What’s interesting to me is rate which AI is improving. The new extensions being released on a daily basis is overwhelming.

Writing it off is mistake imo.

Here’s a year difference in Midjourney art AI.

IMG_4495 (resized).jpeg

Has AI found the cure to cancer yet?

Of course they are as AI is the big catch phrase right now in tech. Before it was "The Cloud" which was supposed to replace everyone as well in tech but that never happened either.

Who knows, maybe Twitter had it's big meltdown this week (not mentioning topic here as it's political related) as Elon laid off half the company. AI, automation, or whatever being used really helped that day.

Yesterday was carry a towel day. Douglas Adams was the shit!

I'm waiting for AI like Chat to become "monetized" just like Google, its only a matter of time before humans use it for marketing.

New Drake song

Written and preformed by AI

Imagine taking a Boston dynamics robot and sticking an advanced AI in its head.

Personally, I don't consider ChatGPT to be AI as it does not actually reason but is simply mere data processing. While it does contain machine learning (both through parsing articles and evaluating user feedback), this is only one of many elements of true AI. Much of what it is programmed with dealing with cultural issues are politically correct opinions. The system is not truly autonomous, but fully relies on programming constructs. However, it is certainly a giant step towards the development of AI.

I've both seen and had personal "conversations" that clearly bring out the inconsistencies and show programming biases in the system. However, this doesn't mean that it is not valuable. I use it on occasion for what I would consider to be advanced Google searching. In other words, you can find answers to a question/topic, but then quickly and easily pose follow-up questions to drill down or broaden your inquiry while remaining on topic. Also, it consolidates the information in a convenient response instead of the user having to read and navigate through the search results to compile valid answers to their questions.

As far as the art rendering AI engines, these are pretty sweet and I use them frequently for design. Sometimes the results are perfect, and others times they get me 90% to where I want. Music AI engines and things like Auto Tune have come a long way, but it seems to me for now they are too perfect (e.g. perfect pitch, etc.) I'm sure it won't be long before they are able to program the imperfections of humans to make such products truly indistinguishable from reality. After all, it even took The Matrix several iterations before it realized it needed to purposefully program imperfections.

BTW, "AI" seems to be the trendy buzzword these days. Whether articles, news stories, commercials, etc. that I see, it seems that everyone has tagged their products and solutions with either "AI" or "Powered by AI", even though just a week ago many of the same things were simply products with programmed logic in them.

Whether it was "Terminator", "The Matrix", "I,Robot", or countless other movies on the topic, this never turns out well for humanity.

Well, this thread does reveal one thing. A very small number of you understand ChatGPT and the current state of AI. And the majority do not.

https://news.mit.edu/2020/artificial-intelligence-identifies-new-antibiotic-0220
and a more recent example: https://www.news-medical.net/news/20230525/Scientists-use-AI-to-identify-new-antibiotic-that-could-fight-drug-resistant-infections.aspx

Very intersting.

Seems...sketchy to me. Notice it doesn't mention if it has access to the nuke codes.

pasted_image (resized).png

Process that through a little more. Keep in mind, you and I are not AI experts, but run through how the above COULD happen: AI looks at EVERY rulesheet. It watches every video of pin gameplay. It reads reviews...every review ever published...to understand what people like and don't like. I don't think it's farfetched to think this is a possibility in the next decade.

To quote Bill Gates: "We always overestimate the change that will occur in the next two years and underestimate the change that will occur in the next ten."

SkyNet has been around since the 1960's Britain - satelites..

Made popular in The Terminator movies.

