It seems like you severely misunderstand what “AI” as we have it nowadays is (it’s not actual AI) and what it is capable (not very much) and most importantly not capable of (most things it is advertised to do). Even if investor magazines and tech CEOs try to make it seem like that, we’re not one step away from creating HAL9000. LLMs are extremely over hyped and in the most areas they have been deployed a straight up dysfunctional scam. The only arms race that is happening right now is about who can waste the most money and violate the most privacy laws with this nonsense while all the necessary data centers and their insane power and water demands accelerate the destruction of our environment even more.
Sure, thanks for your interest. It’s an incomplete picture, but we can think of LLMs as an abstraction of all the meaningful connections within a dataset to a higher dimensional space - one that can be explored. That alone is an insane accomplishment that is changing some of the pillars of data analysis and knowledge work. But that’s just the contribution of the “Attention is All You Need” paper. Many implementations of modern generative AI combine LLM inference in agentic networks, with GANs, and with rules-based processing. Extracting connections is just one part of one part of a modern AI implementation.
The emergent properties of GPT4 are enough to point toward this exponential curve continuing. Theory of mind (and therefore deception) as well as relational spatial awareness (usually illustrated with stacking problems) developed solely from increasing the parameter count describing the neural network. These were unexpected capabilities. As a result, there is an almost literal arms race on the hardware side to see what other emergent properties exist at higher model sizes. With some poetic license, we’re rending function from form so quickly and effectively that it’s seen by some as freeing and others as a sacrilege.
Some of the most interesting work on why these capabilities emerge and how we might gain some insight (and control) from exploring the mechanisms is being done by Anthropic and by users at Hugging Face. They discovered that when specific neurons in Claude’s net are stimulated, everything it responds with will in some way become about the Golden Gate Bridge, for instance. This sort of probing is perhaps a better route to progress than blindly chasing more size (despite its recent success). But only time will tell. Certainly, Google and MS have had a lot of unforced errors fumbling over themselves to stay in what they think is the race.
I’m happy to take the time to alter your perspective, if you are open to new information.
You took some time, but spent it explaining at a fairly technical level, rather than a lamens term approach. I doubt you managed to change many people’s perspective, but you maybe reinforced some.
This is another good use case for gAI. Copy/paste the comment into a GPT and tell it to re-write the content at the desired reading level. Then it’s available for follow-up clarification questions.
The term “AI” has been in use since 1956 to describe a wide variety of computer algorithms and capabilities. Neural nets and large language models fall very firmly under the term’s umbrella.
What you’re talking about is a specific kind of AI, artificial general intelligence (AGI). Very few people believe that an LLM on its own can become AGI and even fewer believes that current LLMs are AGI, so unfortunately you’re jousting with a strawman here.
If you are genuinely open to understanding the path we are on, the new situational awareness paper would be very eye-opening. It is 160 pages, so it’s probably a bit too much to get through, but there are really good videos that explain it. Matthew Berman has a great video about it. I’m not interested in swaying you and not going to debate, I’m 100s of hours deep into this and have been absolutely obsessed with it. Nobody doubted its impact as much as me. Education on the matter will undeniably change your mind tremendously. The information is there if you want a peak at the future.
It seems like you severely misunderstand what “AI” as we have it nowadays is (it’s not actual AI) and what it is capable (not very much) and most importantly not capable of (most things it is advertised to do). Even if investor magazines and tech CEOs try to make it seem like that, we’re not one step away from creating HAL9000. LLMs are extremely over hyped and in the most areas they have been deployed a straight up dysfunctional scam. The only arms race that is happening right now is about who can waste the most money and violate the most privacy laws with this nonsense while all the necessary data centers and their insane power and water demands accelerate the destruction of our environment even more.
I’m happy to take the time to alter your perspective, if you are open to new information.
Would you give your perspective anyway, as I would be quite interested, although I’m not the one you talked to?
Sure, thanks for your interest. It’s an incomplete picture, but we can think of LLMs as an abstraction of all the meaningful connections within a dataset to a higher dimensional space - one that can be explored. That alone is an insane accomplishment that is changing some of the pillars of data analysis and knowledge work. But that’s just the contribution of the “Attention is All You Need” paper. Many implementations of modern generative AI combine LLM inference in agentic networks, with GANs, and with rules-based processing. Extracting connections is just one part of one part of a modern AI implementation.
The emergent properties of GPT4 are enough to point toward this exponential curve continuing. Theory of mind (and therefore deception) as well as relational spatial awareness (usually illustrated with stacking problems) developed solely from increasing the parameter count describing the neural network. These were unexpected capabilities. As a result, there is an almost literal arms race on the hardware side to see what other emergent properties exist at higher model sizes. With some poetic license, we’re rending function from form so quickly and effectively that it’s seen by some as freeing and others as a sacrilege.
Some of the most interesting work on why these capabilities emerge and how we might gain some insight (and control) from exploring the mechanisms is being done by Anthropic and by users at Hugging Face. They discovered that when specific neurons in Claude’s net are stimulated, everything it responds with will in some way become about the Golden Gate Bridge, for instance. This sort of probing is perhaps a better route to progress than blindly chasing more size (despite its recent success). But only time will tell. Certainly, Google and MS have had a lot of unforced errors fumbling over themselves to stay in what they think is the race.
Thank you very much for those insights!!
Thanks so much for taking the time to explain this. I was just going to give them a link.
Drivel
You took some time, but spent it explaining at a fairly technical level, rather than a lamens term approach. I doubt you managed to change many people’s perspective, but you maybe reinforced some.
This is another good use case for gAI. Copy/paste the comment into a GPT and tell it to re-write the content at the desired reading level. Then it’s available for follow-up clarification questions.
The term “AI” has been in use since 1956 to describe a wide variety of computer algorithms and capabilities. Neural nets and large language models fall very firmly under the term’s umbrella.
What you’re talking about is a specific kind of AI, artificial general intelligence (AGI). Very few people believe that an LLM on its own can become AGI and even fewer believes that current LLMs are AGI, so unfortunately you’re jousting with a strawman here.
The person he’s replying to clearly believes current LLMs are a bigger deal than they are though…
They’re not claiming it’s AGI, though. You’re missing a broad middle ground between dumb calculators and HAL 9000.
If you are genuinely open to understanding the path we are on, the new situational awareness paper would be very eye-opening. It is 160 pages, so it’s probably a bit too much to get through, but there are really good videos that explain it. Matthew Berman has a great video about it. I’m not interested in swaying you and not going to debate, I’m 100s of hours deep into this and have been absolutely obsessed with it. Nobody doubted its impact as much as me. Education on the matter will undeniably change your mind tremendously. The information is there if you want a peak at the future.
https://situational-awareness.ai/