@robmurrish.bsky.social
What editor is this? So I can never
November 23, 2025 at 3:03 AM
What editor is this? So I can never
Makes all the difference if it's once a year to advance a project you really care about, or monthly (or more) for what amounts to understaffing or poor planning.
November 22, 2025 at 6:16 PM
Makes all the difference if it's once a year to advance a project you really care about, or monthly (or more) for what amounts to understaffing or poor planning.
Turning to trump and asking "Are you?"
would have been great
would have been great
November 22, 2025 at 4:23 AM
Turning to trump and asking "Are you?"
would have been great
would have been great
I'm inspired to spend some time on a theme for myself!
November 22, 2025 at 1:24 AM
I'm inspired to spend some time on a theme for myself!
I like diffuse. Like, "here's a little tension for you, and some for you, and I'll put some over here in the corner"
November 22, 2025 at 1:20 AM
I like diffuse. Like, "here's a little tension for you, and some for you, and I'll put some over here in the corner"
The root-cause analysis will miss this point.
November 18, 2025 at 11:44 PM
The root-cause analysis will miss this point.
The runes must be carved *precisely*
November 18, 2025 at 2:39 AM
The runes must be carved *precisely*
Is this the best place to follow you if we want to see all the details for this project?
November 17, 2025 at 5:48 PM
Is this the best place to follow you if we want to see all the details for this project?
That was what I was trying to say. The resource use is not your main point against it (it seems).
November 15, 2025 at 6:58 PM
That was what I was trying to say. The resource use is not your main point against it (it seems).
Then say in 3 years, there's specialized hardware and genAI isn't using any more resources to train or use than other software. Are you still against it like you are today?
November 15, 2025 at 6:55 PM
Then say in 3 years, there's specialized hardware and genAI isn't using any more resources to train or use than other software. Are you still against it like you are today?
Lol. To some degree I just embrace it. The internet still feels positive sum even knowing many facts in my head are false (though I feel a little called out, the same argument could apply to LLMs).
Also, prediction markets. I quantify the calibration of my internal world models. Highly recommend.
Also, prediction markets. I quantify the calibration of my internal world models. Highly recommend.
November 15, 2025 at 6:52 PM
Lol. To some degree I just embrace it. The internet still feels positive sum even knowing many facts in my head are false (though I feel a little called out, the same argument could apply to LLMs).
Also, prediction markets. I quantify the calibration of my internal world models. Highly recommend.
Also, prediction markets. I quantify the calibration of my internal world models. Highly recommend.
I fully expect I'll use code gen eventually, I give it a fresh shot every 6 months or so, but it's not there for me yet. I think it's more than just "activation energy" needed, but your experience is a point to the contrary.
November 15, 2025 at 6:37 PM
I fully expect I'll use code gen eventually, I give it a fresh shot every 6 months or so, but it's not there for me yet. I think it's more than just "activation energy" needed, but your experience is a point to the contrary.
My experience wrt staff is much more sr engineer than manager, so I admit negative-sum-first-staffer isn't something that came to mind in the comparison.
November 15, 2025 at 6:37 PM
My experience wrt staff is much more sr engineer than manager, so I admit negative-sum-first-staffer isn't something that came to mind in the comparison.
RE: "it's like having staff". Interesting, but jr staff can learn. Specific issues take an incident or two, and it's just never so long before the mistakes are so clever and subtle they become sr.
It's better and worse with LLMs, harder to fix specific issues, but always seeing global improvement.
It's better and worse with LLMs, harder to fix specific issues, but always seeing global improvement.
November 15, 2025 at 6:05 PM
RE: "it's like having staff". Interesting, but jr staff can learn. Specific issues take an incident or two, and it's just never so long before the mistakes are so clever and subtle they become sr.
It's better and worse with LLMs, harder to fix specific issues, but always seeing global improvement.
It's better and worse with LLMs, harder to fix specific issues, but always seeing global improvement.
