Hacker News Reader

Comment by mpweiher | Hacker News Reader

This is not EU R&D budget. This is a small delta, extra funding specifically for attracting US scientists who are looking to move. Comparing this to the total expenditure by the US federal government seems...odd.

I asked ChatGPT:

"Combining both EU-level and national R&D expenditures, the total R&D spending across the EU in 2023 was approximately €505 billion." But that appears to be total, both government and industry.

Spending by national government was apparently around € 123 billion. In addition, the EU spent ~ € 13 billion a year. So a total of € 136 billion in government spending.

https://www.eureporter.co/economy/eurostat-economy/2024/08/0...

https://eufunds.me/what-is-the-budget-of-horizon-europe/

View on HN

For the love of everything, CHATGPT IS NOT A PRIMARY SOURCE. Always assume every fact it spits out is made up.

Good thing then that I didn't use it as a primary source. :-)

it's not a source at all, it's definitionally bullshit because the LLM has no concern for the truth. Never post LLM text. Anyone can get that for themselves. To do so is an insult to the comment section. I'm not here to read bot summaries. If you ask the bot something and then it leads you to a source and then you read that and then you put that in your own words, that's a comment worth reading.

what you did is like farting in a crowded space. STINKY AND RUDE

unless you know, it gives you verifyable sources to dig in deeper to verify - much like Google search (minus the ads on top) :)

But you wouldn't post your google search either.

Lmgtfy links were extremely common around 2010, so ppl totally did

(Which is besides the point even, as the comment referencing chatgpt also provided links to sources)

why wouldn’t I if asked to provide source(s) for my “claim(s)”?

I think that the US strongly outperforms the EU by that metric too,

https://ncses.nsf.gov/pubs/nsb20246/cross-national-compariso... ("Cross-National Comparisons of R&D Performance" (2024))

In the NSF's specific definition, the US greatly outspends the EU even normalizing by GDP—3.46% of GDP in the USA, against 2.16% in the EU-27.

(Notably, China also recently surpassed the EU, at 2.43%).

The US is undoubtedly ahead at the moment, but the point is that this moment is developing into a turning point where the US is reducing science funding while simultaneously being openly hostile to both scientists and very concept of science itself. If US scientists feel this not just a transitory bump but a genuine change in the political climate going forward, then Europe is going to look inviting, especially if they start offering incentives.

Hmm...Wikipidea says "$135.110B in R&D spending" for federal funds.

Now that was 2015, but the number is very similar the EU figure of €136 billion.

There will be be differences, but the point stands that comparing the tiny delta provided by this specific program to total spending is not serious.

Right; that's the federal government spending, while the other's combined R&D from all sources (aligning with 'mpweiher 's comment and their metric).

The nsf.gov page has a breakdown table, too, of government vs. industry spending.

That turns out not to be the case. The €136 billion was EU + national governments.

I'm not sure if you can separate this as easily. In Germany for example a lot of funding comes through Max Planck institutes (and Fraunhofer and Leibniz centres and I' forgetting one), not sure if those are counted as government, but they are basically on par with the US national labs (but less military contacts)

I salute you for being open about the chatgpt use! Do you really trust what it says? That quoted chatgpt-number has zero value to me.

The other numbers are valuable, since they come from actual sources.

What if the links came from the ChatGPT answer as well?

For me that does not change anything really. I assert the trustworthyness of the webpage as usual.

My problem is the statement from chatgpt. I have seen it invent enough bullshit that if it was a person I would have labeled them as untrustworthy a long time ago. Yes yes, it's also amazing and all that jazz, but I still don't know how to trust a 'Chatgpt told me this' - quote.

I do get it. However it's hardly any different or less trustworthy than a random person making a random claim identical to what ChatGPT would say.

Of course a 'Chatgpt told me this' disclaimer does indicate something, i.e. either that person has no clue about the topic and is unable to verify the answer at least to some extent on their own and is just blindly copy pasting something and/or believes that anything LLMs say is inherently credible on its own without extra verification.

Good grief.

It's exactly the opposite.

Then why even bother with the "I asked chatgpt"? Just cross reference the links and credit the original sources. It just adds verbosity and doubt to the statement.

It's well known for making the stuff up because this is how it works

The Guardian found a article attributed to them, generated (not "written") by chatgpt.

A silly lawyer got into trouble trying to use chatgpt-generated precedences in court.

Everything chatgpt prints out is made up, and that includes links.

Seriously, heed the warning the company itself prominently prints I app and in the webui.

Chatgpt may print out mostly true made-up sentences, but by definition, because oh how it works, it doesn't generate truth. It generates tokens that make up words.

Chatgpt is not a RAG, come on, it's 2025!

It can (and does actually) open and verify the links it provides so it's not as bad anymore, at least when it's using real existing articles/papers/etc. it find as sources inside its context.

They did.

> Do you really trust what it says?

"Really trust"? Nope. But I think it gives me a good ballpark estimate and ways to check if that estimate is about right or not.

Checking the answer is quicker and potentially less error-prone than compiling the answer.

Unless you verify the ballpark figure you shouldn't really use it in the conversation.

Chatgpt is a glorified autocomplete. Don't share it's output unverified like that was some sort of an oracle.

If you really HAVE TO resort to "AI", at least use Perplexity.

Ok, so did you verify the 500 billion? If so, then that's really the relevant part for me. But then I trust you, not Chatgpt.

Where did you take the 500 billion figure from?

It's from the chatgpt quote my comment was an comment to: "[...] in 2023 was approximately €505 billion"

Thanks.

> I asked ChatGPT:

Funny I asked it about your info and it says it has discrepancies. :-))

"I asked chatgpt" is an immediate downvote.

NOBODY CARES what chatgpt says, and EVERYONE has the ability to ask it themselves.

You aren't adding to the conversation, you're just taking up space.

> I asked ChatGPT

irony alert

Reply on news.ycombinator.com