A future that includes ChatGBT

I have been a hobbyist programmer since 1996. In that time, I've learned a ton from YouTube and various tutorials scattered over the internet. I don't have any formal training. As I get older, I feel the need more and more to pass all the knowledge I have accumulated over the last 27 years down to the next generation that will rise in my place. After all, everything I have learned, I learned from people who were willing to share their knowledge for free. It just feels like sharing completes the circle and will allow me to rest in peace after I am gone.

My sister-in-law's boyfriend (he was maybe 19-20 at the time) and I had a conversation shortly after they first met a couple years ago about game development. He was just getting his feet wet with Python. I suggested he try Unity but he was dead set on learning Python. To each their own.

I recently learned that this fledgling programmer has given up his pursuit of learning Python and instead just tells ChatGPT what he wants and ChatGPT poops out code that is already debugged. By the time my educational project is finished, there will only be a handful of true programmers left in the world and they will all be at least in their fifties.

So I put it to the Unity community: How do we encourage young people to spend years developing and honing their skills in this fascinating area of computer science when they can just type a couple sentences into ChatGPT and produce an entire game without any knowledge of programming? Am I just wasting my time with this tutorial series? Do young people even have an interest in learning anymore or are they just looking for the quickest, easiest route to riches and fame?

 

you've invented a fantasy surrounding chatgpt. It doesn't work like that.

"True" programmers use it to save time either with busy-work, or research new ideas, or act like a rubber duck on steroids. Still have to problem solve and use logic part of the brain.

and yes, everybody is looking for the easiest route in whatever they do. Thats what computers were built for. the prize for doing things in a slower or harder way than necessary is that somebody who did it faster or cheaper beats you. people are making throw-away entertainment, not the roman aqueducts.

 

Lets say for the sake of argument that ChatGPT really could write an entire game, then why should we encourage young people to waste their time?

How much assembly language do you know? Back in the 80s there'd be plenty of people who'd tell you that python is not "real" programming. Why do young people not want to learn anymore? Instead of machine-level programming they're just taking short-cuts like learning python and Unity.

 

ChatGPT is not intelligent, it spits out words that have a high probability of making sense to a human based on content humans have produced in the past. This means it can't work with things that are not part of its training dataset such as cutting edge technology. It also makes stuff up quite often so you still need to know how code to debug ChatGPT results. You also need to wire the code it gives you into software that actually runs it such as Unity so you need to be knowledgeable in that as well.

Paid ChatGPT can write code for a simple version of Pong in single prompt. Its memory is too limited for anything more complex so for the tool to be useful in game development scenario you need to know how to code so that you're able to ask very specific, technical questions across many new prompts due to memory limits. We're still very far from the fabled "Make Game" button.

And even when GPT4 32k API becomes available, that's only 4 times more than what's currently available and still wont be enough for anything more complex than simple hyper casual games.

For now this tech only serves as a speed boost in some scenarios for professional game developers. It's also useful for learning to program. I've seen some posts from translators and writers that are losing work already due to ChatGPT. It remains to be seen if it will happen for programmers as well. I presume, less creative industries will be affected to a higher degree than game development since automation there is easier to achieve.

It's also unknown how far this tech will go - technology plateau is a thing so unless singularity happens, perhaps this is as disruptive as it gets.

 

Why do you think you need to encourage anyone more now than in the past?
Personally ChatGPT would have been motivation on its own because I would not have had to ask on StackOverflow or google for solutions this much if I had had such a good explainer with infinite patience while studying.

ChatGPT feels very smart but never ever allows it e.g. a manager who studied business administration to develop a game or any software on their own. They still need us software devs and that will not change.

Similar to how AI art also doesn't replace artists fully. Software can only do so much. The last 5-10% will always be missing and require somebody with in-depth knowledge.
It's a tool.

That said, I personally don't encourage studying something as narrow as games dev. It's a job every gamer and their dog dream of pursuing and thus the studios have free reign over employees. Other fields of software dev is just more lucrative.

 

IMO, we don't. There's no point if they don't want to learn.

Whoever is truly interested in computers will find the way, meanwhile toys made by ChatGPT will not scale past certain point and will require human input.

Let people type sentences, as they'll be the ones paying premium to human programmers later.

