Bay 12 Games Forum

Please login or register.

Login with username, password and session length
Advanced search  
Pages: 1 ... 27 28 [29] 30

Author Topic: Reactions to Chinese White Balloon of AI Awareness  (Read 28509 times)

Great Order

  • Bay Watcher
  • [SCREAMS_INTERNALLY]
    • View Profile
Re: Reactions to Chinese White Balloon of AI Awareness
« Reply #420 on: October 11, 2024, 10:39:45 am »

That is not a goddamn finger.

How are men meant to compete with that once it's mainstream? AI-powered prehensile penises are going to be the doom of this species.
Logged
Quote
I may have wasted all those years
They're not worth their time in tears
I may have spent too long in darkness
In the warmth of my fears

King Zultan

  • Bay Watcher
    • View Profile
Re: Reactions to Chinese White Balloon of AI Awareness
« Reply #421 on: October 12, 2024, 02:43:16 am »

Damnation they're giving robots penises, how are we supposed to compete with their robo boners?
Logged
The Lawyer opens a briefcase. It's full of lemons, the justice fruit only lawyers may touch.
Make sure not to step on any errant blood stains before we find our LIFE EXTINGUSHER.
but anyway, if you'll excuse me, I need to commit sebbaku.
Quote from: Leodanny
Can I have the sword when you’re done?

zhijinghaofromchina

  • Bay Watcher
  • I am支景灏
    • View Profile
Re: Reactions to Chinese White Balloon of AI Awareness
« Reply #422 on: October 12, 2024, 12:28:14 pm »

I think we may need to change the title since what is talked about has nothing to do with the Chinese white balloon , which makes it look like a common AI thread.
« Last Edit: October 12, 2024, 12:31:19 pm by zhijinghaofromchina »
Logged
问苍茫大地,谁主沉浮?我主沉浮!

Starver

  • Bay Watcher
    • View Profile
Re: Reactions to Chinese White Balloon of AI Awareness
« Reply #423 on: October 12, 2024, 12:59:04 pm »

I think we may need to change the title since what is talked about has nothing to do with the Chinese white balloon , which makes it look like a common AI thread.
Well, not really an AI thread (not everything is AI![1]). Maybe more of an AI thread...

[1] Everything is being 'given' AI, these days. Even(/especially) when it really doesn't need to be. +++Out Of Cheese Error ???????+++
Logged

lemon10

  • Bay Watcher
  • Citrus Master
    • View Profile
Re: Reactions to Chinese White Balloon of AI Awareness
« Reply #424 on: December 20, 2024, 04:45:53 am »

In distant eons past I commented that I felt like we were on the eve of revolution; the beast slowly slouching towards Bethleham.
AI progress in the interim has done nothing but strengthened these thoughts; the footsteps of the beast echoing ever louder.

We are *still* on the S-curve.
Speed has increased, what you can fit into a model has increased (notably its fairly common for a new small model to be roughly equal to a bigger model from the same company just a few months old), cost has decreased, increases in science/math/coding abilities, intelligence has increased, general utility have increased, ect.
Not just a little progress either, every month or two has seen significant increases across the board in basically all of the above.
---
The progress we are seeing is remarkable, a clear refutation against the claim that I see in places that AI development has stopped due to the datawall and the end of scaling.
Scaling needs three things to work properly: Scale (larger AI), compute, and data.

Since we've used almost all of the data, scaling models further is pretty weaksauce and doesn't do all that much, so there is *some* truth in the claim.

But it turns out to not have been that much of a wall really, sure, naive scaling hit its limit, but there are *so* many other things to do and try that progress has kept up just fine.
---
For instance AI can now think for a variable amount of time before it responds (o1), turns out not vomiting out your stream of consciousness and spending some time thinking first gives some pretty huge gains.
It also came with OpenAI announcing a $200 monthly subscription to use the best version of it, so its not all roses.

