Unless you’ve been air-gapped from society this year, you know that chatbots powered by LLMs (large language models) are everywhere, from search engines to your email inbox. You can perform a lot of interesting tasks with ChatGPT, Bing Chat or Google Bard -- but if you’re looking for answers or advice, you can’t always trust them.
The big AI bots grab information from all over the Internet and mash it together, often without citing their sources. If they don’t have information, they can “hallucinate” or make things up to fill in gaps.
Readers deserve a tool that combines the interactive, natural-language capabilities of AI with advice and testing from humans that they can trust. That’s why we’re rolling out HammerBot: a chatbot trained on the expert reviews, buying guides, tutorials and news from Tom’s Hardware. The bot answers your questions both with text and – so you can get all the details – a series of links to our articles on the topic you asked about.
Like everyone else in tech – Google and Microsoft included – we’re still learning how to use AI best. Right now, the HammerBot is in early beta form as we refine the training, the code and the overall UI. But we want to share the bot with you and get your feedback now, not only about how it works now but how you’d like it to work.
Please share your feedback on this forum thread. Scroll past the chat window for more information on how HammerBot works, its known issues and our goals for it.
How the HammerBot Works
We also want to share details about how HammerBot works on the backend. Right now, our developers are using OpenAI’s text-davinci-003 model, otherwise known as GPT-3, trained on a custom dataset of our articles. The data is stored in a vector database from Weaviate and the bot is coded primarily with Python, using LangChain, a framework that makes it easy to customize AI output.
When you enter a prompt, the server queries against the dataset that’s stored in Weaviate to get the search results. Those are then sent to the LLM to help it develop a consistent response to your question.
A lot of chatbots will talk to you about anything on Earth. But don’t ask HammerBot for a knitting pattern! It’s designed to have limits on what it will talk about; it focuses on the expertise you can only get from Tom’s Hardware, so it may say it doesn’t know or can’t answer if prompts that fall outside of its training.
We know you’ll find ways that the output of HammerBot could be better. Here are some you may encounter:
- Answers sometimes out of date: occasionally recommends last-gen products
- Answers not always the top link: The top link in the search result may not be the one most directly related to the chat answer (ex: the answer mentions Ryzen 5 5600X but the top result isn’t the Ryzen 5 5600X review).
- Recommended offers: section may not have a product that matches the chat.
- Fairly short answers: Most of the answers are a bit terse right now.
- Chat window may zoom in when you enter input on mobile.
- May express opinions that aren’t necessarily those of Tom’s Hardware.
More and Better Coming Soon
Aside from just working out some of the known (and unknown issues), we want to make HammerBot a much more powerful tool by adding new capabilities including access to structured data: benchmarks, product specs and up-to-date pricing. Right now the model gets some of this data from article texts, but it’s not as organized in that format.
We want you to be able to ask our bot to compare two specific CPUs and be shown a chart with both of them in it and a side-by-side table with the specs of each. And we want the bot to be able to help you build a PC shopping list, but we’re not there just yet.
What would you like the HammerBot to do for you? Tell us in this thread.