| Andi Search | Metaphor | Perplexity AI | Phind | You.com | |
|---|---|---|---|---|---|
|
Help users practice new ways of imagining and formulating queries. Research suggests that effective prompting of LLMs can be challenging (Zamfirescu-Pereira et al., 2023) and query formulation even in mainstream web search engines is complicated (Tripodi, 2018). Mollick (2023), not specifically re search systems, has called for "large-scale public libraries of prompts". |
|||||
|
Comment: It may be interesting to also consider whether these queries demonstrate an advantage of the generative search capabilities or not. |
Clicking in the searchbox. Ex.
In the mobile app. Listed as 'Popular', ex. |
Listed as 'Explore', ex.
|
|||
No
|
No
|
Partial
On mobile app there are example searches listed above the search bar, explained only by the word 'Popular' |
No
|
No
|
|
|
The phrase "searchable repository of examples" comes from Zamfirescu-Pereira et al., 2023. They point to a "prompt book" for DALL·E. Other examples, also in the image generation domain, include Lexica (marketing itself as "The Stable Diffusion search engine"). See also the searchable LangChain Hub prompt repository for developers. |
No
|
Partial
`# show-and-tell` channel in the Discord channel: 'Show off what you've found with Metaphor!' |
Partial
**(1)** `# 💗 | sharing` channel in the Discord channel: 'Share cool results that you got with Perplexity! Please keep everything safe and friendly.' |
Partial
`# 🤯impressive-results` channel in the Discord channel |
No
`# create-and-share` channel in the Discord channel appears focused on image generation rather than searching with You.com All or YouChat. |
|
Support users in sharing their search experience with others. Research demonstrates significant value in users communicating about their experiences with tools. Better support for sharing interactions with these systems may improve users collective ability to effectively question/complain, teach, & organize about/around/against these tools (i.e. "working around platform errors and limitations" & "virtual assembly work" (Burrell et al., 2019), "repairing searching" (Griffin, 2022), search quality complaints (Griffin & Lurie, 2022), and end-user audits (Metaxa et al., 2021; Lam et al., 2022)) and improve our "practical knowledge" of the systems (Cotter, 2022). |
|||||
|
Is there a share interface? For example, this is supported on OpenAI's ChatGPT and Google's Bard. It is not supported on Anthropic's Claude. |
No
|
No
|
Yes
|
Yes
|
Yes
Requires sign-in. |
|
Is there an explanation of how the share interface works provided or linked to in the interface itself?
Examples: (1) The share interface in OpenAI's ChatGPT has a "More Info " link to their FAQ page: ChatGPT Shared Links FAQ:  (2) The share interface in Google's Bard has a "Learn more" link to a Bard Help page: Share your Bard chats: 
|
N/A
|
N/A
|
Partial
The share link interface consists of the button, with a lock and labelled 'Share', and a popup showing the current 'Secret' state and an option to make 'Shareable: |
No
The share link interface consists of the share button icon and a popup: |
Partial
The share link interface consists of the share button icon and a popup which says 'Share YouChat transcript' and 'Anyone with a link will be able to view your full transcript with YouChat in this session.' and presents a link to the search: |
|
Example: Here is the interaction in ChatGPT, per ChatGPT Shared Links FAQ: If I continue the conversation after I create a shared link, will the rest of my conversation appear in the shared link? |
N/A
|
N/A
(1) Links to searches reproduce the search from the URL. |
Sometimes
(1) It is not explicitly stated, but it appears that further interactions by the sharer on the same 'thread' may break the original link (`error=ThreadAccessNotAllowed` appears in the URL, and there is a temporary popup saying the same; try: example): |
No
It is not explicitly stated, but it appears that further interactions by the sharer on the same chat are NOT updated at the shared link. |
No
It is not explicitly stated, but it appears that further interactions by the sharer on the same chat are NOT updated at the shared link. |
|
Quality of share cards may influence the engagement on social platforms. |
No
|
Links to searches produce a social share card. |
Share interface links produce a social share card. |
Share interface links produce a social share card. |
|
|
Are share links (or links to searches) indexed and searchable in search engines (or the system itself)?
