It seems like we actually do not have a good name for the things that AI companies are building, weirdly enough.
LLMs? A hard-to-pronounce acronym for Large Language Models, which they were a couple of years ago, but now kind of aren’t.
Frontier Models? Then how do we talk about models which are not frontier? Also, “models” of what? Also, does this term include AI systems which reach the frontier by a very different paradigm?
Reasoning Models? Better (best on this list?), though it seems to refer specifically to the models for math / coding problems, that use e.g. chain-of-thoughts, not the agents.
Models? I think this is what we are settling on? Seems fine, but again, not a crisp semantic boundary.
LLM Agents? Same problems as “LLMs,” though it at least indicates that there exist later stages of training on top of language modeling.
AIs? Not specific.
AGI / ASI? Disputed, probably not yet, and also not specific about the algorithms / technology.
Prosiac AI Systems? That is descriptive, but awful(ly wordy).
Shoggoths? Makes a specific claim about what’s going on, also seems rather unlikely to catch on.
Transformers? We would probably like a term that also includes e.g. Mamba, also it seems that the current frontier models are sometimes not just one transformer.
Generative AI? This is sounds and feels like a buzzword, and also I think the “generative” part misses the point (it seems like something that will make videos, not take over the world). But this DOES seem like it captures the rough boundaries of the current paradigm pretty well. They are predictive models, but we often use them as engines for generation. Overall still not a fan.
This actually slows down my reasoning or at least writing about the topic, because I have to choose from this inadequate list of options repeatedly, often using different nouns in different places. Seems useful to have a more wieldy handle for the concept I’m trying to point at (an instance of Patrick Henry Winston’s Rumplestiltskin Principle—naming something gives you power over it). I do not have a good suggestion. Any ideas?
General-purpose AI Systems—Unwieldy. Possibly overemphasizes their tool nature.
Digital Minds / Digital Brains—Very accurate in some important ways, allergically disputed in others. Not technical.
Some further shots from the hip:
Broad AI—Not narrow, without claiming full generality. Highly unspecific.
Digital Cognition Engines—Anything “engine” doesn’t acknowledge the system as being whole unto itself. Also this is sci-fi name territory.
Cognition Manifolds—Also sci-fi, but scratches an itch in my brain. I like this one a little too much and I am a little sad now that this isn’t the accepted term.
A really good name choice would work for the annoying people on reddit I sometimes talk to who insist that these things aren’t intelligent. I have interviewed a few and the consensus seems to be that they are powerful, but the thing they do to get that power is not intelligence. I suspect the crowd with this opinion defines intelligence in a way that means it can only be achieved by something I’d call an aligned ai, or perhaps they define it in a way that can only be achieved by an uncontrollable AI, somewhere in that space.
...anyway, how about… deep learning! or large ML? or sequential token generators. depends how specific you want to be. should alphafold count? it’s a transformer. should wavenet count? should sora count? should deepmind gato count?
re: “models of what”—most naively, they’re models of the dataset.
For those who insist these aren’t intelligent I like my extremely general term “Outcome Influencing System (OIS)” pronounced “oh-ee” and defined as “any system composed of capabilities and preferences which uses its capabilities informed by its preferences to influence reality towards outcomes it prefers”. Then it becomes trivially true that these things have capabilities and the capabilities are improving.
It seems like each of those terms does have a reasonable definition which is distinct from all of the other terms in the list:
LLMs: Neural networks trained (usually via autoregressive next-token prediction) on text corpora
Reasoning models: LLMs which have special delimiters marking the start and end of a chain of thought, and post-training to teach it to use those special delimiters
LLM agents: LLMs with access to tools invoked repeatedly until some stopping condition is met (or indefinitely, though we don’t see much of this yet)
Frontier Models: The subset of models trained with the largest compute budgets at any given time
Models: Neural networks trained to minimize some loss function over a dataset
AIs: Systems that perform tasks which required human intelligence last year
AGI/ASI: Systems matching or exceeding human cognitive capabilities across most/all domains respectively, but with the definition of which domains “count” gerrymandered such that no existing system counts as AGI
Prosaic AI Systems: Systems built by scaling up existing deep learning techniques rather than novel architectural insights
etc etc
It seems like you’re hoping for some term which encompasses all of “the AI systems which currently exist today”, “AI systems which can replace humans in all tasks and roles, which the frontier labs explicitly state they are trying to build”, and “the AI systems I expect will exist in the future, which can best be modeled as game-theoretic agents with some arbitrary utility function they are trying to maximize”. If that’s the case, though, I think you really do need multiple terms. I tentatively suggest “current frontier AI agents”, “drop-in-replacement-capable AI” or just “AGI”, and “superhuman wrapper minds” for the three categories respectively.
