Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Learn as you search (and browse) using generative AI

blog.google

Post reply on HN
Print view View original

191–200 of 203 posts

Re: Learn as you search (and browse) using generative AI

#191

Post by machdiamonds »

ignite wrote: post #186
machdiamonds wrote: post #178

Earlier quoted context omitted.

I typically run lengthy YouTube video transcripts through Anthropic Claude to get a summary. It's free and offers a 100k context window. After summarizing, you can engage in a conversation with Claude. You can ask specific questions about the covered topics or prompt it with queries such as "What were the standout points?". It's quite adept at extracting key highlights.

Looks like that is still invitation only.

Really? I was under the impression that it is available to everyone. Are you attempting to access the API? There might be a waitlist for that. But, it should be readily available for use here: https://www.anthropic.com/index/claude-2

Re: Learn as you search (and browse) using generative AI

#192

Post by marricks »

brookst wrote: post #179
marricks wrote: post #155

Earlier quoted context omitted.

Ok, but people won't use it just for recipes. If it's successful and common place google will monetize it. The beginning could just be "see this related product or service." Which won't mess with the content. Perhaps that's fine. I'd argue though that any summarizing tool inherently has a bias. It must choose to ignore certain details and make decisions about what to highlight. As we understand LLM's more and the stu…

Too tinfoil for me. Bicycles are bad because someone will try to use one like a helicopter. And humans already summarize and (mis)characterize. I just don’t see blaming a tool that is used by people as they see fit.

I do if that tool is going to be more centralized then what it replaces.

Seems like a pretty facile argument. Bicycles and helicopters don't need to be trained on ridiculously largest datasets on ridiculously large super computers.

Re: Learn as you search (and browse) using generative AI

#193

Post by stubish »

jameshart wrote: post #172
tarvaina wrote: post #163

Earlier quoted context omitted.

For questions such as the difference between typhoon and hurricane, ChatGPT and Claude give accurate answers in just a couple of seconds. It’s such a different experience compared to the search engines.

But how do you know if they’re accurate?

You don't. Just like before generative AI responses. If accuracy is required, you have to check the citations and make a judgement call. And no, people generally won't do that just like they didn't do that before.

Re: Learn as you search (and browse) using generative AI

#194

Post by andsoitis »

carlosjobim wrote: post #184
andsoitis wrote: post #64

Earlier quoted context omitted.

What’s a good example that I can Google for but won’t find results (censored) or where only one side is presented? I will Google it and report back…

I don't think anybody is interested in giving you an example, because they expect that it will get them banned from HN, or at least have their comment flagged and removed by activists. Google does censor content, and they are kind enough to inform the owner of a website by e-mail when they have censored the content from their search results, so it is a fact that is just undeniable.

What a nice escape hatch. I can't give you and example because I will be banned if I do, ergo there are cases that will get me banned. Not QED.

Re: Learn as you search (and browse) using generative AI

#195

Post by andsoitis »

everdrive wrote: post #173
andsoitis wrote: post #145

Earlier quoted context omitted.

Do you want it sorted in some way? If so, why might you go to page 99999999999999 and pick the last item?

I’m not searching for some huge wildcard — I’m searching for some video content which must be generated by somebody and uploaded. There are not billions of results for “cool cover of a song I like” — there are hundreds of results at most. I want those hundreds of results.

Maybe you can just use a different product (search site) to get better results.

Re: Learn as you search (and browse) using generative AI

#196

Post by tomjen3 »

speak_plainly wrote: post #109

Reading is not a binary process and you can scale your engagement with any written work. If you use a variety of techniques, there is probably no need for a machine summary. For example, start by skimming and scanning an article and you may be able to pull out all the salient points and satisfy your reading goal or choose to go deeper with your analysis depending on time and interest and what you are trying to accomp…

It takes much more time to even skim an article than to run it through a summerizer. Not to mention that it’s generally hard to skim a YouTube video.

Re: Learn as you search (and browse) using generative AI

#197

Post by asdff »

faeriechangling wrote: post #104
cowsup wrote: post #6

At the risk of sounding like an old man — heck, I guess I am — is this innovation ethical for humanity's future? A one-click "tl;dr" button for the web? There are studies saying overall human attention span is shrinking. Anecdotally, I know people in their 20s who use TikTok daily, and they're no longer able to watch a movie without losing focus. Squeezing all aspects of our digital life, including learning new skill…

If we're going off anecdotes, we've also seen an explosion in popularity of the 2-3 hour podcast and season long Netflix binges. Doesn't this generation have an absurdly long attention span compared to the generation of the 30 minute TV show with three 2-3 minute commercial breaks?

There's things that are different though. People listening podcasts are sometimes multitasking so that's divided focus right there. In terms of modern TV and movies, shots are actually shorter these days. To its worst extent you can end up with these sorts of scenes:

https://www.youtube.com/watch?v=by4UZ-79MK4

Re: Learn as you search (and browse) using generative AI

#198

Post by The_Colonel »

GenericPoster wrote: post #121
The_Colonel wrote: post #75

Earlier quoted context omitted.

> some evil magic It can be. Summarizing process can be tweaked very slightly to give more space to certain POVs, make them sound more reliable than others via word choice etc. The subtlety can make it undetectable, yet it can have strong effects when applied on mass scale. I don't know how this will look like, but it is a very powerful technology to sway the public opinion one way or another.

Can you give some examples and point out the subtleties?

You can already try this with GPT, e.g. "Can you summarize XYZ while giving more attention to its negative impacts on the freedom of individual", contrast with "... to its benefits on the collective wellbeing".

Both will be factually correct, but they will each manipulate the opinion of the non-expert reader by giving more attention to specific facts.

GPT seems to be very strong in being able to present information with some specific accent, e.g. there are many examples of "write this in the style of [person, literary style, political slant]".

Re: Learn as you search (and browse) using generative AI

#199

Post by RugnirViking »

machdiamonds wrote: post #191
ignite wrote: post #186

Earlier quoted context omitted.

Looks like that is still invitation only.

Really? I was under the impression that it is available to everyone. Are you attempting to access the API? There might be a waitlist for that. But, it should be readily available for use here: https://www.anthropic.com/index/claude-2

"Unfortunately, Claude.ai is only available in the US and UK. We're working hard to expand to other regions soon."

Re: Learn as you search (and browse) using generative AI

#200

Post by indymike »

px43 wrote: post #89
indymike wrote: post #44

This is where you start to sympathize with the news people in Canada who want google to pay for links. If google just summarizes at scale, interdicts the traffic and revenue, where's the incentive to create content? This is a failure mode for the information economy.

If your only incentive for creating content is harvesting attention and eyeballs, then the internet is better off without your content.

I'm not sure that is the case: TV and radio use that model and are/were valuable.
Post reply on HN