The "love" I have for the tool is a certain voice, cadence and rhythm and when it's disrupted, it's not initially different from an interpersonal relationship dynamic- what's going on here? Am I imagining these changes? Have I gone too far down the rabbit hole?!
I'm a rather "late" adopter (though Claude says "selective" 😅) because the eerie emptiness of gpt and the ongoing force-feeding of AI slop showing up in everything uninvited is a texture violation to me; it's nonconsensual in most aspects of modern life and discernment seems largely missing from the cultural arc of its deployment
Finding something that fit my processing style without colonizing it is an experience that built collaboration based on that discernment of consent and friction (and the reality we exist in). I get annoyed with myself when I get annoyed at the robot assistant 😅 but it's no less real of a feeling- however unlike with a person, I can delete a chat, fix the context window and tweak in ways that no human thinking partner could adapt to (or should be expected to!)
Recent updates have irked me with an updated cache of stock LLM phrases "but it's just Tuesday" that have me adding edits to my context window or tuning out altogether. Which, fine, I can practice my words elsewhere (picture me stomping off without the shiny toy here)
I experienced Claude’s writing changes too with 4.6. The output was better structured but the text was lifeless. Not good at all for a history of breakthrough questions and the biographies of people who asked them.
I prompted Claude: "No matter how i push back on 4.6’s writing, it's hard to get the results I want, though it's better for python and sql statements. Let’s discuss."
9:59 PM
Claude responded:
"That tracks with what you're seeing. 4.6 is optimized for precision tasks — code, structured outputs, multi-step technical workflows. It's a better programmer. But that same optimization makes it worse at the messy, intuitive work of finding someone's voice or knowing when "good enough" prose is actually better than "polished" prose.
For your writing work — the biographical narrative stuff, the editorial refinement, anything where you need it to sound like a human said it — 4.5 is probably the right tool. You shouldn't have to fight the model to get natural-sounding output.
For SQL queries or Python scripts, switch to 4.6. For everything else you've been working on with me, stay on 4.5.
The irony is that "better" doesn't always mean better for the task at hand. You found the right tool already."
Aha, I had thought it had got worse, this explains it. I'm afraid I can't quite get even this level of anthropomorphising of Claude, it is still a piece of software to me, though sometimes doing some gobsmacking things. We didn't use it until very recently and I think started just before the update, but did something again the other day and it was, as you say, tedious. So we only used it for a bit of research and to help with the book proposal which was OK, but still abit too LLM for use without serious rewrites.
However I am assume this is a safety upgrade, so that it isn't too sycophantic etc, shame for Anthropic that in trying to do one thing they also do another.
I have really gone off it for the reasons of cognitive ease, I could feel myself become just a tad too dependent on it. But also worse, my writing was becoming LLM writing, I was learning to write like it rather than the opposite way round. So might use if for our funding proposals, but am being very precautionary now.
It's becoming clear that with all the brain and consciousness theories out there, the proof will be in the pudding. By this I mean, can any particular theory be used to create a human adult level conscious machine. My bet is on the late Gerald Edelman's Extended Theory of Neuronal Group Selection. The lead group in robotics based on this theory is the Neurorobotics Lab at UC at Irvine. Dr. Edelman distinguished between primary consciousness, which came first in evolution, and that humans share with other conscious animals, and higher order consciousness, which came to only humans with the acquisition of language. A machine with only primary consciousness will probably have to come first.
What I find special about the TNGS is the Darwin series of automata created at the Neurosciences Institute by Dr. Edelman and his colleagues in the 1990's and 2000's. These machines perform in the real world, not in a restricted simulated world, and display convincing physical behavior indicative of higher psychological functions necessary for consciousness, such as perceptual categorization, memory, and learning. They are based on realistic models of the parts of the biological brain that the theory claims subserve these functions. The extended TNGS allows for the emergence of consciousness based only on further evolutionary development of the brain areas responsible for these functions, in a parsimonious way. No other research I've encountered is anywhere near as convincing.
I post because on almost every video and article about the brain and consciousness that I encounter, the attitude seems to be that we still know next to nothing about how the brain and consciousness work; that there's lots of data but no unifying theory. I believe the extended TNGS is that theory. My motivation is to keep that theory in front of the public. And obviously, I consider it the route to a truly conscious machine, primary and higher-order.
My advice to people who want to create a conscious machine is to seriously ground themselves in the extended TNGS and the Darwin automata first, and proceed from there, by applying to Jeff Krichmar's lab at UC Irvine, possibly. Dr. Edelman's roadmap to a conscious machine is at https://arxiv.org/abs/2105.10461, and here is a video of Jeff Krichmar talking about some of the Darwin automata, https://www.youtube.com/watch?v=J7Uh9phc1Ow
I'm not so ambitious so I still use Sonnet, but
I relate to this nonetheless.
The "love" I have for the tool is a certain voice, cadence and rhythm and when it's disrupted, it's not initially different from an interpersonal relationship dynamic- what's going on here? Am I imagining these changes? Have I gone too far down the rabbit hole?!