 

Exactly. As long as scripting is useful people will want to learn it and they will, whether we encourage them or not. If there's ever a point that it's not useful, then they won't learn it and it won't matter that they don't.

 

Alot of good points made here, tech is advancing very fast, but even new tech has limits. When it really comes down to it... it pays to know scripting base functions, and not rely on a prog to do it for you. Those that dont will be completely lost...

 

This looks like neural-network generated text.

 

I felt this way when I first learned the WWW Consortium introduced the image tag to the HTML codebase, THIS INTERNET THING, WILL CHANGE THE WORLD. This is bigger, so much bigger!..The near future? LangChain! Learn it, use it, be a part of it now, Becoming a logic orientated AI translator, 'Prompt Programmer' and 'Agent Developer' are looking at up to $300,000 a year.

Every single small, medium and large business will have a logic coordinator who interfaces and creates custom prompts, agents and fine-tuning tasks improving every part of business workflow, There is a large middleware space, ready for developers to push this tech to the next level,

How to create and maintain an internal secure LLM? What logic is needed to make use of that data set? How about a Simple Agent?; Find me a recipe for dinner, create a grocery list, best price based on coupons, order online and pay and have it delivered, and send a text that it has arrived.

 

So, you're saying there will be another dot com bubble followed by a crash, centered around Language Models. Well, this is certainly a possibility.

Local LLMs are pretty much available. There's Vicuna, WizardLM, and ton of others. Apparently someone came up with a way to quickly finetune quantized models and called that QLora. So you can finetune something almost equal to chatgpt in hours, although you'd still need 48gb gpu for it if you want 65 billion parameter model and not 7. .

 

For all the whiz bang hype remember..these LLM GPT based AI's are just stochastic parrots. They basically utter the best statistical find which is why their "lies" sound plausible. The ai can only contextualize by the words in cojunct to each other. Until the AI can understand what words actually mean/reference and their particular dynamical representation in the 3D Laboratory Universe they will be stuck in stochastic parrot mode.

 

*sighs*

You might as well start a cult at this point. I mean, the only thing that is missing from this statement are a couple of words about "enlightenment". I do hope you snap out of it, eventually.

Your video is filled with clickbait on its title page. That alone gives a reason not to watch it.

Additionally OpenAI is a walled garden, out of which I'm personally locked out unless I jump over t he fence or dig under it. So if you're looking for an AI future, OpenAI is not the answer and neither is ChatGPT, which is being continuously dumbed down since its initial release.

 

We're not close at all but then we don't need to be close. We just need idiots to hook up the existing bots to systems they have no business being hooked up to and wait for the wrong hallucination.

 

When ChatGPT hype started I had a thought experiment.

Imagine a robot with sensors, cameras.
Now, each five seconds it reports sensor readings via text. (hooking cameras into classifiers shouldn't be hard).
You can submit basic commands to it via text. Move forward, turn left, etc.

Now, the fun part is that you an plug an LLM here. Explain it what needs to be done, run it every five second feeding it directives along with current reading, and have it produce commands for the robot.

That will work. And you'll have an autonomous agent. You can even implement a semblance of memory, have the LLM produce a piece of text that will be fed back to it as a prompt during next update cycle.

This is all fun and games, but there's one nagging thought I can't quite get rid of. For some reason I'm absolutely sure, that at some point this scheme will be definitely used to drive a killer bot, robot turret or a fully armed nuclear sub. And then this thing will hallucinate.

 

I come at this from the perspective of un-boxing the first IBM PC Jr, with DOS. Before windows even existed, I worked on PageMaker 1.0, Photoshop 1.0, and from long term experience in fortune 500 companies as a Data Analyst, I certainly ain't no fan boy of anything, but when I recognize potential, its from decades of history, not from social media.

I admit my journey is just beginning and I feel late, but just from an initial research into the potential, I am both impressed and frightened, I am no longer invested in the sunk cost fallacy of manual repetitive tasks.

I see the potential, of what it can be, not what it is now, Far from being my personal opinion, the facts of the environment, speaks for itself. It is coming, like or not. Not you, me or any government can stop it now,.

Prepare yourselves.

 

I come from the perspective of learning to program on a programmable calculator which did not have DOS on it and then training on ZX-spectrum which used audio tapes and had no floppy drive.