AI can also use your computer via looking at your screen as well (claude computer use). Not the best idea to let it run around buying stuff off amazon for you yet, but we are very much getting there.

LLM's can now also talk and sing and draw and analyze videos and pictures (4o) and all manner of new and novel things (advances in medicine, biotechnology, ect) that have mostly just passed without a peep.

Oh, and Sora finally came out, but its not even the best video generator anymore, so meh.
---
https://thezvi.substack.com/p/ais-will-increasingly-attempt-shenanigans

AI also has its first recorded escape attempts. Which... worrying? Especially given that models are now getting to the point where they are competent enough to self replicate.

We aren't *quite* at the point where we have self-replicating AI floating around the internet doing whatever it wants, for a wide variety of reasons, but its almost certainly coming soon.
The closed source ones would have to break out (difficult) which means that open source will probably be running around making botnets first. So Llama 4? 5 maybe?

I would bet on Llama 4, since its probably going to be roughly capable enough, and people will 100% do that shit on purpose to get it started.
---
Quote from: Joshua Achiam
A strange phenomenon I expect will play out: for the next phase of AI, it's going to get better at a long tail of highly-specialized technical tasks that most people don't know or care about, creating an illusion that progress is standing still.
Researchers will hit milestones that they recognize as incredibly important, but most users will not understand the significance at the time.
Robustness across the board will increase gradually. In a year, common models will be much more reliably good at coding tasks, writing tasks, basic chores, etc. But robustness is not flashy and many people won't perceive the difference.
At some point, maybe two years from now, people will look around and notice that AI is firmly embedded into nearly every facet of commerce because it will have crossed all the reliability thresholds. Like when smartphones went from a novelty in 2007 to ubiquitous in the 2010s.
It feels very hard to guess what happens after that. Much is uncertain and path dependent. My only confident prediction is that in 2026 Gary Marcus will insist that deep learning has hit a wall
(Addendum: this whole thread isn't even much of a prediction. This is roughly how discourse has played out since GPT-4 was released in early 2023, and an expectation that the trend will continue. The long tail of improvements and breakthroughs is flying way under the radar.)
A bit of prophecy about the future that (to me) seems fairly inevitable at this rate.
Logged
And with a mighty leap, the evil Conservative flies through the window, escaping our heroes once again!
Because the solution to not being able to control your dakka is MOAR DAKKA.

That's it. We've finally crossed over and become the nation of Da Orky Boyz.

King Zultan

  • Bay Watcher
    • View Profile
Re: Reactions to Chinese White Balloon of AI Awareness
« Reply #425 on: December 21, 2024, 03:54:55 am »

I thought this thread was the joke thread and the other AI thread was the serious one.
Logged
The Lawyer opens a briefcase. It's full of lemons, the justice fruit only lawyers may touch.
Make sure not to step on any errant blood stains before we find our LIFE EXTINGUSHER.
but anyway, if you'll excuse me, I need to commit sebbaku.
Quote from: Leodanny
Can I have the sword when you’re done?

Starver

  • Bay Watcher
    • View Profile
Re: Reactions to Chinese White Balloon of AI Awareness
« Reply #426 on: December 21, 2024, 01:03:40 pm »

I thought this thread was the joke thread and the other AI thread was the serious one.

That's what thAI want you to think!
Logged

eerr

  • Bay Watcher
    • View Profile
Re: Reactions to Chinese White Balloon of AI Awareness
« Reply #427 on: December 22, 2024, 04:49:45 pm »

bear in mind that there's also benchmarks that basically define how an ai is rated.
what is testable is teachable, or so cgp grey says

so if people start making incredibly useful rating systems that apply to ai outputs, we'll have a huge rise in ai tech.

there was some article about how ai's usefulness was approaching the limits set by the current system, but i can't be bothered to find it.
Logged

lemon10

  • Bay Watcher
  • Citrus Master
    • View Profile
Re: Reactions to Chinese White Balloon of AI Awareness
« Reply #428 on: December 23, 2024, 05:27:59 am »

