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

The danger of relying on OpenAI's Deep Research

economist.com

Post reply on HN
Print view View original

21–30 of 63 posts

Re: The danger of relying on OpenAI's Deep Research

#21

Post by cratermoon »

Workaccount2 wrote: post #12
azinman2 wrote: post #3

The world is going to become stupider because it’s easier. This will be most dramatic for the youth in both school learning as well as in the field. Unfortunately I don’t realistically see another path in any reasonable time unless the world takes a radical turn away from AI (extremely unlikely).

The models that exist right now would still be considered alpha LLMs. Even a half decent video game takes longer to make than the time since GPT3.5 launched. If they still are at this level in 2-3 years, then yeah, it will likely cause a bunch of bad downstream effects. But at least recognize that it is still a few years too early to judge.

Allow me to introduce The First Step Fallacy. https://thebullshitmachines.com/lesson-16-the-first-step-fal...

Re: The danger of relying on OpenAI's Deep Research

#22

Post by giaour »

seanmcdirmid wrote: post #10
azinman2 wrote: post #3

The world is going to become stupider because it’s easier. This will be most dramatic for the youth in both school learning as well as in the field. Unfortunately I don’t realistically see another path in any reasonable time unless the world takes a radical turn away from AI (extremely unlikely).

They probably said the same thing when the printing press came out. “How can you really know the book if you didn’t copy it down yourself during a reading?” But the printing press won out in the end.

People didn't read books by making copies of them. Producing new copies of manuscripts was a skilled occupation.

Re: The danger of relying on OpenAI's Deep Research

#23

Post by seanmcdirmid »

giaour wrote: post #22
seanmcdirmid wrote: post #10

Earlier quoted context omitted.

They probably said the same thing when the printing press came out. “How can you really know the book if you didn’t copy it down yourself during a reading?” But the printing press won out in the end.

People didn't read books by making copies of them. Producing new copies of manuscripts was a skilled occupation.

Nope, that’s not true. You had monks copying books of course (mostly bibles), and very rich people could afford very expensive books made by scribes, but students at universities were expected to make their own copy during a reading/lecture run by a reader/professor. The words are still used in some older universities.

Re: The danger of relying on OpenAI's Deep Research

#24

Post by giaour »

seanmcdirmid wrote: post #23
giaour wrote: post #22

Earlier quoted context omitted.

People didn't read books by making copies of them. Producing new copies of manuscripts was a skilled occupation.

Nope, that’s not true. You had monks copying books of course (mostly bibles), and very rich people could afford very expensive books made by scribes, but students at universities were expected to make their own copy during a reading/lecture run by a reader/professor. The words are still used in some older universities.

I was a medievalist before moving into tech, and I was always taught that students usually could not afford parchment on which to take notes, much less to copy down every word read out to them. Please cite a source for this claim.

Also, people bought books to read them outside of the context of universities.

Re: The danger of relying on OpenAI's Deep Research

#25

Post by dlahoda »

mrtesthah wrote: post #18
aeternum wrote: post #6

We could fix it all with a single law: All federally funded research papers must be published with full open access. The issue is the models can mostly only access abstracts (same as the majority of humans!) and thus we are back to the dark ages where knowledge is hoarded by the aristocrats in their ivory towers of academia.

> We could fix it all with a single law: All federally funded research papers must be published with full open access. Two things wrong with this sentence: 1. All federal funding for research is in jeopardy. https://www.nbcnews.com/science/science-news/trumps-nih-budg... 2. The executive branch now considers itself superior to the legislative and judicial branches, so actions of the legislative branch (i.e., laws ) m…

1. tax funded researchers did not publish non pay walled. right?

2. as per link text, there was denial to access treasure dep records. so what is secrecy in there?

so let just append '... and data' to the root of thread

Re: The danger of relying on OpenAI's Deep Research

#26

Post by birdfood »

Klaster_1 wrote: post #5
azinman2 wrote: post #3

The world is going to become stupider because it’s easier. This will be most dramatic for the youth in both school learning as well as in the field. Unfortunately I don’t realistically see another path in any reasonable time unless the world takes a radical turn away from AI (extremely unlikely).

In your opinion, what's the key difference that makes AI lead to world becoming stupider compared to other technologies that lead to more widespread knowledge, such as writing, book printing, calculators, computers and Internet?

With having internet in our pocket we no longer value being able to recall information. There is no need when anything is a query away. I suspect a similar thing will happen with our ability to think. What is the value in being able think through and solve a problem when you can just get your phone to do it instead? And I would argue in reality most people won't care if the results aren't correct so long as they are plausible and fast.

Re: The danger of relying on OpenAI's Deep Research

#27

Post by beeflet »

seanmcdirmid wrote: post #17
azinman2 wrote: post #15

Earlier quoted context omitted.

The printing press takes your ideas that you already wrote and increases distribution. Deep research is writing the text for you. That’s completely different.

