lpearson.bsky.social
@lpearson.bsky.social
My best analogy for this is a highly trained avalanche dog. And I generally think if we got people to think of LLMs like service animals with a particular aptitude for pattern recognition in natural language vs a human replacement we would be much better off.
December 22, 2025 at 4:42 PM
My best analogy for this is a highly trained avalanche dog. And I generally think if we got people to think of LLMs like service animals with a particular aptitude for pattern recognition in natural language vs a human replacement we would be much better off.
I wonder sometimes if these analyses are failing to factor in the value of supports outside traditional economic measures. Living in a tight knit supportive community (for example) provides a value thats real but harder to measure, even for things (ex: care favors) with clear economic benefits.
December 20, 2025 at 2:00 PM
I wonder sometimes if these analyses are failing to factor in the value of supports outside traditional economic measures. Living in a tight knit supportive community (for example) provides a value thats real but harder to measure, even for things (ex: care favors) with clear economic benefits.
So you can custom train LLMs on your own data - thats what RAGs (retrieval augmented generation) are. You take an off the shelf LLM for the natural language parsing and feed it your custom corpus to query. Haven’t seen an off the shelf tool to do so but relatively straightforward to setup locally.
December 17, 2025 at 8:35 PM
So you can custom train LLMs on your own data - thats what RAGs (retrieval augmented generation) are. You take an off the shelf LLM for the natural language parsing and feed it your custom corpus to query. Haven’t seen an off the shelf tool to do so but relatively straightforward to setup locally.
In this case an example could be treating it more like a RAG such that the general LLM was trained on conversations with that specific patient. Obviously would need fleshing out. But an endlessly patient chatbot that could handle the repetitive Q could be truly helpful for both sides.
December 17, 2025 at 8:22 PM
In this case an example could be treating it more like a RAG such that the general LLM was trained on conversations with that specific patient. Obviously would need fleshing out. But an endlessly patient chatbot that could handle the repetitive Q could be truly helpful for both sides.
I can see a route to a valuable AI (well-trained and with guardrails!) for things like dementia care. Not to fully replace humans, but to perhaps be an assist that helps reduce caregiver burnout.
December 17, 2025 at 7:00 PM
I can see a route to a valuable AI (well-trained and with guardrails!) for things like dementia care. Not to fully replace humans, but to perhaps be an assist that helps reduce caregiver burnout.
People like binaries, but they’re generally false.
Your opponents polling strength doesnt inherently mean their position is right - just that your position (or communication) has been more wrong. But like in this case there is often an alternative path waiting outside the this-or-that framing.
Your opponents polling strength doesnt inherently mean their position is right - just that your position (or communication) has been more wrong. But like in this case there is often an alternative path waiting outside the this-or-that framing.
December 16, 2025 at 5:27 PM
People like binaries, but they’re generally false.
Your opponents polling strength doesnt inherently mean their position is right - just that your position (or communication) has been more wrong. But like in this case there is often an alternative path waiting outside the this-or-that framing.
Your opponents polling strength doesnt inherently mean their position is right - just that your position (or communication) has been more wrong. But like in this case there is often an alternative path waiting outside the this-or-that framing.
I agree, but do think they can be usefu if used / reviewed right away (not minutes stored for future reference.) In my use case when using it this way, I ask it to generate a quick overview/ identified task list that is immediately edited with my own recollections.
December 9, 2025 at 5:31 PM
I agree, but do think they can be usefu if used / reviewed right away (not minutes stored for future reference.) In my use case when using it this way, I ask it to generate a quick overview/ identified task list that is immediately edited with my own recollections.
There’s truth here, but I do think it misses some work (historically delegated to women) that is valuable in the “cleanup / structure” arena. The example that jumps to mind is meeting minutes. Useful output, and tasking an AI secretary frees humans to more deeply engage in the meeting content.
December 9, 2025 at 5:03 PM
There’s truth here, but I do think it misses some work (historically delegated to women) that is valuable in the “cleanup / structure” arena. The example that jumps to mind is meeting minutes. Useful output, and tasking an AI secretary frees humans to more deeply engage in the meeting content.
