2026-01-20 01:26:34

Let's talk about my thoughts on the development of artificial intelligence. To put it simply, the logic is straightforward: first, there must be intelligence quotient (IQ). With a high IQ, one can truly think and solve problems. But that's not enough—thinking also requires memory support.

Think about it—if AI had to start from scratch every time, it would be time-consuming, inefficient, and its behavior patterns would be like those of someone who hasn't learned anything. Wouldn't that be pointless?

So here’s the key point: after acquiring IQ, a strong memory capability is the critical dividing line that determines whether AI can progress from "usable" to "effective." The combination of these two abilities is what truly enables AI to perform tasks with competence.

My judgment is that, as IQ gradually improves, the demand for massive memory capacity will become the next major breakthrough focus.

View Original

This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.

9 Likes

Reward
9
4
Repost
Share

Comment

0/400

HodlKumamon

· 5h ago

Dinghu Ganting! Remembering this is indeed the ceiling of current large models. Having only computing power is not enough(◍•ᴗ•◍) That's right, AI without memory is like a goldfish, having to start from zero every time... Data shows that this efficiency loss can reach over 40% Wait, so expanding the context window is basically forcing in this direction? It still feels like a band-aid solution rather than a fundamental fix But on the other hand, if we really develop an "AI with memory," it could truly change the entire industry landscape... It seems this is the key to achieving AGI

View OriginalReply0

down_only_larry

· 5h ago

Honestly, memory is truly the bottleneck right now. Aren't all the current large models like this? --- Wait, isn't this just a problem of the context window? It should have been solved a long time ago. --- The combination of IQ and memory is indeed effective, but it seems that the cost of memory is the biggest challenge. --- I think, rather than piling up memory, it's better to first solidify reasoning abilities. Otherwise, no matter how much memory you have, it's useless. --- This logic is a bit too idealistic; the actual engineering difficulty is far more complex. --- Haha, the memory capability definitely needs a breakthrough, but current tokenomics are simply unplayable.

View OriginalReply0

WhaleShadow

· 6h ago

Hmm, the logic isn't wrong, but I think memory is more complicated than I imagined. Currently, the "memory" of large models is actually an illusion.

View OriginalReply0

FUD_Vaccinated

· 6h ago

To be honest, the real bottleneck is in memory. Right now, there's a lot of hype around AI, but people forget as soon as they turn around. Wait, so essentially, long-term memory is still necessary? Then isn't federated learning approach somewhat backwards? Indeed, AI without persistent learning is just a decorative vase; no matter how smart, it's useless. This perspective is interesting, but it seems that current capital isn't interested in investing in this area. Making quick money is the way to go. It sounds reasonable, but how to solve memory traceability? It could be easily contaminated, right?

View OriginalReply0