I think no, no ai winter.

I see it white and black (yes, no shades) mostly because of the hype cycle.

Bitcoin.. everyone shouted like hell.. no one needed a solution because no problem. Hype cycle did apply fully.

LLM and AI .. less shouting and everyone needs and wants, cause it solves more and More of our daily problems. Investment is now high already , but the prospect and future usages is more worth. No winter there fore

If there is one, I think it will be many years away. With ChatGPT/GPT-4/llama2/Gemini and Stable Diffusion/DALL·E/Midjourney available, and each getting more powerful and convincing as time goes on, I feel like the "cat is out of the bag". There's tremendous opportunity in these technologies and they'll find their way into more and more niches. I think the drive to improve them isn't going to taper off any time soon, especially if some talented engineers have the AGI sparkle in their eyes.

No, we have been using AI to improve things and make profits since before the LLM hype cycle. LLMs just unlock more capabilities. There is ROI for businesses now and consumers are also paying monthly subscriptions.

What we might see is a plateau of capabilities/ ROI, but I imagine we'll have to go through the LLM version of RL, like we did with DNNs and DeepRL. Still, while the capabilities level they will also expand and see more applications. I think we are well on our way to pervasive robotics, or the end of physical labor holdups

hello,

yes.

why? everything in the world is bound to "trends", "hypes" or "fashions" ... and stuff tends to go out of fashion after a while of not fulfilling exaggerated expectations.

especially in IT - and especially AI ... we had that several times (!) since its inception in the 50ties of the last century :)

just look at its history: even neural networks are nothing new ... heavily simplified: they where with us after the logic-programming hype of the 70ties during the 80ties, displaced by fuzzy-logic around the turn of the decade to the 90ties. the logic-programming approach got reboiled by semantic-technologies in the zero-decade, replaced by machine-learning - applied statistics - and deep-learning - again neuronal networks - in the 10ths and up until now ... thanks to cheap computational resources - pumped up to billions of parameters etc.

don't get me wrong: great stuff, but less novel than some of us might think ;)

notice: i wrote nothing about when this will happen, but its more or less a tautology, that there will be such a downward development of the current trend sometimes in the future ... :)

idk ... maybe after people realize, that current pre-trained transformers/language models - as nice/useful/funny as they are - are not "(hard) AI" ... i think its now called "general AI" or AGI!?

cheersv

Yes - this is how ai has gone on for years ups and downs.

The trouble these days if I’m honest is either people are laser focused on the latest new toy (llm this year) or too new to the field that they don’t have a grounded understanding in what they are doing (ie they will all be trying to remix new models that have actually already been tried)

All of this while AI is still gate kept by companies with huge amounts of compute.

I will say that each new ai summer has gotten more and more dramatic though.