nowfound

AI · December 8, 2022

WS

Web search using a ChatGPT-like model that can cite its sources

We’ve trained a generative AI model to browse the web and answer questions/retrieve code snippets directly. Unlike ChatGPT, it has access to primary sources and is able to cite them when you hover over an answer (click on the text to go to the source being cited). We also show regular Bing results side-by-side with our AI answer. The model is an 11-billion parameter T5-derivative that has been fine-tuned on feedback given on hundreds of thousands of searches done (anonymously) on our platform. Giving the model web access lessens its burden to need to store a snapshot of human knowledge…

Visit beta.sayhello.so ↗Alternativestop 4% of December 2022

In plain words

This is a generative AI search tool that answers questions and retrieves code snippets by browsing the web and citing its sources. Users can hover over answers to see which sources are being cited and click through to the original content. The tool displays AI-generated responses alongside traditional Bing search results. Built on an 11-billion parameter T5-based model fine-tuned from hundreds of thousands of anonymous searches, it prioritizes accuracy and cost-efficiency by accessing current web information rather than relying solely on stored knowledge.

written from the facts on this page · September 2026

From the sources

In the maker’s words, at launch

We’ve trained a generative AI model to browse the web and answer questions/retrieve code snippets directly. Unlike ChatGPT, it has access to primary sources and is able to cite them when you hover over an answer (click on the text to go to the source being cited). We also show regular Bing results side-by-side with our AI answer. The model is an 11-billion parameter T5-derivative that has been fine-tuned on feedback given on hundreds of thousands of searches done (anonymously) on our platform. Giving the model web access lessens its burden to need to store a snapshot of human knowledge within its parameters. Rather, it knows how to piece together primary sources in a natural and informative way. Using our own model is also an order of magnitude cheaper than relying on GPT. A drawback to aligning models to web results is that they are less inclined to generate complete solutions/answers to questions where good primary sources don’t exist. Answers generated without underlying citable sources can be more creative but are prone to errors. In the future, we will show both types of answers. Examples: https://beta.sayhello.so/search?q=set+cookie+in+fastapi https://beta.sayhello.so/search?q=What+did+Paul+Graham+learn... https://beta.sayhello.so/search?q=How+to+get+command+line+pa... https://beta.sayhello.so/search?q=why+did+Elon+Musk+buy+twit... Would love to hear your thoughts.

More ai this month

the category →
  • I trained a 125M model to autocomplete piano on-device▲598

    I trained a 125M-parameter transformer to autocomplete piano performances in real time (~108 notes/sec on an iPhone 15). The idea is basically GitHub Copilot or Tabnine, except instead of prompting it with code, you prompt it by playing a few notes on a MIDI piano. The model then continues what you played, entirely on-device. The app is free if anyone wants to try it. Happy to answer questions about the model, training, Core ML, or the many things that didn't work.

    AI · 17d ago · simedw.com

  • Astute▲585

    Automate your B2B brand going viral, with new media creators

    AI · 18d ago · company-app.joinastute.com

  • Grok Bot▲547

    AI teammates that you can give real work to

    AI · 25d ago · x.ai

  • Needle2: 14MB agentic LLM for phones, wearables, smart home and robots▲537

    Hey HN, Henry from Cactus here! We previously released Cactus Needle, a 14MB agentic LLM for tool call, device use, and structured extraction for phones, wearables, smart homes, small robots and microcontrollers. We got really great feedback here, and have now incorporated the suggestions to release Needle 2. The whole model is a single 14MB binary that runs a full session in 28MB of RAM; 45m parameters at 2bit compression. Needle hits 500 tokens/sec decode speed on a Raspberry Pi 5, sits between 400-1,500 tokens/sec on VR devices like Meta Quest 3S and Apple Vision Pro, and ranges…

    AI · 27d ago · cactuscompute.com

  • Clara AI SDR▲460

    Turn website visitors into qualified pipeline

    AI · 19d ago · clarasdr.ai

  • Kane CLI▲446

    Natural language browser & mobile app tests from terminal

    AI · 24d ago · testmuai.com

Launched alongside, December 2022

the whole month →