Is the word “systems” required? “prosaic AI” seems like it’s short enough already, and “prosaic AI alignment” already has an aisafety.info page defining it as
Prosaic AI alignment is an approach to alignment research that assumes that future artificial general intelligence (AGI) will be developed “prosaically” — i.e., without “reveal[ing] any fundamentally new ideas about the nature of intelligence or turn[ing] up any ‘unknown unknowns.’” In other words, it assumes the AI techniques we’re already using are sufficient to produce AGI if scaled far enough
By that definition, “prosaic AI alignment” should be parsed as “(prosaic AI) alignment”, implying that ” “prosaic AI” is already “AI trained and scaffolded using the techniques we are already using”. This definition of “Prosaic AI” seems to match usage elsewhere as well, e.g. Paul Christiano’s 2018 definition of “Prosaic AGI”
It now seems possible that we could build “prosaic” AGI, which can replicate human behavior but doesn’t involve qualitatively new ideas about “how intelligence works:”
It’s plausible that a large neural network can replicate “fast” human cognition, and that by coupling it to simple computational mechanisms — short and long-term memory, attention, etc. — we could obtain a human-level computational architecture.
If that term is good enough for you, maybe you can make a short post explicitly coining the term, and link to that post the first time you use the term each time.
I do note one slight issue with defining “prosaic AI” as “AI created by scaling up already-known techniques”, which is that all techniques to train AI become “prosaic” as soon as those techniques stop being new and shiny.
2025!Prosaic AI, or, if that’s not enough granularity, 2025-12-17!Prosaic AI. It’s even future-proof if there’s a singularity, you can extend it to 2025-12-17T19:23:43.718791198Z!Prosaic AI
To my understanding, Reasoning Model refers to models that have been explicitly fine-tuned to perform chain-of-thought reasoning effectively. Something analogous to DeepSeek’s GRPO. For instance, even if I told GPT-3 to show its work, it would not be a reasoning model.
As for LLM, it’s admittedly on the path to being a holdover term, but not for lack of effort. Some people tried to make VLM (Vision-Language Model) more common, but I’ve mainly heard it used in the context of lightweight models used to graft vision capability onto systems driven by single-modality LLMs.
It’s probably too late now, but maybe what we should have done is:
AGI = artificial general intelligence = an AI system that is generally capable rather than just narrowly capable ASI = AGI that is better than the best humans at everything while also being faster and cheaper
We could then say that these are AGI companies building lots of little AGIs, and that they are on a path to ASI but aren’t there yet and probably won’t be for several more years. Claude is an AGI. GPT5 is an AGI. Etc. They keep improving in capability and generality and one day they’ll be ASIs, but they aren’t yet.
If AGI is human-equivalent for the purposes of developing a civilization, a collective of AGIs is at least as capable as humanity, plus it has AI advantages, so it’s much more capable than a single AGI instance, or any single human. This leads to ASI being often used synonymously with AGI lately (via individual vs. collective conflation). Such use of “ASI” might free up “AGI” for something closer to its original meaning, which didn’t carry the implication of human-equivalence. But this setup leaves the qualitatively-more-capable-than-humanity bucket without a label, that’s important for gesturing at AI danger.
I think the other extreme for meaning of “ASI”, being qualitatively much stronger than humanity, can be made more specific by having “ASI” refer to the level of capabilities that follows software-only singularity (under the assumption that it does advance capabilities a lot). This way, it’s neither literal technological maturity of hitting the limits of physical law, nor merely a collective of jagged-human-level AGI instances wielding their AI advantages. Maybe “RSI” is a more stable label for this, as in Superintelligence Strategy framing where “intelligence recursion” is the central destabilization bogeyman, rather than any given level of capabilities on its own.
What do you think of “Locked In AI (LIAI)” for when an AI becomes sufficiently capable that it’s preferences / utility function are “locked in” and can no longer be altered or avoided by other agents? This “locking in” is how I refer to the theoretical point of an RSI when it becomes too late to stop or alter course.