Pins are going to become self aware

I honestly think that the internet has evolved in the last month or so, and most people have no idea. ….fall of the Berlin Wall, 9/11, Covid, and now GPT-4. I’ve lived through a couple globe changing events and this is one of them even if you haven’t realized it yet. I’m sure a couple of the more senior pinside members will even remember a time before JFK. We’re entering another new world and the next couple years of change are going to be a little rocky

Tech fads come and go.
Generative AI!
The Cloud!
The metaverse / VR
Crypto (which isn't used for anything except securities fraud)
Mobile Apps (2012 everyone was making apps and it was how you get rich)
Web 2.0
Pets.com

Some of the above are viable businesses (AWS makes money for sure), some are not. The Cloud did not eliminate sysadmins (in fact, it introduced an entirely new field of developers, DevOps engineers). AI won't eliminate programmers. Midjourney won't eliminate designers. But these tools will make them more efficient. I don't believe it will make them so efficient we will only need 10% of the current engineers or designers that we have in the industries now.

Conflict leads to good stories. The peaceful march of progress makes for a boring action movie.

Yeah, as The Terminator theme plays in my head , I'm thinking the soundtrack to such a movie wouldn't be as emotive either.

Dang it, now after replying to you the Superman movie theme is stuck in my head. BTW, I rewatched the movie a month or so ago after being decades since I last saw it. I have to say, I still enjoyed it and appreciated Christopher Reeve's portrayal even more.

Yeah... sketchy. Pretty deflective that these people don't think some stupid engineer wouldn't bridge the gap between virtual and physical worlds. Then "boom" skynet.

I just interacted with it tonight (my first time) to try and reverse engineer a 2FA algorithm for a takeout restaurant I ordered from tonight... just to see if I could make it work. My data set wasn't big enough... but I image when I have more data and could provide it with enough info - It might be able to identify the algorithm. IT was only an example case as I can see no reason someone would spend time to hack a local Chinese restaurant's 2fa order system.

That said; I could see someone abusing it to maybe do something more financially viable.

What I want to see is these "https://www.youtube.com/c/ScammerPayback" type heros use ChatGPT/AI to keep these damn scammers so busy wasting time with AI that they can't scam their victims. I know I can't be the first one to come up with this idea. Could you imagine how frustrated they'd get with an nearly flawless AI.

Keep in mind it's been trained on GitHub. There's billions of lines of code in there to learn from. There's nowhere near 0.001% of that amount of information for pinball tech. Plus it hasn't been trained specifically for it.
Also, even if they tried to, the code committed to GitHub is generally working, whereas it's harder to be sure that technical discussions on pinball forum end with a working fix.
One thing I did try that's pretty impressive is that I fed it a schematic from ipdb and it made it sound like it sorta understood it, although I'm pretty sure it wouldn't really have been able to find an actual issue.

Thank you ChatGpt, I now know how to use The 3 shells ....

finally-the-mystery-of-the-three-seashells-has-been-solved-v0-mkscv1k80rga1 (resized).png

What you're missing is all the misinformation here. The general suggestion here for most every repair is to reflow the solder on the header pins, which is rarely needed and rarely works. Too many titles from too many manufacturers (860 currently on IPDB and growing) over almost 100 years of games for AI to work.

I don’t buy it. Chat GPT or whatever AI solution will never replace actual human ingenuity and artistic ability. Chat GPT isn’t going to create the next LOTR or Godzilla ruleset. Also, there’s currently no way for AI to analyze pinball gameplay videos. I doubt there ever will be and if it ever does will it know what good gameplay is?

This AI hype is overblown IMO and is the current big tech buzzword. It’s a tool and nothing else.

AI is just the new Elon Musk.

In that countless people want to sound smart will randomly insert something about AI into their Pinside posts. Like they did with Elon Musk for several years.

Its role isn’t to replace. In more like this. Someone who is a pinball coder can use ai to create code 10x faster than before. Or one coder can do the work it used to take 10. Human in the loop in still needed. For now.

LOL. Ya'll saw it here people... PanzerFreak 's Bill Gate 640k moment.

I have an older buddy of retirement age that used to get his Son In Law to help him write C code for the Arduinos to create Pinball Light shows. I met with him today to deliver some parts from my webstore and we visited for a couple of hours. Apparently; he's graduated to using Chat GPT to help him generate the code/lightshows for his wall art.

I have another friend whom is using Chat GPT to come up with plot ideas for a scifi book he's writing. He uses it to remove writers block and to come up with different plots. He does the acutal "writing".

