claims to be able to do all this and much more and, according to its own benchmarks, it should also be faster than the competition. In a video for the market launch, the AI agent is shown performing three tasks at top speed.

Multi-tasking

It is intended to take on a variety of tasks, including analyzing job applications, creating property reports and carrying out stock correlation analyses.

According to the creators, the special thing about this is that the tool goes beyond simple queries by expanding tasks independently. For example, it can not only rank CVs, but also analyze skills and compare them with market trends in order to make optimized hiring decisions.

For real estate inquiries, Manus considers additional factors such as crime rates, rental trends and weather conditions to provide a customized selection of properties.

While ChatGPT-4 and Google’s Gemini rely on human prompts, Manus is designed not to wait for instructions. Instead, it is designed to tackle tasks independently, evaluate new information and adapt its approach dynamically. According to Forbes, Manus should not only focus on the task at hand, but also independently expand the task with additional, helpful parameters.

Cloud-based Ubuntu workstation

Manus presents itself with a chatbot user interface including a blank text field for prompts. Early testers have described their experience to news portal The Register as being like sitting with someone at a keyboard, turning vague instructions into precise output with extraordinary speed.

The service runs on “Manus’ Computer,” which is apparently an Ubuntu workstation in the cloud. The launch video states that the service operates as a multi-agent system based on several different models. Some of these models are to be released as open source later this year.

Reality check

However, testers have already expressed their first doubts about Manus: The Register, for example, had the AI agent create an itinerary for a two-month trip to “Australia, then New Zealand, Argentina (and other parts of South America) and Antarctica”. Citing just 17 sources, Manus claimed that a flight in business class was only twice as expensive as economy class.

In reality, however, such a flight costs at least three times as much. According to the developers, the bot arrives at such results by opening several browser windows simultaneously to collect data and analyzing them at lightning speed. The AI also did not provide the promised downloadable summary.

The key to its performance is said to lie in its multi-agent architecture. When Manus is assigned a complex task, it divides the problem into manageable components, assigns them to the appropriate agents and monitors their progress. This structure allows the tool to handle multi-step workflows that previously required multiple AI tools to be manually stitched together.

Cloud-based asynchronous operation is said to be another advantage. With conventional AI assistants, users have to lend a hand themselves — Manus does not need them, according to the developers. It carries out its tasks in the background and only reports back when the results are available.

Party-poopers not allowed

What the agent cannot do, however, is criticize the Chinese Communist Party. Prompts that lead to an answer that puts those in power in a bad light lead to mistakes.



READ SOURCE

This website uses cookies. By continuing to use this site, you accept our use of cookies.