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

Claude can now search the web

anthropic.com

Post reply on HN
Print view View original

571–580 of 758 posts

Re: Claude can now search the web

#571

Post by ryukoposting »

tcdent wrote: post #102

Searching the web is a great feature in theory, but every implementation I've used so far looks at the top X hits and then interprets it to be the correct answer. When you're talking to an LLM about popular topics or common errors, the top results are often just blogspam or unresolved forum posts, so the you never get an answer to your problem. More of an indicator that web search is more unusable than ever, but inte…

I reiterate: https://news.ycombinator.com/context?id=42012631

RAG was dead on arrival because it uses the same piss-poor results a human would, wrapped in more obfuscation and unwanted tangents.

My question is why the degradation of search wouldn't affect LLMs. These chatbot god-oracle businesses are already unprofitable because of their massive energy footprint, now you expect them to build their own search engine in-house to try to circumvent SEO spam? And you expect SEO spam to not catch up with whatever tricks they use? Come on, people.

Re: Claude can now search the web

#573

Post by bobxmax »

aprilthird2021 wrote: post #474
bobxmax wrote: post #293

Earlier quoted context omitted.

It's quite odd that people think of hallucinations as a dealbreaker for LLMs. Have they ever even met a human being?

The problem is that human beings are far more likely to know what they don't know. And we build a lot of our trusting work environments around that feature. An LLM cannot know what it doesn't know by definition.

I don't believe that's true at all. LLMs, especially reasoning models, tend to be quite good at calling out gaps in their knowledge and understanding.

LLMs also don't have the ego, arrogance and biases of humans.

Re: Claude can now search the web

#574

Post by nicce »

pests wrote: post #360
nicce wrote: post #65

Earlier quoted context omitted.

But how did you find those sites that had the robot.txt to begin with? LLM must somehow find the existence of those pages and store that information before they can crawl them further or mark as acceptable source.

I am a human so I can visit other sites with links or from word of mouth or business cards or literally anywhere? LLM finds out about it from me, when I ask it to go to the link. You don’t accuse browsers of “somehow find[ing] the existence of those pages”. How does a browser know what page to visit? The user tells it to. If I prompt an LLM “go to example.net and summarize the page” how is that any different from me…

That is certainly true. But that is not how these work 99% of the time. This post was originated by "search".

Re: Claude can now search the web

#575

Post by Sammi »

sshine wrote: post #466
labrador wrote: post #440

Earlier quoted context omitted.

Thank you for saying this. I recently heard that story "if you let one Nazi in your bar, pretty soon you have a Nazi bar"

Also, if you call everyone a nazi, all you have is nazi bars. I was called a nazi last week for driving a Tesla, and I have Jewish ancestry. The word hardly makes any sense.

In this case you aren't being called a nazi because of your ancestry. You're being called a nazi for supporting the car brand of a nazi. It does make sense.

A lot of people who definitely were not intending to be nazies are driving swasticars, because they didn't know about how nazi the car company owner was. But here we are. You definitely know now. What you do now matters.

Re: Claude can now search the web

#576

Post by wvh »

Almondsetat wrote: post #111
tcdent wrote: post #102

Searching the web is a great feature in theory, but every implementation I've used so far looks at the top X hits and then interprets it to be the correct answer. When you're talking to an LLM about popular topics or common errors, the top results are often just blogspam or unresolved forum posts, so the you never get an answer to your problem. More of an indicator that web search is more unusable than ever, but inte…

>looks at the top X hits and then interprets it to be the correct answer. LLMs are truly reaching human-like behavior then

Be careful what you call AI, you might just get what you wish for...

Re: Claude can now search the web

#577

Post by infecto »

wongarsu wrote: post #291
johntb86 wrote: post #215

Earlier quoted context omitted.

I've found that OpenAI's Deep Research seems to be much better at this, including finding an obscure StackOverflow post that solved a problem I had, or finding travel wiki sites that actually answered questions I had around traveling around Poland. However it finds its pages, they're much better than just the top N Google results.

Grok's DeepSearch and DeeperSearch are also pretty good, and you can look at their stream of thought to see how it reaches its results. Not sure how OpenAIs version works, but grok's approach is to do multiple rounds of searches, each round more specific and informed by previous results

Grok is still lightyears behind OpenAI when it comes to deep research capabilities. While its model might hold up reasonably well against something like o1, the research functionality feels rudimentary, almost elementary compared to what OpenAI offers. It might serve as a decent middle ground for basic search tasks.

Re: Claude can now search the web

#578

Post by pbronez »

dmazin wrote: post #546
HankWozHere wrote: post #124

Earlier quoted context omitted.

Kagi Assistant allows you to do search with LLM queries. So far I feel it bears reliable results. For instance - I tried couple of queries for product suggestions and came back with some good results. Whilst it’s a premium service , I find the offering to be of good value.

Has anyone compared Perplexity with Kagi Assistant? I am always looking for Perplexity alternatives. I already pay for Kagi and would be happy to upgrade to the ultimate plan if it truly can replace Perplexity.

I got a free year of Perplexity thanks to owning an R1. I already had a Kagi subscription, but decided to give Perplexity a try.

I found Perplexity was slower and delivered lower quality results relative to Kagi. After a week of experimenting, I forgot about Perplexity until they charged my $200 to renew my free year. I promptly cancelled the heck out of it and secured a refund.

Re: Claude can now search the web

#579

Post by infecto »

Sammi wrote: post #575
sshine wrote: post #466

Earlier quoted context omitted.

Also, if you call everyone a nazi, all you have is nazi bars. I was called a nazi last week for driving a Tesla, and I have Jewish ancestry. The word hardly makes any sense.

In this case you aren't being called a nazi because of your ancestry. You're being called a nazi for supporting the car brand of a nazi. It does make sense. A lot of people who definitely were not intending to be nazies are driving swasticars, because they didn't know about how nazi the car company owner was. But here we are. You definitely know now. What you do now matters.

Nope, still doesn’t make sense. It might give you a sense of satisfaction, but targeting the owners of those vehicles is immature. You have no idea what their financial situation is or what else they might be dealing with. Just because something feels good doesn’t mean it’s right.

Re: Claude can now search the web

#580

Post by vidarh »

_heimdall wrote: post #565
ldoughty wrote: post #423

Earlier quoted context omitted.

The idea behind UBI is that people do jobs that they want to do...

How do you make sure that enough people want to do the necessary jobs? And why do you need money at all in that scenario, at least for the basic items the UBI intends to make affordable to all? Why not just make them free and available to everyone?

You pay for them, on top of UBI.

No UBI proposal I'm aware of proposes UBI replaces salaries or is high enough to satisfy everyone. The "B" is for basic. Most people are not satisfied with earning a basic salary.

Post reply on HN