Also, for what it’s worth, I like “artificial general super intelligence (AGSI) which then frees up “AGI” for AI that does general reasoning and language at any level of capability, and hilariously frees up “ASI” to refer to any AI that does what it does better than any human, so a pocket calculator is an ASI because it does arithmetic better than any human. Though more confusing, LLMs would be ASI and not AGI because they are superhuman at text prediction, but chatbots made from LLMs would be AGI and not ASI because they reason and talk with general intelligence, but it seems more limited in some ways than human reasoning.
Another possibility that appears to be in wider use is neuro-symbolic AI.
An invention of my own is a Generative Programming Interface (GPI), analogous to an API, CLI, or GUI. The idea here is that LLMs or other forms of gen-AI can best be thought of as a crucial component of the program interface in this paradigm. I might have said “AII” (AI Interface), but that renders like the word “All.” AI Interface might work if spelled out, though.
The problem with “neuro-symbolic AI” is that Gary Marcus types use it to refer to something distinct from the current paradigm. Even though it is ironically a pretty good description of the current paradigm.
What about “neuro-scaffold AI?” Marcus is loading the “symbolic” part of “neuro-symbolic” with the connotation of “definitionally NOT the current paradigm.” The term “scaffold” gets used to refer to traditional programs that wrap and attempt to refine and be controlled by AI, and is encompassing enough to include reasoning or agentic capabilities without assuming them.
GPTs. Yes, it’s still an initialism like “LLM” but it’s much easier to pronounce (“jee-pee-tee”) and you can call them “jepeats” (rhyming with “repeats”) if you want.
How about “IAC”: “Interactive Associative Composer”? - “Interactive” as indicating there is a complex enough response to human inputs to feel organic, yet without the degree of self-motivation terms stemming from “agency” would connote. - “Associative” as indicating the means by which it works, in a very generic way. - “Composer” as (somewhat redundantly) indicating that it produces responses as the output of its associative process, as well as giving a connotation of quality.
It seems like we actually do not have a good name for the things that AI companies are building, weirdly enough.
LLMs? A hard-to-pronounce acronym for Large Language Models, which they were a couple of years ago, but now kind of aren’t.
Frontier Models? Then how do we talk about models which are not frontier? Also, “models” of what? Also, does this term include AI systems which reach the frontier by a very different paradigm?
Reasoning Models? Better (best on this list?), though it seems to refer specifically to the models for math / coding problems, that use e.g. chain-of-thoughts, not the agents.
Models? I think this is what we are settling on? Seems fine, but again, not a crisp semantic boundary.
LLM Agents? Same problems as “LLMs,” though it at least indicates that there exist later stages of training on top of language modeling.
AIs? Not specific.
AGI / ASI? Disputed, probably not yet, and also not specific about the algorithms / technology.
Prosiac AI Systems? That is descriptive, but awful(ly wordy).
Shoggoths? Makes a specific claim about what’s going on, also seems rather unlikely to catch on.
Transformers? We would probably like a term that also includes e.g. Mamba, also it seems that the current frontier models are sometimes not just one transformer.
Generative AI? This is sounds and feels like a buzzword, and also I think the “generative” part misses the point (it seems like something that will make videos, not take over the world). But this DOES seem like it captures the rough boundaries of the current paradigm pretty well. They are predictive models, but we often use them as engines for generation. Overall still not a fan.
This actually slows down my reasoning or at least writing about the topic, because I have to choose from this inadequate list of options repeatedly, often using different nouns in different places. Seems useful to have a more wieldy handle for the concept I’m trying to point at (an instance of Patrick Henry Winston’s Rumplestiltskin Principle—naming something gives you power over it). I do not have a good suggestion. Any ideas?
Adding some descriptors I have frequently used:
General-purpose AI Systems—Unwieldy. Possibly overemphasizes their tool nature.
Digital Minds / Digital Brains—Very accurate in some important ways, allergically disputed in others. Not technical.
Some further shots from the hip:
Broad AI—Not narrow, without claiming full generality. Highly unspecific.
Digital Cognition Engines—Anything “engine” doesn’t acknowledge the system as being whole unto itself. Also this is sci-fi name territory.
Cognition Manifolds—Also sci-fi, but scratches an itch in my brain. I like this one a little too much and I am a little sad now that this isn’t the accepted term.