I do honestly believe ChatGPT / AI will be used more regularly for things like this. It'll become another tool in the arsenal to get shit done.

Here's a great example of why AI can't be used on something as complicated as pinball repair. Ripped from today's headlines:

'A lawyer used ChatGPT and now has to answer for its ‘bogus’ citations / A filing in a case against Colombian airline Avianca cited six cases that don’t exist, but a lawyer working for the plaintiff told the court ChatGPT said they were real.'

https://www.theverge.com/2023/5/27/23739913/chatgpt-ai-lawsuit-avianca-airlines-chatbot-research

'ChatGPT said they were real.' Hilarious.

I found this on the web to help explain why GPT-4 is better than chatGPT. If you think chatGPT is good then GPT-4 will amaze you. If you think chatGPT has limits than be prepared to watch them disappear.

ChatGPT and GPT-4 both use a transformer-based architecture as part of a neural network that handles sequential data. ChatGPT is based on GPT-3.5 so it is less advanced, has a smaller number of potential parameters included, and its data may be a little more dated. Thus, it is less accurate and lags GPT-4 in results to a greater degree, especially as the complexity of the problem or challenge rises.

ChatGPT in its current form seems to perform well in chatbots, language translation, and answering simple questions. But GPT-4 is smarter, can understand images, and process eight times as many words as its predecessor.

Reporters will no doubt continue to manage to fool both of them in various ways, but GPT-4 is far less prone to errors and hallucinatory responses. It can even explain why jokes are humorous. Thus, users should go straight to GPT-4 as it contains everything that is in ChatGPT (based on GPT-3.5) and far more besides.

I’m sure it’s super fun and efficient debugging code that was written by AI and must be understood by the “coder” (or shall I say “debugger”) first. Oh, that’s going to be AI too?

So in the future, Stern fires every developer and just buys an AI subscription and the CEO types in “develop the new “Back To The Future” pinball machine in the style of the 1985 movie” and bingo, it’s all done?

You can tell who understands current AI and who doesn't in this thread. It is absolutely going to be used, by people, in the pinball business.

This, 100%. The latest models like GPT-4 and Claude are just absolutely fantastic and the next iteration of the models are going to be even better. My efficiency is off the charts. It's like someone connected my brain directly to my dev environments.

There were many times where I would go through a review of a dev's code, and ask them where the code snippet they were using came from, why it was so inefficient, and exactly why they did it that way. "I copy/pasted it from Stack Overflow" was the too often response, and they clearly had no idea how it worked or where the potential pitfalls were.

If I were still in the position of managing a team, I would worry the same thing, only more of it will happen with these new tools. A blow torch can be a really effective tool when you need it, but you wouldn't hand it to a toddler.

In fairness, I'm pretty sure a human made whirlwind 2.0 and I KNOW AI could have done better that stuff.

As usual, the people that know the least talk the most.

That's not saying a lot. Besides repair, how else do you see it being used? Writing rules for games? Designing games? Art? Personally, I like games featuring specific designers, programmers and artists. Guys like Steve Ritchie, Pat Lawlor, Lyman Sheets and Keith Johnson. Steve is the king of flow, Pat's style keeps you coming back for more, Lyman's rules are as close to perfection as you can get, Keith is the master of deep rules. You okay losing all that? Only thing that matters to you is a fun end result?

I did see a fantastic use for AI recently. The Impossible Statue, designed by AI. Best unintentional pinball trophy ever.

Screenshot_20230528-161245~2 (resized).png

I don't think you need a LARGE language model to help pinball players. A small one would work too. We only shuffle through a limited vocabulary of fuck, shits, and god damnits.

There are going to be great places to leverage AI, and others that are not so satisfying. Its human nature to weigh the value of everything. (I could use EXCEL to help me create a grocery list too, but I usually use a scrap of paper. EXCEL just seems unnecessary). In this generation of AI for most of us, the human element is always *implied*. It's a time saver to be sure, but not a replacement for human innovation. Now other commercial applications are already using AI to "drive the car" which is full on AI. That is another whole level of AI discussions, and various applications of it. AI for these all-in applications may also have philosophical and legal concerns as well since human safety is involved. But there are some great uses of AI today....

