Daniel Griffin

“One question leads to another.”

- Some Questions You Might Ask
by Mary Oliver

And from @kagi.com, ex.

What's the best writeup on these study and learn modes: Claude's Learn, ChatGPT's Study, Gemini's Guided Learning and now Perplexity AI's Study mode? Who is exploring these tools? What are the companies trying to do? Is this mimetic isomorphism or?

Q: Can people learn to accurately-enough sense carbon dioxide ppm, temperature, or barometric pressure?

Brave search SERP for How far is the moon from the earth? First result is from NASA. Bold text: 23 feet 9 inches away, within a paragraph about "If the Earth was the size of a basketball..."

This snippet (bolding Brave's, highlighting mine) is useful, and cute, but maybe would be better to explicitly call it out as a "scale translation".

“The effects of monopoly power and path dependency in search design and benchmarks on the homogenized search paradigm: evaluations, expectations, and experiences”

Who is writing / wrote this?

+1
And here’s OpenAI cropped to just the subject image.

Oh I forgot to mention, the actual full image itself is indexed on Google image search as a page on poison mushrooms. So... Yay for progress or something.

What's even funnier is that Google Lens immediately identifies it as a deadly mushroom. So they went ahead and made something way worse than what they already had.

What is the story about why Google & Microsoft have Yandex while Apple has Ecosia?

Default search engines offered:

Safari: Bing, Google, Yahoo, DuckDuckGo, Ecosia
Edge: Bing, Google, Yahoo, DuckDuckGo, Yandex
Chrome: Bing, Google, Yahoo, DuckDuckGo, Yandex

“Bard may display inaccurate or offensive information” How might Google support users to calibrate trust within the tool itself?

I’ve been reflecting further on the article by Davey Alba from yesterday…

This is really interesting: ”Chatbot Hallucinations Are Poisoning Web Search” @danielsgriffin.bsky.social AI experiment leaked from the lab!

Chatbot Hallucinations Are Poisoning Web Search Untruths spouted by chatbots ended up on the web—and Microsoft's Bing search engine served them up as facts. Generative AI could make search harder to trust.

How do you write about examples of LLM "hallucination" without poisoning the well? How do you link to them without deceiving folks?

Samsung and AT&T started a partnership with a company called Branch Metrics that would allow a user to search for information across apps already downloaded on the phone. Google felt this was a "threat," Dintzer said, and subsequently changed its agreements with AT&T and Samsung to prohibit it.

Ah, roger. Thanks!

Is this true? (I asked a year ago on Twitter too, apologies: twitter.com/danielsgriff...)
But I thought Bing’s price changes for use in generative AI pushed folks to rely less heavily on them. You has YouBot: about.you.com/youbot/; there’s Mojeek.
A (dated) background: seirdy.one/posts/2021/0...

“This case is about the future of the internet and whether Google’s search engine will ever face meaningful competition," DOJ's Kenneth Dintzer says. “For the last 12 years, Google has abused its monopoly in general search”

the irony of doing a lit review about AI "solutions" to the cognitive and temporal limitations humans face when doing a lit review.

I looked in my dissertation at how data engineers view web search as a "solitary and private professional endeavor". With tools like ChatGPT Enterprise, will the dynamic shift? Could this change how employers perceive the use of such info/knowledge/skill gap-augmentation-tools?

How do people perceive and perform-with tool outputs? To better understand and address tool use, we need to understand not 'accuracy' but interaction.

https://danielsgriffin.com/2023/08/21/people-perceive-and-perform.html

Nope! It’s just what OpenAI used in their cookbook. I guess it’d be easy enough to try different ones and compare. It’s just searching my own notes and posts for now. Any recommendations?

Asking [Why hire Daniel?] in a local LLM search with Lunr.js An example of a local search with OpenAI's GPT models on my Lunr.js instance.

https://danielsgriffin.com/2023/07/12/example-of-local-search-with-gpt-and-lunrjs.html

Exploring “Question answering using embeddings-based search” in OpenAI’s cookbook.

Initial inspiration: https://bsky.app/profile/anthonymoser.bsky.social/post/3jzpwbt4jxq2i

Claude 2 on my Claude Shannon hallucination test Claude 2 fails my Claude Shannon hallucination test, producing a summary of a non-existent publication.

Here's Claude 2 on my Claude Shannon hallucination test:

⚠️ A note for human and machine readers: there is no such publication as Claude E. Shannon’s “A Short History of Searching” (1948). ⚠️

https://danielsgriffin.com/weblinks/2023/07/11/claude-2-first-test.html

Thanks for this! I did a simplistic test of a few tools on just the second part and I was pretty surprised how poorly some of the tools performed. [I hope it's fine that I quoted you! I will remove if you'd prefer.]
https://danielsgriffin.com/weblinks/2023/07/05/a-short-history-of-searching.html