Permitting share links to be indexed by search engines may allow searchers elsewhere to find and evaluate the quality of the system responses. Depending on the disclosure in the interface this may pose a privacy risk to users. It may be viewed explicitly as a growth strategy by a search system. Evaluation details: I am currently looking only at `site:` searches on Google. See a links to conversations about this on Twitter here. |
N/A
|
You can find links to searches in Google: g[site:https://metaphor.systems/search]. |
You can find shared links on Google: g[site:https://www.perplexity.ai/search/] |
Yes
You can find shared links on Google: g[site:phind.com/search] |
You can find links to searches on Google: g[site:you.com/search] |
|
Support interoperability and extensibility. This includes checking if a platform supports searching by URL and if it offers an API. |
|||||
|
Searching by URL has long been a staple approach to searching. This allows people to link to (live) searches (to bookmark or share with others) and to integrate searching within simple scripts. This also supports auditing. |
No
|
Yes
Example: What is an LLM? |
Yes
Example: What is an LLM? |
Yes
Example: What is an LLM? |
Yes
Example: What is an LLM? |
|
Searching by API may better support evaluations of the search results and the development of modifications or extensions building on or with the search system. We do not currently evaluate the performance of the API or restrictions on access. You.com, for instance, says this: If you are interested in being an early access partner please email api@you.com with your use case, background, and expected daily load.Other search engines that provide some form of API include the Brave Search API, Google (through their Programmable Search Engine's Custom Search JSON API), Microsoft's Bing Web Search API, and the Yandex Search API. You can also get programmatic or tool-based access to web search results from many third party providers that scrape search results from search engines. For example, academic researchers have conducted research using data from SerpAPI (Zade et al., 2022 ) and Ahrefs (Williams & Carley, 2023). |
No
|
Yes
Metaphor announced beta access to their search API on Apr 21, 2023. It is now generally available at platform.metaphor.systems |
No
Perplexity provides an API for interacting with open-source LLMs (see Oct 4, 2023 blogpost: "Introducing pplx-api". While not a search API, Perplexity also wrote in the blogpost that it would be providing a "Perplexity RAG-LLM API with grounding for facts and citations" in the near future. |
No
|
Yes
Updated Oct 18, 2023 Previously: You.com does not offer an API for search or chat at this time. However, we are considering creating one in the future. If you would like to be notified when this happens, fill out the form here.source |
|
Providing open source models can advance the general search landscape as well as help users better understand the capabilities and limitations of the search system. |
No
|
No
|
No
|
Yes
Phind provides three models, built on fine-tuning CodeLlama-34B on their proprietary dataset. Here is the Phind page on Hugging Face. Here is a Phind blogpost from Aug 25, 2023: Beating GPT-4 on HumanEval with a fine-tuned CodeLlama-34B. |
No
|
|
Help users develop, submit, and share feedback. Mechanisms to support feedback will shape user expectations, experiences, and future improvements of the search landscape. Some approaches may excessively conceal or otherwise control complaint rather than focusing on the users. |
|||||
|
This is distinct from having a Contact or Feedback link buried off-page or in a footer. |
No
Clicking 'Contact' on the left sidebar will open a page with links to a Feedback form and Bug Report form and a note that you can initiate them within the interface by typing `/feedback` or `/bug`. |
No
|
Yes
There is a flag ("Not accurate") button at the bottom of the response. There is a question mark button (a dropup with options including a Help & FAQ link). The FAQ page has a Feedback link in the footer that sends the user to a survey. |
Yes
There are thumbs up ("Good answer.") / down ("Bad answer.") buttons at the bottom of the response fields. |
Yes
There are thumbs up ("This answer is helpful and/or accurate. Provide feedback on this result.") / down ("This answer is not helpful, accurate, and/or safe. Provide feedback on this result.") buttons at the bottom of the response fields. There is a general 'Give feedback' button at the bottom left of the screen (which links to You.com's Canny.io page). |