Take for example the hobbyist that wants to get started with LED strips and a Raspberry Pi. Maybe looking for the means to drive an LED pixel strip with patterns. I entered the command: "write some code in micropython that will allow various patterns to be displayed on a ws2812b led strip connected to pin 16 on a raspberry pi pico" and I posted below the results. Is this the END every hobbiest was looking for? No, but it sure gave a head-start and gets the strip doing something. From here, the rest of the coding is up to the hobbyist , who surely appreciates the start.

import time
import neopixel

# Set the number of LEDs in the strip.
num_leds = 10

# Create a NeoPixel object on pin 16.
strip = neopixel.NeoPixel(board.GP16, num_leds)

# Define a function to set the color of an LED.
def set_pixel_color(pixel, red, green, blue):
strip[pixel] = (red, green, blue)

# Define a function to cycle through all the colors in the rainbow.
def rainbow():
for i in range(num_leds):
hue = (i * 360) / num_leds
rgb = neopixel.Color(hue, 1, 1)
set_pixel_color(i, rgb.red, rgb.green, rgb.blue)

# Define a function to flash all the LEDs red.
def flash_red():
for i in range(3):
for pixel in range(num_leds):
set_pixel_color(pixel, 255, 0, 0)
time.sleep(1)
for pixel in range(num_leds):
set_pixel_color(pixel, 0, 0, 0)
time.sleep(1)

# Define a function to blink all the LEDs green.
def blink_green():
for i in range(3):
for pixel in range(num_leds):
set_pixel_color(pixel, 0, 255, 0)
time.sleep(1)
for pixel in range(num_leds):
set_pixel_color(pixel, 0, 0, 0)
time.sleep(1)

# Define a function to fade all the LEDs from red to blue.
def fade_red_to_blue():
for i in range(256):
for pixel in range(num_leds):
set_pixel_color(pixel, i, 0, 255 - i)
time.sleep(0.01)

# Define a function to cycle through all the patterns.
def cycle_patterns():
while True:
rainbow()
time.sleep(2)
flash_red()
time.sleep(2)
blink_green()
time.sleep(2)
fade_red_to_blue()
time.sleep(2)

# Start the cycle patterns function.
cycle_patterns()

Video title: AI is about to change the world forever - And it gives me a headache

This is an overly simplistic example, but still…

The other day, I asked Microsoft / Bing's AI if it was a good day for fishing. It gave a lot of good info on local fishing, ending with the weather forecast high temp of 74F for a pleasant day. It was 99F at the time.

I've been playing around with GPT a lot for the past couple of months. One of the things I've been testing with it is coming up with a pinball skills training plan, my idea being, give me a list of skills with specific exercises to work on to improve those skills with a machine I give it.

In my experience, it works the best if YOU tell it what skill you want to work on. Like, "I own a Deadpool Pro, give me five repeatable combos I can drill that will let me practice live catch at the end." They usually need to be tweaked a bit (sometimes they don't actually feed to where GPT claims), but it usually gives pretty good results, as long as your game was made prior to its training cutoff in 2021 (and it works best on spike2 sterns, since that's the data that's most available online).

Trying to have it give you a specific plan to build your skills is harder, especially when you don't list out those skills, since there aren't very many templates online to base it on. If you ask GPT to "give me an intermediate training plan to cut down my 5k times", it'll nail it out the door. With pinball, it takes more collaboration, telling it to focus on certain things, not others, until you get something that works.

Also, you can't ask GPT for examples or youtube videos. I was curious if it would give me things like AbeFlips videos, but nope, it just gives me videos to search for that don't actually exist, even if I try using browsing mode.

All in all, it's a great tool. It doesn't give you a finished product, but it IS pretty good at giving you rough drafts you can work with, sprinkled with some pretty good ideas you didn't think of yourself.

