Skip to main content


Looking for technical stuff about DHCP. Google was returning lots of junk and so, wary that ChatGPT is a bullshit fountain, I asked it anyway.

It gave me several concise and to the point answers addressing exactly what I asked.

They were all completely wrong. It just confidently spouted bullshit the same way a narcissist bluffing about a subject they know almost nothing about will.

However, the bullshit it gave me allowed me to frame some pertinent google searches to actually get the information I wanted.

So I guess that’s a usage model for it: to guide your google searches.

The issue is that Google will slowly fill up with the output of the bullshit fountains, at which point we won’t be able to check their answers anymore, because they’ll effectively be marking their own homework.

Utter joy abounds (not really).

in reply to Sarah Brown

This…

We implement dnsmasq into our code using the modular nature of the C language. FTL v4.0 has always been multi-threaded for speed and efficiency. On startup, it launches a number of threads, each dedicated for specific tasks. We extend the already existing multi-threading in FTL to provide an even faster experience.

(Emphasis mine, which most of you won’t see because Mastodon has β€œNot Invented Here” syndrome on a huge scale)

…on the other hand, was written by a human, and β€œthe modular nature of the C language” is an interesting way to say, β€œjust write stuff onto the stack and then execute a jump instruction because that works with any old shit”.

in reply to Sarah Brown

I say this as someone who made a lot of money as a C programmer back in the 90s: C was of its time, and that time was 50 years ago. In the third decade of the twenty first century, C does not need defending any more; it needs a stake through its heart.
in reply to Sarah Brown

that's my day job. Except we write C in Java.

I... need to escape from this place.

in reply to Mirrors And Stuff

@mirrorsandstuff My day jobs writing C as well. I really need to get people in the industry to move over to rust.
Unknown parent

Sarah Brown

@Mirrors And Stuff That’s pretty much what we did at ARM, except it was transcribing VHDL and Verilog into C.

Some of the legacy code we worked with was BBC Basic from the early 80s that had been hand transcribed into C.

Unknown parent

Sarah Brown
@Josh Knapp :verified: At least it doesn’t do a Bing and go full-on narcissistic rage, demanding the world’s nuclear arsenal be trained on you THIS FUCKING INSTANT.
in reply to Sarah Brown

What would you like to know about DHCP? It's one of the bits of tech stuff I feel halfway competent to blather about...
Unknown parent

Sarah Brown
@Esther Was just thinking that we’re gonna have to go back to encyclopaedias.
in reply to Sarah Brown

ChatGPT wasn't built for information retrieval. The problem is isn't ChatGPT, it's the masses who are using it for tasks it's clearly not suited for.
in reply to newsorpigal

@newsorpigal
Then it should do a StackOverflow moderator persona "This is not a valid question or prompt for this service".
in reply to Simon Lucy

@simon_lucy

And how would it know? These are language models, they generate plausable text based on what it has seen before. They have no sense of truth or accuracy.

in reply to newsorpigal

@newsorpigal
Because there is governance and one of its use cases is a better search, for instance try a prompt for scientific papers on a reasonably specific question with references.

In that use case it appears to be useful and the papers seem genuine (it's best not to try and get it to make value judgements or make the purpose too opinionated).

in reply to newsorpigal

What is ChatGPT for then?
What's left then is a very fancy and expensive Lore ipsum generator
This entry was edited (1 year ago)
Unknown parent

prasoon
@esther
Or perhaps a good time to bring back trust networks like WebRings and curated URL directories?
in reply to Sarah Brown

I just had a similar experience when asking about a class of locomotive (I am a bit of a train enthusiast).

On a general question, the answer was accurate except for one detail. On a more specific question, all the information it gave me *sounded* plausible but was completely incorrect.

What was 'interesting' was that upon being told the information was incorrect, it apologised for the error, and came back with a different, yet still incorrect answer.

in reply to Sarah Brown

THIS is going to be the real fallout of #chatGPT and others: not students having their essays written by it, but the factually wrong and well-sounding garbage noise making it harder to find facts and actual sources.

Necessary consequence: learning to do better research and to better evaluate sources. Because this isn't new, regurgitating "I found this on Google" was never good research, but needs to be replaced by searching for the actual sources.

in reply to Sarah Brown

now I'm curious what you were searching for ... ISTM that Search has become a victim of its own success, and looking for highly technical info (i.e., searches with jargon and acronyms unique to a technical domain) yields rapidly diminishing returns because there's just _so much stuff_ out there and it's easier than ever to flood the zone ... so I got in the habit of tacking "reddit" on a lot of these searches lately and I'm happier for it
This entry was edited (1 year ago)
in reply to Sarah Brown

that's an interesting observation. It might be that search engines won't have to build in chat bots, they'll instead have to become good at filtering out the bullshit. Of course, that might lead to a useful chat bot.
in reply to Sarah Brown

"bullshit fountain" is a great neologism. HuMansplaining is my fave.
in reply to Sarah Brown

That image -- google filling up with the output of the bullshit fountains -- is going to haunt me for weeks

The truth is: yes, I am afeared

in reply to Sarah Brown

Weirdly it's possible to imagine a world where Wolfram Alpha is the sole survivor of the second search engine war because it's the only one that makes any attempt to trace the provenance of the elements of its inferences.
in reply to Sarah Brown

@esther it's very dated but this covers a lot of ground to help you get started to ask the right questions. Shame you need to know half the answer before you can do that. redbooks.ibm.com/abstracts/gg2…
in reply to Sarah Brown

To be fair, this is a tried and true method of getting quick answers on Stack Overflow: stating a clearly wrong answer gets the right answer much more quickly than asking an honest question!
in reply to Sarah Brown

@cstross We fed it our interview technical questions and it fell in to every trap. Its garbage in and garbage out!
in reply to Sarah Brown

I wonder if the model behind ChatGPT could be adapted to make this feature more explicit. Let's say for a given query it can produce some amount of suggested topics and/or terms that might be relevant for you. Basically strip the "generate bullshit" step and directly expose the "this information is adjacent to what you asked" knowledge encoded in it.
in reply to Sarah Brown

Just to put it in perspective:

So we invented this fancy World Wide Web-thingie with the intention of making it easy to share information.

Instead, apparently, we’re ending up spending an awful lot of energy on a pile of garbage no one can use for anything πŸ€”

Very convenient, indeed πŸ€¦β€β™‚οΈ

in reply to Sarah Brown

I think that maybe Google delivered too well to start with. It never promised to give the best answer (truest?) simply the most relevant for the question (most popular?)

Maybe we are seeing things hitting their 'eternal September' much more quickly?

After all, "Usenet is like a herd of performing elephants with diarrhea. Massive, difficult to redirect, awe-inspiring, entertaining, and a source of mind-boggling amounts of excrement when you least expect it."

— Gene Spafford, 1992

in reply to Sarah Brown

perplexity.ai gives ChatGPT-like answers BUT everything is linked to sources. It's really good.
⇧