Last update: Saturday 7/26/25
A. Browsing | B. Chatbots | C. Assistants/Associates | D. Beyond | E. Takeaways
A. Browsing for information before genAI
The reader should note that this discussion will refer to apps like Chrome, Safari, and Firefox as "browsing tools" because we suggest that the term "browser" was always a misnomer: it was, at best, a shorthand for "browsing tool"; but at worst, it was a misleading term that has recently caused would-be competitors, like ChatGPT and Perplexity, to challenge Chrome's dominance, instead of recognizing that the app to beat in the age of genAI is actually Gemini.
- People have always been "browsers" and they have always used tools, like card catalogs in libraries, to help them browse.
- Before the Internet, one browsed before deciding which materials to read. For example, one might skim through various magazines at a newsstand before deciding which ones to buy.
- Then along came the Internet, a vast repository that promised to place all human knowledge at one’s fingertips. The Internet’s information was organized into pages that had unique identifiers called “hyperlinks” or just “links”. Pages could refer to other pages via these links. Creators placed their pages on Internet servers. Users displayed these pages via client apps called “browsers”, e.g., Mosaic, Netscape, and Explorer .
- The Internet expanded so rapidly that it quickly contained far too many pages for anyone to skim before reading.
- But when a user supplied Google‘s search engine with a few words that named a subject’s key features, the search engine returned a long list of links to pages on the Internet that discussed those features.
- Google always placed the “best” pages at the top of its lists. So most user's skimmed the pages linked to the top of the list before deciding which ones to read in order to obtain the information they were looking for.
- Actually, Google developed two browsing tools: its Chrome “browser” and its search engine for finding the best pages. Other tech companies also invented new “browsers” — e.g., Safari, Firefox, and Edge — and other search engines — e.g., Bing, Brave, and DuckDuckGo. But Google’s tools quickly became and remained the most widely used.
B. Chatbots
The notion of artificial intelligence pre-dates the 20th century, but it was not formalized into an academic discipline into the late 1950s by researchers at M.I.T., Carnegie-Mellon, and Stanford Universities. AI developed in fits and starts with a long period of limited progress called the “AI winter”. But progress accelerated in the early 2000s and moved into high gear after the publication of google’s famous paper, "Attention Is All You Need"
- When OpenAI released its chatbot called ChatGPT running on its GPT-3.5 language model at the end of 2023, Google issued a famous “Code Red” — because it realized that its lucrative search engine business was facing a direct threat. Users could now ask questions and receive g answers, rather than receive links to pages that might contain the information that might answer users’ questions.
- As Microsoft’s CTO, Kevin Scott, proclaimed Chatbots are the “great levelers” that enable ordinary users — users who do not know python or Java or JavaScript or whatever other nerdy computer language — to directly interact with the most powerful technology the human race has ever developed: generative AI.
- ChatGPT's viral popularity conditioned users to expect that chatbots would become the user interface for all subsequent AI mechanisms, including connectors, agents, and everything else that genAI developers might create.
- Google quickly produced its own chatbot called Gemini. At this time, Gemini is the only Big Tech chatbot that has full access to the most powerful search engine that can find the best sources of information on the Internet, Google’s search engine; and Gemini is also the only Big Tech chatbot that can follow a link to every page on the Internet that is not restricted by a paywall or some other prohibition (because of legal constraints).
- Gemini is the app to beat by would be competitors of Google, not Chrome because Gemini’s dominance of these two functions enables it to become world's first genAI browsing tool. The days of Internet browsers being users’ primary interface to the world’s vast repositories of information are over.
- To be sure, Google’s technical advantages may not be decisive for users in densely populated urban areas when users are looking for concise summaries of features or events that have been widely reported in high profile media. But users in suburban or rural areas who are looking for summaries of local phenomena may quickly come to appreciate the encyclopedic scope of Google’s search engine because if it’s out there somewhere, Google’s search can find it.
C. Research Assistant vs Research Associate
For these two roles, Gemini’s access to Google’s search and its ability to follow links to unrestricted pages together with its capacity to remember previous interactions will provide it with a decisive competitive advantage over the chatbots of other Big Tech genAI providers. At this time only Gemini and ChatGPT can remember previous interactions with their users.
Throughput the following discussion ChatGPT is assumed to be running on GPT-4o; Claude is running on Sonnet; and Gemini is running on Plus
- A Research Assistant must be able to fact check a user’s hypotheses against the findings of the top ten, perhaps the top twenty best sources of information on the Internet to determine which ones confirm or refute the user’s hypotheses.
Indeed, fact checking might be an iterative process wherein the user begins by directing Gemini to summarize relevant findings published by specific sources, then summarize all other sources.
Without Gemini’s memory this would be an inefficient process because it would require the user to load Gemini with the fully detailed original prompt during each iteration, rather than merely change the sources to be checked and summarized.
Whereas the activities of a Research Assistant are driven by a user’s hypotheses about a subject, a Research Associate would merely be given an open question, e.g., what factors seem to influence a subject’s behavior and under which circumstances?
- In other words an associate would be given sufficient autonomy to infer its own hypotheses or to modify its user's hypotheses, then act as its own assistant to fact check its hypotheses before presenting its report to its user.
- The user would then engage the associate in a conversation about the report wherein the user might suggest that the associate modify its hypotheses and produce a revised version of its report.
Inferring hypotheses involves chain of thought capacity, a capacity possessed by all three chatbots. In a recent note on this blog, the editor reported that his personal experience suggested that Claude had the strongest reasoning capacity, ChatGPT mid-range, and Gemini the weakest.
Some users might therefore find it productive to ask ChatGPT to provide a critique of the logic of Gemini’s report and to validate the sources of the report (where possible). The editor recently asked ChatGPT to summarize a seven page report written by Claude and to identify any limitations. The Conclusion of ChatGPT‘s summary provided an insightful critique of Claude's reasoning.
(Note: Claude’s lack of memory of previous interactions would limit its capacity to participate in an iterative discussion of Gemini's report. Indeed, Claude has no recollection of the seven page report that it wrote last week ... 😎)
D. Beyond the Internet
The broader user community, including supremely talented professionals in a wide range of other fields, will not want to have direct contact with anything like Anthropic’s MCP connectors or Google’s Agent2Agent A2A protocols. Yes, they will want to use these powerful new AI infrastructures as soon as they get a better understanding of how much more information and real world impact these infrastructures will enable them to obtain. But the broader user community will want to interact with these connectors and protocols via their favorite chatbots using English, French, Chinese, or whatever is their preferred human language.
E. Key Takeaways
- Google should encourage its Chrome users to redirect their requests for information to its Gemini chatbot while Gemini enjoys its current technical advantages. The Chrome browser will still be useful for displaying the Internet pages that are the sources of Gemini’s responses.
- Google should anticipate that its politically well connected competitors in the Big Tech community will use their connections to policy makers at the highest levels in the U.S. government to pressure Google to share its competitive advantages with other U.S. Big Tech firms in order to make the entire American Big Tech community more competitive against all foreign providers of genAI services.
____________________________________
Links to related notes on this blog:
- "The Anthropic tortoise and the Google hare" ... 7/11/25
No comments:
Post a Comment
Your comments will be greatly appreciated ... Or just click the "Like" button above the comments section if you enjoyed this blog note.