I thought this thread was the joke thread and the other AI thread was the serious one.
*Checks last post date on other thread*. Whoops, yeah, should have posted in that one.
Anyways, don't really want to bump that one for no reason when this one is on the front page now.
---
bear in mind that there's also benchmarks that basically define how an ai is rated.
what is testable is teachable, or so cgp grey says

so if people start making incredibly useful rating systems that apply to ai outputs, we'll have a huge rise in ai tech.
Broadly correct, although benchmarks and other tests are merely one piece of the puzzle, their existence allows progress, but like all of the other pieces (data, compute, scale, frameworks, structure, ect), they merely allow progress, and are not enough on their own to guarantee it.
ai's usefulness was approaching the limits set by the current system, but i can't be bothered to find it.
Ehh. Its true that most benchmarks are getting close to saturation (aka, when AI beats it, so it can't get better by using them), but that just means more harder benchmarks are created.

For example the FrontierMath benchmark was just released, with the aim of making *really* hard math problems that take the absolute best mathematicians in the world to solve.

Of course for some things we'll run out of ability to measure how good it is (eg. compressing and quantifying the difference between a monet/rembrant and [other very good paintings in the same field by less good but still well regarded painters] into a benchmark may prove to beyond human capabilities), which yeah, will cause problems in AI progress, but we aren't there yet.
---
Quote from: Sam Altman
today we're going to talk about that next Frontier Model um which you would think logically maybe should be called O2 um but out of respect to our friends at telefonica [as in, they don't want to be sued] and in the grand tradition of open AI being really truly bad at names it's going to be called 03
Spoiler: Benchmark/test images (click to show/hide)
In other really big news o3 was announced, and the difference between it and o1 (already the best AI in the world, notable for being the first AI ever to properly "think" for variable amounts of time) is massive.
Especially notable is its improvement on the new FrontierMath, going from the literal second best AI in the world having 2%->25%, or getting what (apparently) amounts to like a 99.8 percentile (compared to humans, and up from 93% for o1) on codeforce.

It (along with the $200 pro subscription and the $500 devin subscription) puts the final nail in the coffin for everyone rich and poor alike getting access to the same quality of AI.

This change is fairly conclusively proven with nothing more then its performance on the ARC-agi-pub benchmark.


AI can now think for arbitrary amounts of times to solve a problem, and this thinking time directly and continuously equates to being better able to solve a problem in much the same way it would for a human.
And when I say arbitrary I mean it, note the o3 (high) mode, where they spent literally thousands of dollars of compute on every single one of the 500 test problems.
Quote from: Dude from the benchmark company
Despite the significant cost per task, these numbers aren't just the result of applying brute force compute to the benchmark. OpenAI's new o3 model represents a significant leap forward in AI's ability to adapt to novel tasks. This is not merely incremental improvement, but a genuine breakthrough, marking a qualitative shift in AI capabilities compared to the prior limitations of LLMs. o3 is a system capable of adapting to tasks it has never encountered before, arguably approaching human-level performance in the ARC-AGI domain.
...
Passing ARC-AGI does not equate to achieving AGI, and, as a matter of fact, I don't think o3 is AGI yet. o3 still fails on some very easy tasks, indicating fundamental differences with human intelligence.
This is a huge breakthrough in AI. They still don't have the fundamental ability to learn like a human does (due to humans brains changing over the long term as we learn), but its hard to overstate just how important a breakthrough o1 + o3 represent.
---
Its still very early days with o3 and variable thinking, but like the rest of AI over the past year, its going to get much much better in basically every way.
Logged
And with a mighty leap, the evil Conservative flies through the window, escaping our heroes once again!
Because the solution to not being able to control your dakka is MOAR DAKKA.

That's it. We've finally crossed over and become the nation of Da Orky Boyz.