I disagree, it’s just another step in the automation of knowledge distribution that’s been going on for a long long time. Every step someone has lamented that we lost something, maybe starting with speech itself, and the invention of writing taking away from oral tradition, scrolls, books, the printing press, the World Wide Web, etc… I was just wowing to myself the other day that a local 70b 4q 40GB model on my Mac b…

It's like wikipedia except it spits out incorrect information at an alarming rate and has no references. Still, impressive that it only weighs 10s of GB when wikipedia alone is on the order of ~100GB

Re: The danger of relying on OpenAI's Deep Research

#28

Post by seanmcdirmid »

giaour wrote: post #24
seanmcdirmid wrote: post #23

Earlier quoted context omitted.

Nope, that’s not true. You had monks copying books of course (mostly bibles), and very rich people could afford very expensive books made by scribes, but students at universities were expected to make their own copy during a reading/lecture run by a reader/professor. The words are still used in some older universities.

I was a medievalist before moving into tech, and I was always taught that students usually could not afford parchment on which to take notes, much less to copy down every word read out to them. Please cite a source for this claim. Also, people bought books to read them outside of the context of universities.

Most students couldn’t afford books. Here is one ref, which revisits pre-printing press university education in a piece about the last step people complained about (the internet)

https://opencontent.org/blog/archives/6104

> Before the printing press, faculty had to assume that none of their students had books. This led to a widely adopted practice known as dictation, in which faculty slowly read out the text for students so that they could make their own handwritten copies. (This was, of course, not the only mode of instruction. But it was a common one.) Blair writes that dictation was widely believed to have pedagogical merit, as “the act of copying out a text was often considered an essential part of mastering it” (p. 46). She provides a wide range of support for this view, going all the way back to Demosthenes and St. Jerome. However, others argued that dictation hurt student learning as the focus on writing distracted students from paying closer attention to the faculty themselves. (Presumably no one likes to stand in front of an audience and have them all looking down at their phones parchment the entire time.)

> You may be unsurprised to learn that there is a strong economic undercurrent to the conversation about dictation, and that it often pit students against faculty and others. Blair notes that students saw it as “a cheaper way of procuring oneself a classroom text” (p.45). Consequently, a ban on dictations by Arts Faculty at the University of Paris in 1355 “anticipated vehement student resistance to the ban.”

Re: The danger of relying on OpenAI's Deep Research

#29

Post by seanmcdirmid »

beeflet wrote: post #27
seanmcdirmid wrote: post #17

Earlier quoted context omitted.

I disagree, it’s just another step in the automation of knowledge distribution that’s been going on for a long long time. Every step someone has lamented that we lost something, maybe starting with speech itself, and the invention of writing taking away from oral tradition, scrolls, books, the printing press, the World Wide Web, etc… I was just wowing to myself the other day that a local 70b 4q 40GB model on my Mac b…

It's like wikipedia except it spits out incorrect information at an alarming rate and has no references. Still, impressive that it only weighs 10s of GB when wikipedia alone is on the order of ~100GB

Wikipedia isn’t a completely reliable source either. The LLM also includes a search engine. It’s weird since before I started playing around with it I thought it was just a fancy search engine and was grabbing sources from somewhere to shove into its context … but no, it’s all built in (although as deep research shows, it can do a lot better when it stuffs its context via research).

Re: The danger of relying on OpenAI's Deep Research

#30

Post by bruce511 »

aeternum wrote: post #6

We could fix it all with a single law: All federally funded research papers must be published with full open access. The issue is the models can mostly only access abstracts (same as the majority of humans!) and thus we are back to the dark ages where knowledge is hoarded by the aristocrats in their ivory towers of academia.

Right now, today, nothing stops academia from publishing every paper they write to be open to all. The internet is there and publishing is free.

The fact that they don't implies that they perceive some value is provided by publishing in the traditional way.

Assuming that journals die as a result of your law (why would people pay for a journal anymore?) is it worth considering that value (perceived or real) before advocating it is gone?

For example, consider a news article which starts with "a study published in Nature shows that..." versus "a study posted on Joe's blog on the internet indicates..."

For better or worse (and often with mistakes) journals act as a filter. Researchers in a field can "keep up". They can spend their time on "good papers" rather than trying to sift the dross of the internet.

Researchers, and by extension the place they work, gain karma points when publishing in hard-to-publish papers. If other papers cite their work they get even more points. Ultimately academia (employers and peers) ranks people using these points.

Does this mean your idea is bad? Not necessarily. But it's worth understanding that your proposal would "fix" one problem, but introduce a bunch of new ones.

Another (equally flawed) approach to solving the problem is that the govt simply pays for journal subscriptions to anyone who asks. This means "open access", with benefits limited to "their" population. But the values offered by journals are preserved. Of course this incentivizes journals to increase fees.

Post reply on HN