I say “gippity” meaning “generative pretrained transformer” which IIUC is still true and descriptive for most of this, except Mamba.
A really good name choice would work for the annoying people on reddit I sometimes talk to who insist that these things aren’t intelligent. I have interviewed a few and the consensus seems to be that they are powerful, but the thing they do to get that power is not intelligence. I suspect the crowd with this opinion defines intelligence in a way that means it can only be achieved by something I’d call an aligned ai, or perhaps they define it in a way that can only be achieved by an uncontrollable AI, somewhere in that space.
...anyway, how about… deep learning! or large ML? or sequential token generators. depends how specific you want to be. should alphafold count? it’s a transformer. should wavenet count? should sora count? should deepmind gato count?
re: “models of what”—most naively, they’re models of the dataset.
For those who insist these aren’t intelligent I like my extremely general term “Outcome Influencing System (OIS)” pronounced “oh-ee” and defined as “any system composed of capabilities and preferences which uses its capabilities informed by its preferences to influence reality towards outcomes it prefers”. Then it becomes trivially true that these things have capabilities and the capabilities are improving.
It seems like each of those terms does have a reasonable definition which is distinct from all of the other terms in the list:
LLMs: Neural networks trained (usually via autoregressive next-token prediction) on text corpora
Reasoning models: LLMs which have special delimiters marking the start and end of a chain of thought, and post-training to teach it to use those special delimiters
LLM agents: LLMs with access to tools invoked repeatedly until some stopping condition is met (or indefinitely, though we don’t see much of this yet)
Frontier Models: The subset of models trained with the largest compute budgets at any given time
Models: Neural networks trained to minimize some loss function over a dataset
AIs: Systems that perform tasks which required human intelligence last year
AGI/ASI: Systems matching or exceeding human cognitive capabilities across most/all domains respectively, but with the definition of which domains “count” gerrymandered such that no existing system counts as AGI
Prosaic AI Systems: Systems built by scaling up existing deep learning techniques rather than novel architectural insights
etc etc
It seems like you’re hoping for some term which encompasses all of “the AI systems which currently exist today”, “AI systems which can replace humans in all tasks and roles, which the frontier labs explicitly state they are trying to build”, and “the AI systems I expect will exist in the future, which can best be modeled as game-theoretic agents with some arbitrary utility function they are trying to maximize”. If that’s the case, though, I think you really do need multiple terms. I tentatively suggest “current frontier AI agents”, “drop-in-replacement-capable AI” or just “AGI”, and “superhuman wrapper minds” for the three categories respectively.
I agree that those terms all have distinct definitions. I think that was I want is basically a shorter term for Prosiac AI Systems.
Is the word “systems” required? “prosaic AI” seems like it’s short enough already, and “prosaic AI alignment” already has an aisafety.info page defining it as
By that definition, “prosaic AI alignment” should be parsed as “(prosaic AI) alignment”, implying that ” “prosaic AI” is already “AI trained and scaffolded using the techniques we are already using”. This definition of “Prosaic AI” seems to match usage elsewhere as well, e.g. Paul Christiano’s 2018 definition of “Prosaic AGI”
If that term is good enough for you, maybe you can make a short post explicitly coining the term, and link to that post the first time you use the term each time.
I do note one slight issue with defining “prosaic AI” as “AI created by scaling up already-known techniques”, which is that all techniques to train AI become “prosaic” as soon as those techniques stop being new and shiny.
Yeah—I don’t really like that the word “prosaic” has no connection to technical aspects of the currently prosaic models.
I don’t want to start referring to “the models previously known as prosaic” when new techniques become prosaic.
2025!Prosaic AI, or, if that’s not enough granularity, 2025-12-17!Prosaic AI. It’s even future-proof if there’s a singularity, you can extend it to 2025-12-17T19:23:43.718791198Z!Prosaic AI
To my understanding, Reasoning Model refers to models that have been explicitly fine-tuned to perform chain-of-thought reasoning effectively. Something analogous to DeepSeek’s GRPO. For instance, even if I told GPT-3 to show its work, it would not be a reasoning model.
As for LLM, it’s admittedly on the path to being a holdover term, but not for lack of effort. Some people tried to make VLM (Vision-Language Model) more common, but I’ve mainly heard it used in the context of lightweight models used to graft vision capability onto systems driven by single-modality LLMs.