Duuvian

  • Bay Watcher
  • Internet ≠ Real Life
    • View Profile
Re: Reactions to Chinese White Balloon of AI Awareness
« Reply #429 on: February 18, 2025, 05:06:38 am »

I noticed I now had an unsolicited AI on my Start menu. This time I hesitated before uninstalling. Usually I'm just glad that uninstalling is an option and I won't have to figure out how to half-break the thing or learn a magic spellcode from the internet to dispel it. I opened it up without an internet connection to see what it would do. It did nothing. After calling support, I took their advice and plugged in the internet.

THE DUUVIAN MEETS THE ROBOT

I forgot to copy this world's first. However I can summarize it:

First question: Can you draw? <windy yes>

Second question: Draw me a sexy dinosaur
I can't do that

3: Draw me a dinosaur holding a handgun
I can't do that
3a: I managed to find something silly the robot could draw, however this required an account so I skipped that.

4. Summarize Epictetus on Freedom
The robot supplied an accurate and detailed answer

5. Compare the above to Bacon
The robot cleared the low bar of understanding I mean the philosopher. It described that in compatible views one philosopher focussed inward while the other focussed outwards. I was most impressed by this.
-------------------
I think after this I switched to eclectic stuff.
6. Write a new Pootie Tang song
The robot goofed this one. It generated a song it claimed was in "Pootie Tang's style", a style which it described using adjectives and adverbs I can't recall that were generic, like a modern music style random generator or something. It was kind of cool I guess but it seemed like it was generating a format for lyrics so to speak. I pointed out that while the lyrics were acceptable, Pootie Tang does not express his meaning in English words. The robot apologized and generated another song in nearly the same format, but in a moment of triumph at the end it included a new Pootie Tang phrase which was awesome.
------------------
There is an old SCP foundation entry that I can't find anymore. I asked the robot about it.
7. What is the SCP Foundation entry that features a zone that sends out balloons and revifies animals with cybernetics?
The robot hallucinated here. It claimed that this was an SCP known as the Samsara Zone and was SCP number 005. I knew this wasn't correct as it would have to be a much higher classification number. It started making things up here (I was mildly spooked out when I thought it chose the word Samsara after I read the wiki page for it when I searched for the SCP page but I realized later it chose it due to the Samsara SCP team having cybernetics) until I asked it what SCP 005 was, and it listed and described Skeleton Key which is correct. I asked the first question again and it gave SCP 005 again. I asked how both could be 005 and it changed it's answers completely. It said it had made a mistake, the Samsara Zone was not an official SCP entry but there was an SCP containment team called Samsara. I can get that part again actually:

Spoiler (click to show/hide)

Anyways after having it list all the SCP entries that involve balloons, zones, and other stuff that might be related. The robot suggested it may be an unofficial SCP or fanfiction I had read at some point,to which I conceded the possibility.  I eventually asked it if the SCP Foundation has gone through removals and revisions of old SCP entries and it confirmed that it had, so I concluded upon considering that a likely explanation.

8. What is quantum radar
The robot gave a summary that seemed accurate.

9. What are the implications of quantum radar to the US military?
The robot gave a accurate summary as far as I'm aware.

10: What countries have a prototype quantum radar?
The robot provides the list.

11: How to uninstall
Robot apologizes for not being able to remove itself and provides instructions

12: How to uninstall Microsoft Edge
Instructions for this surprisingly complicated task are given

13: Why is Microsoft Edge hard to uninstall?
Reasons given explain why Edge is included on the OS

Spoiler (click to show/hide)

EDIT: I mispelled Pootie Tang this time which is my fault but the robot did not ask to correct as I may have preferred, but it made some acceptable to pretty good lyrics anyway. I'm wondering if I should have told  it what timing or tempo or whatever other stuff so I could plug it in to my music software somehow. I think this thing does audio but I think it required an account so I didn't do that.