- the absolute mess of code provenance, and the equivalent IP issues in creative areas I understand less well than coding
- a general distrust of how early and forcefully it's being pushed into broad public use
- a general distrust of how early and forcefully it's being pushed into broad public use
November 15, 2025 at 6:05 PM
- the absolute mess of code provenance, and the equivalent IP issues in creative areas I understand less well than coding
- a general distrust of how early and forcefully it's being pushed into broad public use
- a general distrust of how early and forcefully it's being pushed into broad public use
- I like the subtle "why" in art. Why that exact word? Why that brush stroke? Why that chord? Even stupid answers- "I ran out of yellow" -are more satisfying than even a technically superior product where "why" always points back to the same matrix. (I did not know this about myself until recently).
November 15, 2025 at 6:05 PM
- I like the subtle "why" in art. Why that exact word? Why that brush stroke? Why that chord? Even stupid answers- "I ran out of yellow" -are more satisfying than even a technically superior product where "why" always points back to the same matrix. (I did not know this about myself until recently).
My own reasons to avoid genAI include
- the cognitive bias of seeing something presented factually "in print" sneaking a "true" label into the brain
- code models don't yet save me time or effort, or do things I can't. I check in every few months, I expect this will switch eventually
- the cognitive bias of seeing something presented factually "in print" sneaking a "true" label into the brain
- code models don't yet save me time or effort, or do things I can't. I check in every few months, I expect this will switch eventually
November 15, 2025 at 6:05 PM
My own reasons to avoid genAI include
- the cognitive bias of seeing something presented factually "in print" sneaking a "true" label into the brain
- code models don't yet save me time or effort, or do things I can't. I check in every few months, I expect this will switch eventually
- the cognitive bias of seeing something presented factually "in print" sneaking a "true" label into the brain
- code models don't yet save me time or effort, or do things I can't. I check in every few months, I expect this will switch eventually
I'm more concerned that @edfinnerty.bsky.social and others use this argument because it's easy, even though it's not their main con.
I'm also concerned about resources going into new data centers, etc., but that's encouraged by paying the big companies (and hype), not by using small local models.
I'm also concerned about resources going into new data centers, etc., but that's encouraged by paying the big companies (and hype), not by using small local models.
November 15, 2025 at 6:05 PM
I'm more concerned that @edfinnerty.bsky.social and others use this argument because it's easy, even though it's not their main con.
I'm also concerned about resources going into new data centers, etc., but that's encouraged by paying the big companies (and hype), not by using small local models.
I'm also concerned about resources going into new data centers, etc., but that's encouraged by paying the big companies (and hype), not by using small local models.
I'm unbothered that the early models were resource intensive to train. That's true of many technologies that I use, but I'm not going to refuse an MRI because superconductor R&D is expensive.
November 15, 2025 at 6:05 PM
I'm unbothered that the early models were resource intensive to train. That's true of many technologies that I use, but I'm not going to refuse an MRI because superconductor R&D is expensive.
I'm going to risk sharing some opinions where I should maybe just leave this alone.
I think @coloradotravis.bsky.social is basically right to use (electrical) cost as a proxy for damage done.
I have reasons not to use LLMs, but I predict resource use will trend to zero.
I think @coloradotravis.bsky.social is basically right to use (electrical) cost as a proxy for damage done.
I have reasons not to use LLMs, but I predict resource use will trend to zero.
November 15, 2025 at 6:05 PM
I'm going to risk sharing some opinions where I should maybe just leave this alone.
I think @coloradotravis.bsky.social is basically right to use (electrical) cost as a proxy for damage done.
I have reasons not to use LLMs, but I predict resource use will trend to zero.
I think @coloradotravis.bsky.social is basically right to use (electrical) cost as a proxy for damage done.
I have reasons not to use LLMs, but I predict resource use will trend to zero.
I also appreciate how deep this thread went. A disagreement this full of considered positions and civility is rare to see.
November 15, 2025 at 5:01 PM
I also appreciate how deep this thread went. A disagreement this full of considered positions and civility is rare to see.
And we moved content to private spaces (like discord) instead of posting to the open web.
November 14, 2025 at 1:04 AM
And we moved content to private spaces (like discord) instead of posting to the open web.