It’s probably too late now, but maybe what we should have done is:
AGI = artificial general intelligence = an AI system that is generally capable rather than just narrowly capable
ASI = AGI that is better than the best humans at everything while also being faster and cheaper
We could then say that these are AGI companies building lots of little AGIs, and that they are on a path to ASI but aren’t there yet and probably won’t be for several more years. Claude is an AGI. GPT5 is an AGI. Etc. They keep improving in capability and generality and one day they’ll be ASIs, but they aren’t yet.
If AGI is human-equivalent for the purposes of developing a civilization, a collective of AGIs is at least as capable as humanity, plus it has AI advantages, so it’s much more capable than a single AGI instance, or any single human. This leads to ASI being often used synonymously with AGI lately (via individual vs. collective conflation). Such use of “ASI” might free up “AGI” for something closer to its original meaning, which didn’t carry the implication of human-equivalence. But this setup leaves the qualitatively-more-capable-than-humanity bucket without a label, that’s important for gesturing at AI danger.
I think the other extreme for meaning of “ASI”, being qualitatively much stronger than humanity, can be made more specific by having “ASI” refer to the level of capabilities that follows software-only singularity (under the assumption that it does advance capabilities a lot). This way, it’s neither literal technological maturity of hitting the limits of physical law, nor merely a collective of jagged-human-level AGI instances wielding their AI advantages. Maybe “RSI” is a more stable label for this, as in Superintelligence Strategy framing where “intelligence recursion” is the central destabilization bogeyman, rather than any given level of capabilities on its own.
What do you think of “Locked In AI (LIAI)” for when an AI becomes sufficiently capable that it’s preferences / utility function are “locked in” and can no longer be altered or avoided by other agents? This “locking in” is how I refer to the theoretical point of an RSI when it becomes too late to stop or alter course.
Also, for what it’s worth, I like “artificial general super intelligence (AGSI) which then frees up “AGI” for AI that does general reasoning and language at any level of capability, and hilariously frees up “ASI” to refer to any AI that does what it does better than any human, so a pocket calculator is an ASI because it does arithmetic better than any human. Though more confusing, LLMs would be ASI and not AGI because they are superhuman at text prediction, but chatbots made from LLMs would be AGI and not ASI because they reason and talk with general intelligence, but it seems more limited in some ways than human reasoning.
BAIR is calling them Compound AI Systems.
https://bair.berkeley.edu/blog/2024/02/18/compound-ai-systems/
They seem to be drawing an important distinction against “pure / monolithic” models, but the name is long and too general.
Another possibility that appears to be in wider use is neuro-symbolic AI.
An invention of my own is a Generative Programming Interface (GPI), analogous to an API, CLI, or GUI. The idea here is that LLMs or other forms of gen-AI can best be thought of as a crucial component of the program interface in this paradigm. I might have said “AII” (AI Interface), but that renders like the word “All.” AI Interface might work if spelled out, though.
Edit: Also found Hybrid Intelligent System → Hybrid AI or Hybrid LLM for short.
Edit 2: Intelligent Control → AI Control or LLM Control.
The problem with “neuro-symbolic AI” is that Gary Marcus types use it to refer to something distinct from the current paradigm. Even though it is ironically a pretty good description of the current paradigm.
What about “neuro-scaffold AI?” Marcus is loading the “symbolic” part of “neuro-symbolic” with the connotation of “definitionally NOT the current paradigm.” The term “scaffold” gets used to refer to traditional programs that wrap and attempt to refine and be controlled by AI, and is encompassing enough to include reasoning or agentic capabilities without assuming them.
That’s very good!
GPTs. Yes, it’s still an initialism like “LLM” but it’s much easier to pronounce (“jee-pee-tee”) and you can call them “jepeats” (rhyming with “repeats”) if you want.
This is reasonable, but includes “transformer” which seems a bit too narrow.
How about “IAC”: “Interactive Associative Composer”?
- “Interactive” as indicating there is a complex enough response to human inputs to feel organic, yet without the degree of self-motivation terms stemming from “agency” would connote.
- “Associative” as indicating the means by which it works, in a very generic way.
- “Composer” as (somewhat redundantly) indicating that it produces responses as the output of its associative process, as well as giving a connotation of quality.
Not pronounceable.