alt Hacker NewsShow HN: Gemini Pro 3 hallucinates the HN front page 10 years from now
Comments
I continue to be unimpressed by LLMs when it comes to creative work, they're certainly useful sometimes for "reference digging", but maybe I just don't understand enough about how they work and this is actually something that can already be "fixed" or at least optimized for; anyway, one of the headlines is:
> Debian 18 "Trixie" released
While it correctly derives that a likely version number in ten years would be 18, as there are new releases approximately every two years which means +5 from today's version 13, it then goes on to "make up" that the name of it would be "Trixie" -- the same name as the current release in 2025.
Debian has never re-used a release name, and I think we can be pretty confident they won't (as will no other Linux distro), so I would expect it to "understand" that:
- The next Debian release always uses a previously non-used Toy Story character
- Based on this information, _any_ name of a Toy Story character that hasn't been used is fair game
- At the very least, it certainly won't be the same name again, so at least make up a name
So in 10 years from now, fusion will still be 10 years away
I tried this on a bunch of chat webuis and they all failed miserably for single html file. Finally went with Replit and allowed it to just default to whatever stack it decided to use (it ignored my single file html directive and used react but said it "functions as a single monolith -- lol ok).
Anyway, turned out decent. Oh I had it do the articles and comments too, that's probably why I mostly got failures.
A similar project of mine (generated back when GPT-4 was out, along with comment threads and articles):
Inspired by this post, I have asked chatgpt what the mail inbox of a university professor will look like in 10 years. Not everything was funny, but here are some good ones:
- “Hi Prof, Sorry for Late Reply—My Smart Fridge Locked Me Out (Assignment Attached)”
- “URGENT: PhD Student Accidentally Achieved Sentience (Needs Committee Approval)”
- “Question About Homework 3 (Submitted by My AI Twin, Is That Okay?)”
- “Re: Firmware Update Bricked Entire Smart Campus — Students Request Deadline Extension”
- “Grant Review Feedback: ‘Too Secure for Real-World Deployment’”
- “Can I Get Partial Credit? My IoT Implant Was Flagged as Malware”
- “Reminder: Mandatory Annual Ethics Training for Autonomous Systems (40 minutes)”
The title is misleading. This isn't the correct use of the term "hallucination". Hallucination refers to making up facts, not extrapolating into the future.
I read 10 comments before I realized that this was referring to 10 years in the FUTURE and not in the PAST (as would be required for it to be a hallucination).
Honestly? I kind of like this hypothetical future this purports to be a snapshot of. Sounds hopeful.
(EDIT: Not so much after seeing the expanded variant with full submissions and comment threads...)
Also, nailed HN perfectly.
I very much hope the cost of Microsoft 365 is actually $40/month/user 10 years from now!
I would have guessed $200+.
Reminds me of "The Onion's Future News From The Year 2137"
I used the same prompt keepamovin used and changed it to CNN, which produced this:
https://realms.org/pics/cnn.html
Some interesting similarities.
Hopefully something as optimistic as this and not the Quiz Broadcast :) https://www.youtube.com/watch?v=22mt0cVyW5c
The only thing that this is missing to make this perfect is "2035: The Year of the Linux Desktop"
ITER 20 minutes net positive in 10 years time?
How does that alter the "fusion in 50 years" effect? I suspect not much, there will be a huge 80/20 element. "next 20 minutes cancelled after extensive embrittlement found, ITER now in decontamination mode for 3 months before examinations can begin"
One of the entries is:
> How to build a Faraday Cage for your bedroom
I'll get there eventually. A friend and I always joke about this since it took me to 2021 to get a smartphone (by choice).
This is awesome and hilarious.
Archived here so we can compare 10 years from now: https://archive.is/CplcH
I'm missing an announcement like: 2nd confirmed person paid for their winrar license!
By the time I read everything and pressed the back button only to be greeted by another hacker news page I felt like I just went back in time 10 years to a more plesant era.
I would appreciate the same thing written by a human. That human would probably have lots of fun writing this.
ITER achieves net positive energy for 20 consecutive minutes (nature.com) 1205 points by physics_lover 12 hours ago | hide | 402
-Sustained fusion still only 10 years away!
