To me, it misses to analyze and address a fundamental problem of why google “got so bad”.
I believe this is mainly a problem of the top 5%. There is reason why google presents their results like they do. It’s not like they’re a bunch of idiots not knowing their craft.
Yes, for those of use around the internet since 1996, the results got worse. But since then the entire population got online. They’re just fine with the top results. They’re good enough. Google’s machine knows and sees this in the data.
I am sure there is value in summarizing and querying the text. But this will increase the convenience not necessarily the quality of results.
Google isn't delivering information based on what's relevant to users, Google is funneling you into their business: to funnel you into someone else's business. And they're doing it not by being very useful, but by being the default you reach for on every product you buy. To set up your life to not use google is a big hassle, I can safely say this because I've done it.
This is not simply an issue of "your needs are simply more advanced than normies", this is an issue of "indexing the internet doesnt work, we can't win the cat and mouse game of SEO, so we will just ensure a userbase by controlling the environment around everyone." Sprinkle in a little bit of gaslighting and information control and you've got the current state of affairs with regard to internet search.
It's pretty easy to create an search engine that outperforms google in dealing with SEO by valuing different things than Google. But as long as you have a system with a strong monopoly in search directing a huge swath of web traffic, there will be a sort of evolutionary pressure in the direction of what we're seeing on Google.
It's not a function of the sort of search enginge Google is, but its market position, and the fact that there is a metric shit-ton of money to be made in successfully adapting to its algorithms.
Google have deliberately stopped trying to provide an index of the web. They now focus solely on revenue (advertising) maximization. A search will show paid sites, then popular sites, up to a few pages, and that's it. For example, search for "brioche buns" - you can't convince me that there are only 420 pages in the entire www that refer to "brioche buns" :(
It is just not true.
Use any of the competing search engines for even a day and you'll come running back to Google. Believe me, I tried it. I really wanted to switch to Bing or the Duck, but they just don't compare. If the search results of Google are not to your liking it is because search is hard. Also login, if you are a super user, Google will catch on to this as well.
For keyword based search (thinking index and crawlers e.g. 'motorized+camera+focus -autofocus') Google has objectively gotten worse since about 2016 (my personal theory is that it was when Giannandrea headed Search and started allowing ML to be used in the search team).
Search for the general population that is familiar with Natural Language queries i.e. 'what is that film where the guy keeps living the same day over and over?' That type of search is probably improved and is likely making Google a lot of money, which is why the keyword based search is degradating.
Also I think currently Search is putting too much enphasis on 'freshness' for technical queries. Most of the time I am looking for what is prior art on a particular topic, it is absolutely fine to show me a blog posted in 2004 by some geek in Slovakia.
The noise you hear online is that most technical people kind of rely on keyword and bolean operators to find what they are looking for and so are suffering the most from the business shift in Search.
Also not affilliated but definitely supportive... I wholeheartedly agree with other commenters, Kagi kicks ass in technical queries.
I haven't. I switched to ddg years ago and I haven't gone back since. I basically forgot that google search exists and that people still use it.
https://en.wikipedia.org/wiki/Doramad_Radioactive_Toothpaste
"""
There are several variants of GPT-3 model, each fine-tuned for specific tasks or industries. Here are a few examples:
- GPT-3 "davinci" is fine-tuned for creative and imaginative writing tasks, such as poetry, short stories, and song lyrics.
- GPT-3 "curie" is fine-tuned for conversational and dialog generation.
- GPT-3 "babbage" is fine-tuned for code generation and language understanding.
- GPT-3 "einstein" is fine-tuned for question answering tasks.
- GPT-3 "jules" is fine-tuned for summarization.
- GPT-3 "parliament" is fine-tuned for legal text generation.
- GPT-3 "bronte" is fine-tuned for creative writing tasks like storytelling and fiction-writing.
"""
I'd almost assume "parliament" it's a leak of some sort if any of the other examples (except for davinci) were accurate.
Ah, i guess that's what you meant by leak. I think some of the others are accurate - babbage in particular is very accurate!
Even if you only get fiction 20% of the time that's still absolute poison for a search engine that sounds completely authoritative. It's like having sand on your toothbrush.
The solution to abundance of bad info on the internet are two things. One, rudimentary search engine skills, i.e. filtering by source, keyword search and some basic literacy because all the good sources are still there and growing. Secondly, trusted sources curated by people that you have confidence in to produce high quality output.
The discourse increasingly reminds me of Alexa/"smart" home assistants which people thought would replace everything because the natural language interaction seems futuristic.
Generative models dont know what they dont know, they will always struggle on new topics and worse still they wont provide feedback to the user to adjust their query as they will simply hallucinate details
[0]: https://kagi.com [1]: https://labs.kagi.com/ai/contextai