 
Spoiler (click to show/hide)

Spoiler (click to show/hide)

Spoiler (click to show/hide)
« Last Edit: February 18, 2025, 06:25:33 am by Duuvian »
Logged
FINISHED original composition:
https://app.box.com/s/jq526ppvri67astrc23bwvgrkxaicedj

Sort of finished and awaiting remix due to loss of most recent song file before addition of drums:
https://www.box.com/s/s3oba05kh8mfi3sorjm0 <-zguit

martinuzz

  • Bay Watcher
  • High dwarf
    • View Profile
Re: Reactions to Chinese White Balloon of AI Awareness
« Reply #430 on: February 18, 2025, 05:56:40 am »

What's more horrifying than an army of terminators sent by Skynet?
An army of terminators sent by Skynet, each chanting their own unique Pootie-Tang inspired phrase.
Logged
Friendly and polite reminder for optimists: Hope is a finite resource

We can ­disagree and still love each other, ­unless your disagreement is rooted in my oppression and denial of my humanity and right to exist - James Baldwin

http://www.bay12forums.com/smf/index.php?topic=73719.msg1830479#msg1830479

lemon10

  • Bay Watcher
  • Citrus Master
    • View Profile
Re: Reactions to Chinese White Balloon of AI Awareness
« Reply #431 on: February 18, 2025, 06:15:59 am »

Quote from: GPT with web search on
What is the SCP Foundation entry that features a zone that sends out balloons and revifies animals with cybernetics?
ChatGPT said:
The SCP Foundation entry you're referring to is SCP-962. This anomaly is a large metal tower that produces cybernetically augmented animals, known as SCP-962-1, which perform various tasks such as mining, repairing, and hunting non-human life. Additionally, SCP-962 releases hydrogen-filled balloons carrying manuscripts of varying content and style.
https://scp-wiki.wikidot.com/scp-962
Its correct btw. Incidentally this is the exact type of problem AI has historically had big problems with. Integrating web search goes a long way to making it more accurate here. Note their thinking model (with web search off) didn't have the answer, but it also said it didn't know and had to look it up yourself since it couldn't.

I don't know what copilot uses for AI, but its certainly behind what you can get for free (pretty sure you don't even need an account for websearch/reasoning, but picture gen needs one) on the GPT website. Of course if you pay you get better models, and if you pay $200 a month you get some stuff way better (eg. deep research; which does an extensive search and writes up a multi-page extensively cited report to answer your question like this one for example).
---
Spoiler: Song gen (click to show/hide)
I know nothing about Pootie Tang, so I have no clue if this is any good. Do you think its better then what you got from copilot?

---
Its important to note that even over the last ~2-3 months there have been rapid and substantial improvements in both cost and capabilities.
In another few months we will get GPT 5 (and GPT 4.5 before that), progress is roaring ahead as fast as ever.
Given a few months is still enough for multiple OOM of effective efficiency and we have still have no clue how these things work in tons of ways I suspect that we still have a quite a long time left in the S-curve as long as the money keeps flowing.
Logged
And with a mighty leap, the evil Conservative flies through the window, escaping our heroes once again!
Because the solution to not being able to control your dakka is MOAR DAKKA.

That's it. We've finally crossed over and become the nation of Da Orky Boyz.

Duuvian

  • Bay Watcher
  • Internet ≠ Real Life
    • View Profile
Re: Reactions to Chinese White Balloon of AI Awareness
« Reply #432 on: February 18, 2025, 06:34:09 am »

If I remember correctly, they're both wrong. I don't know if the movie actually plays a full length song. Pootie Tang speaks in pidgin or patois some such correct term that is understandable by the people around him and can usually be inferred by the audience. AFAIK this extends to his songs, one of which was complete silence if I remember correctly. It's been a while since I've seen the movie.

A lot of the stuff in the second generated song is phrasing pulled from the movie, and it references a villain character in Dirty Dee where it would be more appropriate to reference Biggie Shorty. I would guess it has a format it's mixing references into without understanding the nature of Pootie Tang's speech.


