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

611–620 of 758 posts

Re: Claude can now search the web

#611

Post by Workaccount2 »

user_7832 wrote: post #479
Workaccount2 wrote: post #252

Earlier quoted context omitted.

Claude 3.7 with thinking is a big step up. If you haven't used it I'd suggest giving it a try. I have used grok a bit and it did what I needed it too, so I can't really compare. But 3.7 thinking is crazy strong for coding.

I wonder if you use it exclusively for coding, because for general purpose explanation tasks, 3.7 seems absolutely terrible unfortunately. Back when it was 3.5 you could actually talk and learn things and it felt humane, but now it sounds like a McKinsey-corpo in a suit who sounds all fancy but is only right half the time. I’ve switched back (rather regretfully) to chatgpt, and holy hell is its personality much bette…

I use it just for coding.

For general inference I use 4.5

I think OpenAI (and likely others) are on the right to track to acknowledge that different model tunings are best for different uses, and they interned to add a discriminator that can direct prompts to the best tuned model/change model tuning in real time.

Re: Claude can now search the web

#612

Post by knowaveragejoe »

MattSayar wrote: post #194
gcanyon wrote: post #177

Funny, I literally just two days ago asked Claude to provide an outline of the functionality of a product, giving it the web site. It of course refused. So I downloaded the text of the site and passed that in, and got mediocre results. The results based on giving the source URL directly were better. Still a bit generic and high-level and vague, as LLMs tend to be, but better than the text-download version a couple da…

I had tried using monolith [0] to feed webpages into Claude but all the html was too much token context. I ended up Print > Save as PDF-ing somewhat often and that worked pretty well. But just giving a URL is ideal. [0] https://github.com/Y2Z/monolith

Using the fetch MCP tool(which is very basic) works fairly well for me, though various sites will block it from time to time.

https://github.com/modelcontextprotocol/servers/tree/main/sr...

Re: Claude can now search the web

#613

Post by klabb3 »

sumedh wrote: post #509
klabb3 wrote: post #368

Earlier quoted context omitted.

Fake it til you make it vibes. I understand why he would do it, what I find strange is that it inspires confidence in the market.

What are they faking?

Sama as the spokesperson regularly makes grandeur statements, often very vague, can’t show it because ”safety”, trade secrets etc. I think it’s widespread culturally, especially in earlier VC-centric times when investor fomo and mystique is name of the game. But nowadays even large publicly listed companies like Tesla pull this off and even sell consumer products that don’t exist yet. Do you want specific examples of sama statements that I think are horseshit specifically designed to generate buzz? It’s not hard to find.

Re: Claude can now search the web

#614

Post by vbezhenar »

UnreachableCode wrote: post #510
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…

> web search is more unusable than ever I’m curious why I’m seeing a lot of people thinking this lately. Google definitely made the algorithm worse for customers and better for ads, but I’m almost always able to find what I’m looking for in the working day still. What are other people’s experiences?

My experience is that Google works perfectly for me and I almost never have any issues with it, despite all the doomsaying.

Re: Claude can now search the web

#615

Post by Workaccount2 »

int_19h wrote: post #326
j_timberlake wrote: post #297

Earlier quoted context omitted.

I expect that AI good enough to automate jobs will also be dangerously good at criminal activities. Governments will want to ban them, but there's just too much $$$ to be made from replacing employees, so things will get complicated fast.

They are already good at criminal activities such as phishing. That bar is rather low, especially once you scale up (hitting 100 people and successfully scamming 1 is still great ROI with cheap small models). But I don't see what governments can really do about it. I mean, sure, they can ban the models, but enforcing such a ban is another matter - the models are already out there, it's just a large file, easy to torr…

It's just going to turn into an arms race of AI trying to stop AI.

Re: Claude can now search the web

#616

Post by unshavedyak »

shortrounddev2 wrote: post #419
herdcall wrote: post #362

It badly hallucinated in my test. I asked it "Rust crate to access Postgres with Arrow support" and it made up an arrow-postgres crate. It even gave sample Rust code using this fictional crate! Below is its response (code example omitted): I can recommend a Rust crate for accessing PostgreSQL with Arrow support. The primary crate you'll want to use is arrow-postgres, which combines the PostgreSQL connectivity of the…

> I asked it "Rust crate to access Postgres with Arrow support" Is that how you actually use llms? Like a Google search box?

That's primarily how i do, though it depends on the search ofc. I use Kagi, though.

I've not yet found much value in the LLM itself. Facts/math/etc are too likely incorrect, i need them to make some attempt at hydrating real information into the response. And linking sources.

Re: Claude can now search the web

#617

Post by hooverd »

pixl97 wrote: post #200
jrflowers wrote: post #103

Earlier quoted context omitted.

If this feature isn’t already part of the Claude API it likely will be at some point, in which case many Claude requests will be automated with no way to distinguish between user-driven or otherwise.

Simply put, at the end of the day you lose, AI blocking will not work. I mean, currently the AI request comes from the datacenter running the AI, but eventually one of two things will happen. AI models will get small/fast enough to run on user hardware and use the users resources: End result? You lose. The user will set their own headers and sites will play the impossible game of identifying AI. AI sites will figure…

Ideally bad behavior by AI companies should trigger crushing fines and jail time for executives.

Re: Claude can now search the web

#618

Post by unshavedyak »

whatevertrevor wrote: post #529
globular-toast wrote: post #502

Earlier quoted context omitted.

It's funny because many people type full sentence questions into search engines too. It's usually a sign of being older and/or not very experienced with computers. One thing about geeks like me is we will always figure out what the bare minimum is (at least for work, I hope everyone has at least a few things they enjoy and don't try to optimise).

It's not about being young or old, search engines have moved away from pure keyword searches and often typing your actual query gives better results than searching for keywords, especially with Google.

Wonder if that's why so many people hate its results lol. It shifted keyword searching to full sentence searching, but many of us didn't follow in the shift.

Re: Claude can now search the web

#619

Post by dostick »

Anyone noticed that if you enable the “browse internet” in ChatGPT, it becomes very dumb? It abandons all its intelligence and produces mostly incorrect results. Like it’s being passive-aggressive, “Oh, you don’t like me as I am and want to augment me with search, let me show you how it is if my brain was only search!”
Post reply on HN