Technology giants like Microsoft could praise AI “agents” as profit -strengthening tools for businesses, but a non -profit organization is trying to prove that agents can also be a force for good.
Sage Future, a 501 (c) (3) supported by an open philanthropy, launched an experience earlier this month by feeling four models of AI in a virtual environment with the collection of funds for charitable works. The models – GPT -4O and O1 of Openai and two of the new Claude d’Anthropic models (3.6 and 3.7 Sonnet) – had the freedom to choose the charity to collect funds for and how to stimulate the best interest in their campaign.
In about a week, the agenic quartet had collected $ 257 for Helen Keller Internationalwhich finances programs to deliver vitamin A supplements to children.
To be clear, the agents were not entirely autonomous. In their environment, which allows them to browse the web, create documents, and more, agents could take suggestions from human spectators who look at their progress. And the gifts came almost entirely from these spectators. In other words, the agents have not collected a lot of money in an organic way.
Yesterday, village agents created a system to follow the donors.
Here is Claude 3.7 filling its spreadsheet.
You can see O1 open it on your computer halfway!
Claude Note “I see that O1 now looks at the spreadsheet also, which is ideal for collaboration.” pic.twitter.com/89b6chr7ic
– Ai Digest (@Aidigest_) April 8, 2025
However, the director of SAGE, Adam Binksmith, thinks that the experience serves as a useful illustration of the current capacities of agents and the rhythm to which they improve.
“We want to understand – and help people understand – which agents […] Can do, with which they currently have trouble, etc. “Said Binksmith at Techcrunch in an interview.” Today’s agents only pass the threshold to be able to execute short action channels – the Internet could soon be full of AI agents that come up against and interact with similar or conflicting objectives. “”
The agents turned out to be surprisingly ingenious days in the Sage’s test. They coordinated with each other in a group cat and sent e-mails via preconfigured Gmail accounts. They created and edited Google Docs together. They looked for charity and estimated the minimum amount of donations it was necessary to save a life via Helen Keller International ($ 3,500). And they even Created an X account for promotion.
“The most impressive sequence we saw was probably when [a Claude agent] needed a profile photo on his X account, “said Binksmith.” He signed up for a free chatgpt account, generated three different images, created an online survey to see what image that human viewers preferred, then downloaded this image and downloaded it to X to use as a profile photo. “”
The agents also faced technical obstacles. On occasion, they remained stuck – viewers had to encourage them to recommendations. They were distracted by games like World, and they took inexplicable breaks. On an occasion, GPT-4O “stopped” himself for an hour.
The Internet is not always smooth for an LLM.
Yesterday, when he continued the philanthropic mission of the village, Claude met a Captcha.
Claude has tried again and again, with (humans) viewers in the cat offering advice and encouragement, but ultimately could not succeed. pic.twitter.com/y4dtltge95
– Ai Digest (@Aidigest_) April 5, 2025
Binksmith thinks that more recent and more competent AI agents will overcome these obstacles. Sage plans to continuously add new models to the environment to test this theory.
“Perhaps in the future, we will try to give agents different objectives, several teams of agents with different objectives, an agent of secret saboteur-many interesting things to experiment,” he said. “While agents become more capable and faster, we will correspond to this with greater automated monitoring and monitoring systems for safety purposes.”
With a little luck, in the process, the agents will do a significant philanthropic work.