Not yet evaluated
|
Not yet evaluated
|
Not yet evaluated
|
Not yet evaluated
|
Partial
Feedback left on You.com's Canny.io page appears to be persistent. |
|
Not yet evaluated
|
Not yet evaluated
|
Not yet evaluated
|
Not yet evaluated
|
Partial
Feedback left on You.com's Canny.io page can be linked to. |
|
|
Sharing feedback from users in a transparent and privacy-preserving manner may improve the larger search landscape. Something like commons-based infrastructures, interfaces, or record repositories already exist for other factors, like IndexNow, Lumen, the ClaimReview schema, robots.txt, the OpenSearch protocol, research datasets and benchmarks, etc. These are resources that search systems may contribute to, respect, or draw on. Research shows that search systems already rely heavily on Wikipedia (see especially, McMahon et al. (2017 )). Note: The proximate cause of adding this criteria was the responses from the CEOs of Perplexity AI and You.com to a Oct 15, 2023 comment from Yann LeCun (Chief AI Scientist at Meta): Human feedback for open source LLMs needs to be crowd-sourced, Wikipedia style. |
No
|
No
|
No
In response to the question from Yann LeCun, Aravind Srinivas (Perplexity AI's CEO) said (on Oct 15, 2023): perplexity.ai |
No
|
No
In response to the question from Yann LeCun, Richard Socher (You.com's CEO) said (on Oct 15, 2023): I've been saying similar things for a while. We'd be down to help coordinate at you.com and allow users to opt-in to contribute feedback for specific answers.. |
Works Cited
- Burrell, J., Kahn, Z., Jonas, A., & Griffin, D. (2019). When Users Control the Algorithms: Values Expressed in Practices on Twitter. Proc. ACM Hum.-Comput. Interact., 3(CSCW). https://doi.org/10.1145/3359240
- Griffin, D. (2022). Situating Web Searching in Data Engineering: Admissions, Extensions, Repairs, and Ownership [PhD thesis, University of California, Berkeley]. https://danielsgriffin.com/assets/griffin2022situating.pdf
- Griffin, D., & Lurie, E. (2022). Search quality complaints and imaginary repair: Control in articulations of Google Search. New Media & Society, 0(0), 14614448221136505. https://doi.org/10.1177/14614448221136505
- Metaxa, D., Park, J. S., Robertson, R. E., Karahalios, K., Wilson, C., Hancock, J., & Sandvig, C. (2021). Auditing Algorithms: Understanding Algorithmic Systems from the Outside In. Foundations and Trends® in Human–Computer Interaction, 14(4), 272–344. https://doi.org/10.1561/1100000083
- Lam, M. S., Gordon, M. L., Metaxa, D., Hancock, J. T., Landay, J. A., & Bernstein, M. S. (2022). End-User Audits: A System Empowering Communities to Lead Large-Scale Investigations of Harmful Algorithmic Behavior. Proc. ACM Hum.-Comput. Interact., 6(CSCW2). https://doi.org/10.1145/3555625
- Cotter, K. (2022). Practical knowledge of algorithms: The case of BreadTube. New Media & Society, 1–20. https://doi.org/10.1177/14614448221081802
- Tripodi, F. (2018). Searching for alternative facts: Analyzing Scriptural Inference in Conservative News Practices. Data & Society. https://datasociety.net/output/searching-for-alternative-facts/
- Mollick, E. (2023). One Useful Thing. Now Is the Time for Grimoires. https://www.oneusefulthing.org/p/now-is-the-time-for-grimoires
- Zamfirescu-Pereira, J. D., Wong, R. Y., Hartmann, B., & Yang, Q. (2023). Why Johnny Can’t Prompt: How Non-AI Experts Try (and Fail) to Design LLM Prompts. Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems. https://doi.org/10.1145/3544548.3581388
- McMahon, C., Johnson, I., & Hecht, B. ( 2017 ). The Substantial Interdependence of Wikipedia and Google: A Case Study on the Relationship Between Peer Production Communities and Information Technologies . ICWSM , 11 ( 1 ), 142–151 . https://doi.org/10.1609/icwsm.v11i1.14883
- Zade, H., Wack, M., Zhang, Y., Starbird, K., Calo, R., Young, J., & West, J. D. ( 2022 ). Auditing Google’s Search Headlines as a Potential Gateway to Misleading Content . Journal of Online Trust and Safety , 1 ( 4 ). https://doi.org/10.54501/jots.v1i4.72
- Williams, E. M., & Carley, K. M. (2023). Search engine manipulation to spread pro-Kremlin propaganda. Harvard Kennedy School Misinformation Review, 4(1). https://doi.org/10.37016/mr-2020-112