I made this with Lovable as well, it's interesting how it's a bit more extreme in terms of where we are in 2035.
https://hn-frontpage-2035.lovable.app/
> The last JavaScript framework: Why we stopped at React 47 (overreacted.io)
I posted something similar from Grok 9 months ago, although it was “flagged” for some reason. the link still works.
Great fun! I was hoping to see "Perl 6.1 goes live" though. Maybe later eh? ;)
i'm kind of having trust issues with HN comments now. I can barely detecy anymore which ones are bots or humans.
Y’know what? This feels like the most plausible future outcome in a lot of ways.
* All the big LLM companies are dead/dying or their infra just never panned out (like Google killing Gemini)
* We’re still fighting the advertising wars
* Restoration of kit continues to be a thing we all love/upvote to see
* More focus on local stuff rather than ginormous moonshots
* Speaking of moonshots, SpaceX continues succeeding despite Elmo’s best drug-fueled benders trying to undermine it like he does Tesla and X
* Clickbait science posts that the comments rightly point out as such while bitching about “can’t keep these things off the front page”
* People still trying to make WASM and web-as-an-OS a thing, bless their hearts (shine on, you brilliantly weird diamonds, I love seeing what you build even as I doubt the practicality of it)
If this is even half-accurate, I will be pleased we didn’t obliterate ourselves in the process, though disappointed another decade went by dealing with the same fucking problems we’ve had for twenty-odd years.
Sad to see Half Life 3 hasn't been released yet.
Freaking awesome. You should extend clicking on a link, similar to how this article describes infinite content:
https://worksonmymachine.ai/p/solving-amazons-infinite-shelf...
There were 666 comments here when I opened this post. I hope that is not a foreboding omen. ;-)
Very nice. But posts about Rust and Zig won't make the frontpage anymore of course in 10 years from now, they will be soooo old fashioned.
I can't think of a better example of AI hauntology than this. A future imagined from remixed web scrapes of past events.
> “Why functional programming is the future (again) (haskell.org)”
Ouch, that hurts!
There is going to be a lot more interest in what is happening in China and much less interest in what is happening in Europe. This hallucination has the same western world biases of today.
What I like it most is how "unoriginal" the result is. It is mostly the current trends, but in the future, rather than a blend of that + unknown future.
That was funny. I wish you went ahead and had it also create at least the top comments for each thread!
So awesome, reminds me of Wired's Found from Future series (I don't know why they ever retired it...continuing that series as a sort of fan fiction was the first thing I did with gen AI when image generators started to get good)
Anyone who says LLMs don’t have a sense of humor should be shown this page.
> The unexpected return of server-side rendering (htmx.org)
Glad to know this topic is still thrashing and spasming and refusing to die.
If true, this is the funniest thing I have seen an LLM generate.
While the style and headline seems like Hacker News, the usernames seem increasingly alike Slashdot.
Wait... I had to do a double take for this one.
To me, the funniest thing is the AlgoDrill leet code post being on the front page both today and ten years from now.
Subtle and hilarious.
This is so unbelievably based. Very nice!!!!
I love this but also funny that it includes the Nia and NATS posts from today.
Clicked link as I looked away, looked back and though, "Huh didn't work" Then started reading the headlines...
> Ask HN: How do you prevent ad-injection in AR glasses?
Yeah, this is gonna be an actual problem in the future. See this for the worst case scenario: https://vimeo.com/166807261
Is this really Gemini because it shows a great deal of understanding in terms of time applied to common current headlines that you could argue it is satire. I’m not sure I believed AI could do this…
I realy need this: How to build a Faraday Cage for your bedroom. please upload!
It did a good job with the points-to-comments ratios. The purely technical stories are higher than industry news and clickbait.
Interesting how controversial Zig will be in the future :)
It seems that when we predict the future, we tend to project what is important or trending now as what will be important or trending then. This is showing in this llm produced bit as well, what with the regular topics on rust, zig, why I'm coding with my two human hands, spacex moon exploration and whatnot.
That must be some kind of bias, you find that in sci-fi as well where even futuristic societies usually have the same tech as today, similar systems of governance and structure of society, just incrementally better or marginally changed (unless that's a device essential to the plot like society is now berserk).
Ps: love that #8 is Google killed gemini
lol @ the HTMX link and unexpected return of SSR. We've been talking about that for years at this point, and it's still going on 10 years from now unexpectedly?