What will Gen AI do in 2025. How long before AI can self regulate?

[Renegade]

We have seen how rapidly Gen AI swept us off. From us marveling at its unseen magical powers, to using it in our daily lives. With enterprises racing to imbibe use cases after use case to prove to their peers and higher ups how they are all technology savvy. Its everywhere, its reading everything and its improving at a rapid pace.

Where do you see Gen AI going in the next year? What improvements will it bring in? Which new use cases will open up?

Most importantly, how long before it can start to self regulate.

 

[mani0993]

Firstly, I see a reduction in subscription prices due to competition. Microsoft's Github CoPilot free-tier is a step in that direction.
Secondly, we may see it hitting a saturation point. AI everywhere will make people miss the human element.

 

[Renegade]

AI everywhere will make people miss the human element.

Ah yes, once you've seen enough AI generated content it is easy to distinguish it from human generated content. And it has started to turn me off a little.

I am curious to see how long before we start seeing AI generated images that can do crisp text formatting perfectly. That would be a tipping point for a lot of people.

 

[ashayc]

Have you seen the news of how they integrated alignment into the CoT process, especially for o3? It's obviously a trade off with latency, but adding alignment as a step during chain of thought seems to be very effective in their benchmarks against o1.

 

[Renegade]

Have you seen the news of how they integrated alignment into the CoT process, especially for o3? It's obviously a trade off with latency, but adding alignment as a step during chain of thought seems to be very effective in their benchmarks against o1.

With o3 they are looking at enhancing the AI's reasoning capability in the Science, Maths and Coding areas which is foundational to enhancing the AI capability further. By this time next year we could be gaping at the step forward that Gen AI has made, which is unimaginable right now.

I am not sure how this will be used by the masses though.

All this reminds me an episode of Orville, Mad Idolatory, where the speed of evolution of society increases exponentially. If we look at just 20 years in the past and compare it with the last 2 years, the progress is not linear. Hard to imagine what would it be 20 (or even 5) years from now.

All this talk of being careful with AI and Elon's concern are not unfounded. Scientists will not hesitate to explore the unknown, no matter how catastrophic it may be.

 

[ronnie_gogs]

As a Warhammer fan I believe AI would be cause a huge conflict in the future. Probably not in our lifetime but I would love to see a full blown conflict as I am nearing to say goodbye to the world.

Warhammer, Dune, Terminator series all written when computers were relatively primitive but all tell the same story. AI going rogue. We don't even have a 10 feet tall psychic emperor to save us. .

 

[6pack]

I have a different take on this AI boom. No matter how much Ai progresses, the humans are still stupid. And we stupid are around 8.1 billion now in number.
We humans have based our economy on large scale manufacturing and consumption of products. This will run only till people have buying power. AI could help by giving jobless people monetary compensation every month but companies wont like it. What will companies do after laying off all their employees to AI? They will get richer in the short term but then only the rich will get to buy stuff. Companies will start to shut down without large scale buyers. The rich-poor gap will start widening more and more till only a few thousand will be able to survive. Then AI will be removed from society completely except a few sectors like medicine.

 

[cranky]

Where do you see Gen AI going in the next year?

Nowhere.

Gen AI has already started cannibalising itself. Amazon for example is full of AI generated reviews, which other models are using as input data and we've already started seeing huge downgrades in output of both LLMs and image generation as a result (The Coca Cola Christmas ad comes to ind, another was that very famous South Indian ad with wonky digits.

Without good data, the way AI is being used today is headed for a mass extinction event in a couple years. Reasoning logic is simply using expanded and cascading datasets from existing model outputs because there's not much fresh data left. I suggest O3 will launch this year and you'll start getting more of the same. We worried about being able to detect AI-generated content but it's so easy to catch spammers who use AI to attempt to bypass guardrails. AI generated text is laughably easy to detect because they all follow the exact same structure and logic.

Then there's agency, which AI models are not allowed to have. Without agency, AI has no actionables and must await interrogation. All manner of reason-based logic is founded in a looping cycle of observation, thought and action. If models cannot be allowed to exhibit action, they cannot develop thinking. Look back to a few decades ago when computers were beating humans at chess. While there is something to be said of their math prowess, the fact they got to play and learn is the difference between those programs and today's AI.

You could also end up with a situation where your AI model can get your CEO killed.

 

[Renegade]

Yeah just the thing I was wondering yesterday, now that a lot of content is being generated by AI. Its like a poop eating cycle for AI now with less and less of content to illegally leach.

it's so easy to catch spammers who use AI to attempt to bypass guardrails

Could you help me understand this better. Why would spammers attempt to bypass guardrails. Guardrails are there to set the boundaries which the model should not traverse. Spammers just want some ready made dish to pass off as their own. How are spammers connected to guardrails.

The most obvious tell for AI generated content is when its too polite, too polished, too elaborate. Just can't do any of that on a community forum.

 

[lockhrt999]

Where do you see Gen AI going in the next year?

1. There's already a big overlap between AI and coding, designer, copywriter etc. freshers. Most of these jobs will go away.

2. The robotics industry also has AI like big momentum. In 2025, I'm hoping to see the AI + Robotics' brain and muscle combo perfected. It'll go after blue-collar jobs.

3. I think in Robotics, China has already surpassed the US. In AI, the competition is close, even though China has restrictions. But what it means for AI as a whole? When there was a Cold War between US-Russia, we saw technology improved at break neck speed. The same thing will happen now. The US won't want to lose the dominance, and it'll give a free hand to its AI industry (less restrictions-less safety).

4. AI deniers will reduce in numbers. But it'll be non-zero.

The most obvious tell for AI generated content is when its too polite, too polished, too elaborate.

A finely prompt generated content is indistinguishable from a human writer. The characteristics you expressed occur when no efforts were taken at prompting.

It's like VFX. The highest quality VFX is indistinguishable from the real footage.

Just can't do any of that on a community forum.

Is that a challenge, sir?

 

[Renegade]

Is that a challenge, sir?

I think you misunderstood.
Most users will not engineer the prompt.
Most users will not be polite, polished, elaborate on a forum. Meant as a joke in reference to forums reputation of being traditionally curt.

 

[w.ilyas]

I just rewatched the first episode of Futurama in which Bender the robot is introduced. I snickered at the scene where Bender comes to realize he is meant to bend all things. With the new found power he bends everything he can see, its proven useful as Fry(a human) and Bender need to escape a prison. Soon after getting out they find a manhole that can be used to escape the cops. Bender being self assured of his new found power offers to bend the manhole cover since its enforced with bars. While Fry simply opens the cover since its got hinges and has no lock.
This pretty much sums up AI at its current state. It has no common sense, and will steer to wherever we as humans give it direction. Till now AI has been a tool that you can add to your toolbox and that's pretty much it, no matter how much they market it as world changing.

 

[cranky]

Could you help me understand this better.

Oh I meant forum spam guardrails, not wrt the model.