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

411–420 of 758 posts

Re: Claude can now search the web

#411

Post by rank0 »

dataflow wrote: post #389
rank0 wrote: post #385

Earlier quoted context omitted.

> Could you at least explain the logic that you believe implies this would occur with such certainty? Because we haven’t actually created anything. Supply is the same, demand is WAY up.

That doesn't follow. It's a reason to believe prices will increase, not that prices will increase roughly in line with the income increase . This distinction is not a minor detail, it's pretty crucial. If you give people $3k and the prices go up by $2k... that's a very different scenario from one where the prices go up by $3k.

It should all even out in the long run.

As long as we’re in a deficit, spending for this program would directly increase the money supply. Of course there are other factors like velocity of money and elasticity of good/services but at the end of the day we’re increasing the amount of money (aka cash + credit) with no change to supply AND we’re going into debt to do it.

Re: Claude can now search the web

#412

Post by prawn »

rank0 wrote: post #375
prawn wrote: post #343

Earlier quoted context omitted.

One suggested weakness of UBI is a lack of purpose. I wonder if the "solution" is somewhat as you implied: jobs without a strict return on investment. You get your stipend, but you're keeping your block clean by sweeping and mulching. They're getting theirs in exchange for cranking out sourdough at cost for the neighbourhood. Someone else gardens for elderly residents.

How do we determine who gets what job?

Let people choose if they want to do something, but have a concerted effort to encourage/suggest things that might give them purpose and build a community. Leave them to decide their hours and effort. Maybe someone wants to clean the gutters for their entire block at 6am and then go tinker in the shed for half the day. I'm sure that sounds really lazy, but this concept is working up from a default UBI that is pay-for-no-job.

I can imagine loads of tasks or jobs that would be quite pleasant if it weren't for stressing over efficiency or business admin.

Re: Claude can now search the web

#414

Post by prawn »

sanex wrote: post #405
jmrm wrote: post #382

Earlier quoted context omitted.

Not UBI per se, but this exists in rural parts of Southern Spain in some way, and is called Rural Employment Plan (PER in its Spanish initials). The give simple jobs, like cleaning or painting, to people on the lower bottom of earnings. Most people in that plan are people with low formation, like those who left school in their mid teens.

Sounds like a minimum wage

Maybe a bit like the CCC. https://en.wikipedia.org/wiki/Civilian_Conservation_Corps

Re: Claude can now search the web

#415

Post by MuffinFlavored »

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…

> the top results are often just blogspam

top results are blogspam but the LLM isn't? /s

Re: Claude can now search the web

#416

Post by shortrounddev2 »

ggm wrote: post #407

So in many respects, search the place it used to construct the model? Isn't that functionally bias-reinforcing? "Look what I synthesise is correct and true because when I use the same top 10 priming responses which informed my decision I find these INDEPENDENT RESULTS which confirm what I modelled" type reasoning. None of us have a problem with an LLM which returns 2+2 = 4 and shows you 10 sites which confirm. What w…

We will very quickly enter a Kepler effect of information on the internet. All text on the internet will become AI slop being parsed by AI. Real information and human beings will be drowned out by the garbage. The internet will cease to be useful and we will retreat to corners of the web or to walled gardens. I'm seeing more and more online communities these days enforce invite only because there's just too much AI slop everywhere now.

Re: Claude can now search the web

#417

Post by rank0 »

prawn wrote: post #412
rank0 wrote: post #375

Earlier quoted context omitted.

How do we determine who gets what job?

Let people choose if they want to do something, but have a concerted effort to encourage/suggest things that might give them purpose and build a community. Leave them to decide their hours and effort. Maybe someone wants to clean the gutters for their entire block at 6am and then go tinker in the shed for half the day. I'm sure that sounds really lazy, but this concept is working up from a default UBI that is pay-for…

Nobody is going to choose to be a ditch digger without a financial incentive. Most jobs worth doing are unpleasant or difficult. Thats why people pay for the labor!

I mean think about it…when was the last time you heard of charity gutter cleaning services? People would much rather enjoy their leisure time on hobbies or with family/friends.

Re: Claude can now search the web

#418

Post by manquer »

JimDabell wrote: post #231
NBJack wrote: post #10

I wonder if it will actually respect the robots.txt this time.

The LLM shouldn’t. robots.txt is intended to control recursive fetches. It is not intended to block any and all access. You can test this out using wget. Fetch a URL with wget. You will see that it only fetches that URL. Now pass it the --recursive flag. It will now fetch that URL, parse the links, fetch robots.txt, then fetch the permitted links. And so on. wget respects robots.txt. But it doesn’t even bother lookin…

While robots.txt is not there to directly prevent automated requests, it does prevent crawling which is needed for search indices.

Without recursive crawling, it will not possible for a engine to know what are valid urls[1]. They will otherwise either have to brute-force say HEAD calls for all/common string combinations and see if they return 404s or more realistically have to crawl the site to "discover" pages.

The issue of summarizing specific a URL on demand is a different problem[2] and not related to issue at hand of search tools doing crawling at scale and depriving all traffic

Robots.txt does absolutely apply to LLMs engines and search engines equally. All types of engines create indices of some nature (RAG, Inverted Index whatever) by crawling, sometimes LLM enginers have been very aggressive without respecting robots.txt limits, as many webmasters have reported over the last couple of years.

---

[1] Unless published in sitemap.xml of course.

[2] You need to have the unique URL to ask the llm to summarize in the first place, which means you likely visited the page already, while someone sharing a link with you and a tool automatically summarizing the page deprives the webmaster of impressions and thus ad revenue or sales.

This is common usage pattern in messaging apps from Slack to iMessages and been so for a decade or more, also in news aggregators to social media sites, and webmasters have managed to live with this one way or another already.

Re: Claude can now search the web

#419

Post by shortrounddev2 »

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?

Re: Claude can now search the web

#420

Post by notepad0x90 »

if everyone is using LLMs to solve problems, in a few years, won't LLMs run out of content to mine? In short, how can the general dumbing down of LLMs and degradation of content used to solve problems be avoided over the long term?

For questions about events and problems that arose after 2025, where would LLMs get information to solve those? and who would be asking those at a forum LLMs can access going forward?

Is the snake eating it's own tail?

Post reply on HN