华盛顿邮报在其主页、应用和文章中嵌入新的气候聚焦AI聊天机器人Climate Answers，利用其丰富报道回答气候变化等问题，该工具基于OpenAI模型，也在试验其他模型，且强调避免错误信息。

🌍Climate Answers是华盛顿邮报推出的AI聊天机器人，旨在利用该报的大量报道来解答关于气候变化、环境、可持续能源等方面的问题。它可以根据用户的提问，如是否应为家安装太阳能板、美国哪里海平面上升最快等，利用训练数据提供总结回答。

📄该聊天机器人使用华盛顿邮报气候板块的文章作为信息来源，最早可追溯到2016年该板块的启动。在回答下方，会提供用于生成答案的文章链接及相关片段。

🤖Climate Answers基于OpenAI的大语言模型，同时华盛顿邮报也在试验Mistral和Meta的Llama等AI模型。对于可能的错误信息，该报表示若不知道答案，会如实告知，而非编造。

📣华盛顿邮报并非唯一依靠信息存档来驱动AI聊天机器人的新闻媒体，其他媒体也有类似举措。此外，华盛顿邮报还在逐步扩大AI的应用，该聊天机器人未来可能扩展到其他主题。

The Washington Post is sticking a new climate-focused AI chatbot inside its homepage, app, and articles. The experimental tool, called Climate Answers, will use the outlet’s breadth of reporting to answer questions about climate change, the environment, sustainable energy, and more.

Some of the questions you can ask the chatbot include things like, “Should I get solar panels for my home?” or “Where in the US are sea levels rising the fastest?” Much like the other AI chatbots we’ve seen, it will then serve up a summary using the information it’s been trained on. In this case, Climate Answers uses the articles within The Washington Post’s climate section — as far back as the section’s launch in 2016 — to answer questions.

Image: The Washington Post

“We have a lot of innovative and original reporting,” Vineet Khosla, The Washington Post’s chief technology officer, said during an interview with The Verge. “Somewhere in the years and years of the data-rich reporting we have done, there is an answer buried in one of the things we have written.”

Beneath the answer, you’ll find links to the articles that the chatbot used to produce its answer, along with the relevant snippet it pulled its information from. The tool is based on a large language model from OpenAI, but The Washington Post is also experimenting with AI models from Mistral and Meta’s Llama.

Image: The Washington Post

When asked about the possibility of misinformation, Khosla said Climate Answers won’t produce a response for questions it doesn’t have an answer for. “Unlike other answer services, we really are baking this into verified journalism,” Khosla said. “If we don’t know the answer, I’d rather say ‘I don’t know’ than make up an answer.” However, we plan to try the tool when it launches today to get a sense of its guardrails.

The Washington Post isn’t the only news outlet that’s relying on its archive of information to power an AI chatbot. In March, the Financial Times started testing Ask FT, a chatbot that subscribers can use to get answers about topics related to the outlet’s reporting. Meanwhile, other publishers, like News Corp, Axel Springer, Dotdash Meredith, and The Verge’s parent company, Vox Media, have jumped into licensing partnerships with OpenAI.

The Washington Post has been gradually building on its use of AI; according to Khosla, the outlet has also rolled out AI-powered summaries for some of its articles. Even though The Washington Post’s new chatbot is only able to field climate-related questions for now, Khosla didn’t rule out the possibility of expanding it across other topics the outlet covers. “We absolutely expect this experiment to extend and scale to everything The Washington Post does,” Khosla said.