I spent several months with ChatGPT, before the hype started, and saw how OpenAI progressively nerfed it.. Sadly I didn't get to play with the version 1.0, which was the most powerful one.

You, right now, are in a honeymoon phase, riding the hype train.

Save this thread for later, revisit it in 6 months, then 1 year later, then 5. See what actually happens.

 

Huh? By this definition we are entirely there. The current 8 billion humans could not be sustained without digital technology. There already is no way back without substantial losses and technology keeps growing.
By Singularity I rather understand that technology develops itself somehow. That I do not see all that close since the last few percent to full autonomy is the by far hardest.


However the impact for people definitely will be there soon. I can see a future where every customer service is powered by some ChatGPT-like system and many people use such a system of their own to communicate with those.
Want to cancel some subscription? Tell your personal assistant to talk it out with the AI of the subscription company.
Want to order food delivery? Tell your AI which already knows what you love, to talk to the restaurants AI. No more stiff, proprietary APIs (which companies use to lock you into their golden cage or force you to use their ad-loaded app)! Only human language. Possibly enhanced by some robotic standard but with regular text as a fallback.

Those are certainly interesting times.

Sure, something can always go wrong, but humans make mistakes as well and do not have the patience of AI.

 

"the computing power used by machine learning algorithms since 2012 has been growing seven times faster than Moore’s Law."

 

I highly advise to play spinnortality on steam.

Because the next thing would be "Thinking about a gift? Don't worry, the assistant already bought them on your behalf, using your credit card"

There's no singularity, though. Because one thing people forget about, is that if you look outside the window, there's no ChatGPT there. Most people will be completley unaffected, whiel the "bleeding edge AI crowd" will think the world matches what they're imagining.

And regarding ChatGPT the plausible scenario is that it will follow fate of Dall-E. A competitor will spawn along with opensource model and takes over.

 

Computation power is not everything. It is important, but consciousness or actually perfect results (and technology aka programming needs perfect results) do not happen out of nowhere with computation alone.

 

I will, actually I am in the Python scripting and LangChain development phase. Working on real world solutions, and fine tuning results. ..For small token transactions, its all free. So free training, free development environment, free tools, free plugins..why not? Don't believe me or anyone, see for yourself, try it, invest a little time and explore what's out there and how it can work for you.

Example: My ISP wants extra monthly cash for a detailed web log report, I just shove that raw data into my system, not only do I get those same reports, but more [It has created the Python scripts that pull and process that data, you no longer have to use AI], for free, locally and to not only to my web site..for all 12 domains, at the same time. Anytime I want or scheduled. - From one single Prompt.

 

Because that's OpenAI and ChatGPT, which locks out a lot of people out of the service.

For things like "Find me a recipe for dinner".. I'm think that's what Prolog was designed for, 51 years ago. Given a database of conditions, it was designed to arrive at a solution for a query.

> "Becoming a logic orientated AI translator, 'Prompt Programmer' and 'Agent Developer' are looking at up to $300,000 a year."

You cannot be paid a lot for a job anyone can learn quickly. neither of those will require a lot of time to learn, meaning anyone will be able to do it, so it will turn into a minimal wage job. What's more both jobs can go extinct in months once the NNs improve. There's already no point in designing prompts for both Stable Diffusion and LLMs, because you can use smarter "brain" instead.

That's why I brought up dot com bubble. There will be a gold rush, everybody will think they'll become a rock star, then t her will be too many rock stars and it will come crashing down.

Meanwhile people will be quietly running finetuned MPT-7B on their toaster and discussing coding with it, so an expensive mainfraime-style saas like ChatGPT will simply become unnecessary.

At least that's how I see it.

 

"anyone will be able to do it, so it will turn into a minimal wage job. What's more both jobs can go extinct in months"-

Funny, that's exactly what I thought when heard the role of 'Social Media Manager' and 'Influencer', as an actual job!

Oh, how wrong I was.

"Prompt engineering is now considered one of the hottest tech jobs as companies look for ways to help train and adapt AI tools to get the most out of new large language models, which can provide results that are not always correct or appropriate." - https://time.com/6272103/ai-prompt-engineer-job/

 

I posit a scenario. A 5 year old boy with a bucket of seawater and a squirt gun vs the server bank of the "singularity". Natural intelligence will drown and short circuit that "singularity".