At this point, AI doesn't have clear boundaries for when it can 'hallucinate' and when it has to stick strictly to the facts.

Also, so far it hasn't got a large enough 'reputation' understanding.

I trust the pinball advice from Ken Layton, from Vid, from Dumbass, from GPE, from a dozen sources here on Pinside. They have developed a reputation for providing information that is far more reliable than even the Wiki (largely because it's more focused on the specific question being asked).

So, when AI has a better understanding of which information comes from a higher reputation source (weighted relative to the most number of sources that say this...), and it understands that some questions need to be weighted less on 'creative' answers (hallucinations) and more to what is factual and certain...

Yeah AI is going to be pretty darned helpful. And it'll be helpful soon!

In the mean time, ask questions here about your repair, and hopefully someone with a TON of experience that isn't generally available to web-page-scraping algorithms will be able to helpfully provide you a repair procedure that gets your machine fixed.

I suppose once the majority of humanity is using AI to answer old questions we will have no way of teaching the AI how to answer new ones.

Actually, it's more than this.

Something unknown is happening with these AI's. It's new. It's interesting. It isn't just 'scrape the web and average the responses to get the average answer'. It's adding in 'hallucinations'. An admixture of invented answers and algorithmically derived answers. (I'm no expert, but I like the word 'hallucinations' to describe the 'creative' part of the answers.)

The problem is that for some questions, I like a unique understanding. An essay on St Francis that provides a mixture of facts presented in a new, unique way along with a certain amount of opinion (not merely regurgitating what others have said) might be just what I'm looking for.

On the other hand, if someone asks what do I do when my General Illumination plug is burnt and browned on a Roadshow pinball, I don't want any 'creativity' at all. I want to be told to replace J120 and J121 pins on the board, and put crimp-and-stuff connectors with trifurcon on the plugs.

Don't underestimate what is happening here. It's not a definable process with simple 'if-then' logic anymore. Decisions are being made with weights that the human programmers don't have the ability to understand. It's only going to get more complex from here.

Yeah, kinda. Just as easy as you can ask GPT to write code, you can then tell it, you are doing pair programming and your partner just developed this code. Debug it. <feeds GPT the code it just generated>

Is the result good? Original? Entertaining? I don't know.

But to the people saying GPT can't create a new pinball game (playfield layout, ruleset, software) without engineers and developers, you're wrong. And we all know, when businesses figure out they can do something without paying people, their first move is to do that thing without paying people.

Seriously, this may be just the latest tech fad, but it is a fad very different from the cloud or blockchain. There's some real Philip K Dick potential here.

No offense, but this isn't correct. Ultimately, the thing I think everyone needs to understand is that EVERY AI answer is invented and algorithmically generated. ChatGPT doesn't function like a human, it functions like the autocomplete option on your phone. The thing about it is, it's got so much training data and processing power behind it, it's also often correct. However, like your phone's autocomplete, it doesn't know anything about the stuff that it's providing you, it's just built a framework of what a "correct" answer looks like by analyzing the stuff posted online, and is giving you something that fits that criteria, but it doesn't actually know if it's right or not, or even has the ability to judge that.

For your specific example, if there were enough repair forum posts and guides posted online about fixing Roadshow pins, there's a pretty good chance it'd do a good job at giving you your specific answer. It has the data needed to make you happy with that autocomplete. But if it hasn't seen that data, then you'll probably get an answer that doesn't specifically do what you need it to do.

It's complicated, but it's not magic. People have been building this technology since the 70s and know what's going on, it's just that we only recently reached a point where we had the processing power to build a usable solution with it. When I was a grad student in the 00s, I participated in a lab that did computational linguistics, and we were addressing these types of questions then, and I know it goes all the way back to the MIT AI lab.

Google updated their chatGPT competitor, Bard, with their latest model codenamed Gemini. The big update was "multimodal" support for more than just text. If you point it at a machine's user manual, it does a decent job of answering basic questions like parts, common problems, etc