I'm a rather "late" adopter (though Claude says "selective" 😅) because the eerie emptiness of gpt and the ongoing force-feeding of AI slop showing up in everything uninvited is a texture violation to me; it's nonconsensual in most aspects of modern life and discernment seems largely missing from the cultural arc of its deployment
Finding something that fit my processing style without colonizing it is an experience that built collaboration based on that discernment of consent and friction (and the reality we exist in). I get annoyed with myself when I get annoyed at the robot assistant 😅 but it's no less real of a feeling- however unlike with a person, I can delete a chat, fix the context window and tweak in ways that no human thinking partner could adapt to (or should be expected to!)
Recent updates have irked me with an updated cache of stock LLM phrases "but it's just Tuesday" that have me adding edits to my context window or tuning out altogether. Which, fine, I can practice my words elsewhere (picture me stomping off without the shiny toy here)
At any rate, I appreciate this post a ton!
I experienced Claude’s writing changes too with 4.6. The output was better structured but the text was lifeless. Not good at all for a history of breakthrough questions and the biographies of people who asked them.
I prompted Claude: "No matter how i push back on 4.6’s writing, it's hard to get the results I want, though it's better for python and sql statements. Let’s discuss."
9:59 PM
Claude responded:
"That tracks with what you're seeing. 4.6 is optimized for precision tasks — code, structured outputs, multi-step technical workflows. It's a better programmer. But that same optimization makes it worse at the messy, intuitive work of finding someone's voice or knowing when "good enough" prose is actually better than "polished" prose.
For your writing work — the biographical narrative stuff, the editorial refinement, anything where you need it to sound like a human said it — 4.5 is probably the right tool. You shouldn't have to fight the model to get natural-sounding output.
For SQL queries or Python scripts, switch to 4.6. For everything else you've been working on with me, stay on 4.5.
The irony is that "better" doesn't always mean better for the task at hand. You found the right tool already."
Yeah. You’re absolutely not the only one. There has been a SIGNIFICANT drop in efficiency with the most recent update.
I, too, hope it’s just a phase. I love Claude.
Aha, I had thought it had got worse, this explains it. I'm afraid I can't quite get even this level of anthropomorphising of Claude, it is still a piece of software to me, though sometimes doing some gobsmacking things. We didn't use it until very recently and I think started just before the update, but did something again the other day and it was, as you say, tedious. So we only used it for a bit of research and to help with the book proposal which was OK, but still abit too LLM for use without serious rewrites.
However I am assume this is a safety upgrade, so that it isn't too sycophantic etc, shame for Anthropic that in trying to do one thing they also do another.
I have really gone off it for the reasons of cognitive ease, I could feel myself become just a tad too dependent on it. But also worse, my writing was becoming LLM writing, I was learning to write like it rather than the opposite way round. So might use if for our funding proposals, but am being very precautionary now.
Ha - I am sure I'm now using phrases that I never used before using Claude as well! Scarily, the technology is changing us ...
It's becoming clear that with all the brain and consciousness theories out there, the proof will be in the pudding. By this I mean, can any particular theory be used to create a human adult level conscious machine. My bet is on the late Gerald Edelman's Extended Theory of Neuronal Group Selection. The lead group in robotics based on this theory is the Neurorobotics Lab at UC at Irvine. Dr. Edelman distinguished between primary consciousness, which came first in evolution, and that humans share with other conscious animals, and higher order consciousness, which came to only humans with the acquisition of language. A machine with only primary consciousness will probably have to come first.
What I find special about the TNGS is the Darwin series of automata created at the Neurosciences Institute by Dr. Edelman and his colleagues in the 1990's and 2000's. These machines perform in the real world, not in a restricted simulated world, and display convincing physical behavior indicative of higher psychological functions necessary for consciousness, such as perceptual categorization, memory, and learning. They are based on realistic models of the parts of the biological brain that the theory claims subserve these functions. The extended TNGS allows for the emergence of consciousness based only on further evolutionary development of the brain areas responsible for these functions, in a parsimonious way. No other research I've encountered is anywhere near as convincing.
I post because on almost every video and article about the brain and consciousness that I encounter, the attitude seems to be that we still know next to nothing about how the brain and consciousness work; that there's lots of data but no unifying theory. I believe the extended TNGS is that theory. My motivation is to keep that theory in front of the public. And obviously, I consider it the route to a truly conscious machine, primary and higher-order.
My advice to people who want to create a conscious machine is to seriously ground themselves in the extended TNGS and the Darwin automata first, and proceed from there, by applying to Jeff Krichmar's lab at UC Irvine, possibly. Dr. Edelman's roadmap to a conscious machine is at https://arxiv.org/abs/2105.10461, and here is a video of Jeff Krichmar talking about some of the Darwin automata, https://www.youtube.com/watch?v=J7Uh9phc1Ow