It is not a ghost in a machine. It's intelligence cannot leap out of the circuits and control anything. It is a fancy self-correcting abacus comprising data massaged into numerics, stats, conditional logic and lots of for loops. Most of the jargon is hand waving for basic gameplay tracking, finding means, maxes, massaging number ranges to extract a better signal, transforming and comparing vectors to derive a tensor which in NLP is basically pointing back to words as nGrams of various window sizes and pre-categorized on an indexed data layer. It still doesn't have a clue what words mean to a human brain and its hosting organism and the multiple ways we can process data in a split second. When a computer processes words it does not "say" them and comprehend the audio, nor does it link to a series of archetypal or specific visual icons or animated sequences that we use imaginatively to dynamically process our relationship to the data, our current environment and any remote environments or related by personal tangents objects or events.

 

Influencers are a good illustration of t he "rock star" mentality I spoke of. Few people are successful, the rest is dreaming of catching the blue bird of happiness and money.

Influencers also illustrate SaaS part well, as an "Influencer" is one dependent on underlying platform. Once it changes the rules or stop paying, they're screwed. This happened several times with twitch, patreon, and so on. This is incredibly similar to dependence on proprietary solution you can't even run locally.

The problem here is that the article you linked is not a neutral and is trying to ride the hype train and earn money in the process. The headline falls into typical clickbait along the lines of "Here's how you could get rich quick easily and without special skills". This kind of thing always sells, but the amount of rich people doesn't seem to be quickly increasing.

I'd recommend to recall blog craze, cryptocurrency popularity, NFTs and so on. it is all the same. Another bubble, saying "the next rich person could be you!" and then bursting.

 

The models will only be as truthful as the prompt engineer is and as informed as the data it was trained with. Political and personal bias will be built in for better or worse. It is great for trinket selling. But a much better AI future of intelligent assistants would be Tiny AI that works on an iDevice or standard desktop. No connection needed. The AI becomes a savant in it's particular knowledge domain. If you are interested in European history then your language model should have no need of feminist Micronesian basket weaving, some knowledge of world geography, a model of an historical timeline and possibly translational features. It should be trained on a corpus of factual materials matching the specific domain and considered reputable amongst that scholarly or ad hoc community as the case may be. It should have the ability for the end user to input further texts/data for ongoing extension of its specific knowledge domain.

 

The big round of tech layoffs targeted censor teams..whatever fancy cute name they had..that is what they were.. Social media influencers are big in the news lately. They have lost companies 10's of billions of dollars. A Lawyer in NYC submitted a case that cites law based on bogus cases that never were written by ChatGPT. He is in hot water. Social Media Managers are akin to Talent Managers/Secretaries. It is just a shift of media and savvy within that construct that made those managers successful clients where publicized.

If yer job is rote and verse dogmatic text churning you may be worried by the advent of AI. Stringing buzzwords into a sing song of drivel is simple for an NLP AI. If your work requires hands on manipulation of hardware in varied environments or uses intuitive creativity to engineer constructs then AI may be of assistance in much the same manner a handbook or repair manual is but it will not usurp that job. If the AI runs a manual labor job then someone has to replace worn out and broken parts, upgrade logic boards or install software updates and there will need to be ongoing monitoring. An AI robot that can do construction can kill somebody quickly on a jobsite. They had AI warrior bots pitted against some marines. The Marines used Warner Bros tactics to sneak up on it. They hid in a box or behind a bush and moved in on it. It had no cognition of them as it has no inherent perspective.

 

Don't underestimate that effort too much.
With AI art for example we are already beyond the point where you can learn to get top notch results (that are possible with AI art) within a few days.

 

Training time for Stable Diffusion is in ballpark of 1 month, which is a joke, and past that point you hit diminishing returns.

However, since the release "prompt engineering" became significantly less important.

The issue with prompt engineering is that rules of prompt engineering change along with neural network brain. You swap the network, and prompts behave differently. Hence the skepticism about "prompt engineer". Locally-tuned models are pretty much on the horizon.

 

Hasn't been about games since the necro. Done.