I also forgot he mixes in some English words when I ran it through Copilot, I told it to take those out so it may have been actually doing a better job than I thought.

EDIT: Thanks for the info, I guess I should check these out at some point. I never used the big social medias and those just sound like nightmares according to all other media. Should I tell the robot someone found the thing I was asking it to find? Will it appease or anger it? Will that help the next questor in search of wisdom? Does it waste my time to do that because there are safety controls to keep me from training the robot? I assume this is the case, but it seems like it might to some degree. I am not sure how to distinguish polite dismissal from this happening. I guess I will have to settle for increasing power cost to society by generating 20 million Pootie Tang phrases. Just kidding, 1000 is enough for my plans (whatever number it provided without account might suffice... maybe... no guarantees) and more than I could have ever reasonably hoped for in the past.

EDIT2: To clarify a bit, the Copilot song I generated seems to "get" Pootie Tang as a character better than the second one you provided. However yours pulls more references to include, though as I noted it placed a villain where it would make more sense for an ally of the protagonist. It later includes both the ally and the villain in the same location in a way that wouldn't make sense, because Dirty Dee would likely only be referenced in a negative sense.
« Last Edit: February 18, 2025, 07:44:46 am by Duuvian »
Logged
FINISHED original composition:
https://app.box.com/s/jq526ppvri67astrc23bwvgrkxaicedj

Sort of finished and awaiting remix due to loss of most recent song file before addition of drums:
https://www.box.com/s/s3oba05kh8mfi3sorjm0 <-zguit

lemon10

  • Bay Watcher
  • Citrus Master
    • View Profile
Re: Reactions to Chinese White Balloon of AI Awareness
« Reply #433 on: February 18, 2025, 07:43:55 am »

EDIT: Thanks for the info, I guess I should check these out at some point. I never used the big social medias and those just sound like nightmares according to all other media. Should I tell the robot someone found the thing I was asking it to find? Will it appease or anger it? Will that help the next questor in search of wisdom? Does it waste my time to do that because there are safety controls to keep me from training the robot? I assume this is the case, but it seems like it might to some degree. I am not sure how to distinguish polite dismissal from this happening. I guess I will have to settle for increasing power cost to society by generating 20 million Pootie Tang phrases. Just kidding, 1000 is enough for my plans (whatever number it provided without account might suffice... maybe... no guarantees) and more than I could have ever reasonably hoped for in the past.
I wouldn't bother telling the robot it was found, but you can if you want. Putting the correct answer would give the company more high quality training data, which would indeed to some infinitesimal degree help after the next training run.
One conversation would make no difference given the sheer quantity of training data, but if you did thousands or tens of thousands of posts with it about Pootay-Tang it could totally help the next version of it be more tangy.

You could also use the API and spend time writing a good card for Pootie; you might even include the entire movie script in context. Good prompting and context makes a world of difference to getting what you want.

Finally if you want you could throw the screenplay and stuff on the movie into the AI and finetune it; bit overkill for something like this and by far the most annoying way to do so, but gives the best results.
« Last Edit: February 18, 2025, 07:52:13 am by lemon10 »
Logged
And with a mighty leap, the evil Conservative flies through the window, escaping our heroes once again!
Because the solution to not being able to control your dakka is MOAR DAKKA.

That's it. We've finally crossed over and become the nation of Da Orky Boyz.

Duuvian

  • Bay Watcher
  • Internet ≠ Real Life
    • View Profile
Re: Reactions to Chinese White Balloon of AI Awareness
« Reply #434 on: February 18, 2025, 07:54:31 am »

Swippity-snap!
Logged
FINISHED original composition:
https://app.box.com/s/jq526ppvri67astrc23bwvgrkxaicedj

Sort of finished and awaiting remix due to loss of most recent song file before addition of drums:
https://www.box.com/s/s3oba05kh8mfi3sorjm0 <-zguit
Pages: 1 ... 27 28 [29] 30