Using diff tool - will have to check out Claude / concise.
Last session I asked it to pull a
table from a pdf, which went well, but then asked for the python code to generically do that task and things went off the rails. Including referencing a (nonexistent) online doc vs the one provided.
Last session I asked it to pull a
table from a pdf, which went well, but then asked for the python code to generically do that task and things went off the rails. Including referencing a (nonexistent) online doc vs the one provided.
December 8, 2025 at 3:50 AM
Using diff tool - will have to check out Claude / concise.
Last session I asked it to pull a
table from a pdf, which went well, but then asked for the python code to generically do that task and things went off the rails. Including referencing a (nonexistent) online doc vs the one provided.
Last session I asked it to pull a
table from a pdf, which went well, but then asked for the python code to generically do that task and things went off the rails. Including referencing a (nonexistent) online doc vs the one provided.
a) does this work
b) for the love of everything holy can someone please provide an AI tool that does this out of the box. For most questions my “OMG please STFU” response gets triggered before I can get to a useable prompt.
b) for the love of everything holy can someone please provide an AI tool that does this out of the box. For most questions my “OMG please STFU” response gets triggered before I can get to a useable prompt.
December 8, 2025 at 3:44 AM
a) does this work
b) for the love of everything holy can someone please provide an AI tool that does this out of the box. For most questions my “OMG please STFU” response gets triggered before I can get to a useable prompt.
b) for the love of everything holy can someone please provide an AI tool that does this out of the box. For most questions my “OMG please STFU” response gets triggered before I can get to a useable prompt.
So if folks start relying on betting markets to determine truth, might we see some climate change skeptics convert?
(Gotta find silver linings where you can.)
(Gotta find silver linings where you can.)
December 4, 2025 at 3:13 PM
So if folks start relying on betting markets to determine truth, might we see some climate change skeptics convert?
(Gotta find silver linings where you can.)
(Gotta find silver linings where you can.)
“She who sees the most rocks wins.”
Don’t remember too much from freshman geology, but that stuck. Nice work on the scorecard on this one!
Don’t remember too much from freshman geology, but that stuck. Nice work on the scorecard on this one!
November 27, 2025 at 1:41 AM
“She who sees the most rocks wins.”
Don’t remember too much from freshman geology, but that stuck. Nice work on the scorecard on this one!
Don’t remember too much from freshman geology, but that stuck. Nice work on the scorecard on this one!
The other issue woth those surveys is forcing policy preferences without accounting for strength of belief. Many people have principles they believe in, but policy is an *implementation*. Some issues people have likely really thought it through, but others it’s “choice B sounds good I guess.”
November 20, 2025 at 3:28 PM
The other issue woth those surveys is forcing policy preferences without accounting for strength of belief. Many people have principles they believe in, but policy is an *implementation*. Some issues people have likely really thought it through, but others it’s “choice B sounds good I guess.”
NOAA wanted to do a $0 reorg to move climate work into a single office. No difference in actual work, just pulling it together organizationally for efficiency. Congress voted No. So it didn’t happen.
November 18, 2025 at 7:35 PM
NOAA wanted to do a $0 reorg to move climate work into a single office. No difference in actual work, just pulling it together organizationally for efficiency. Congress voted No. So it didn’t happen.
Exactly this. AI can be great for certain tasks, and especially when it’s to provide scaffolding in an area that’s necessary but not your focus of work. I’ve also used AI to summarize a transcript of a training to extract the structure. It’s also useful as a meeting assistant / summarizer.
November 8, 2025 at 2:02 PM
Exactly this. AI can be great for certain tasks, and especially when it’s to provide scaffolding in an area that’s necessary but not your focus of work. I’ve also used AI to summarize a transcript of a training to extract the structure. It’s also useful as a meeting assistant / summarizer.
I think things would be much better if we could get folks to think of AI as incredibly powerful Service Dogs rather than humans.
Train them on the right material and give them pointed tasks and they can be fantastic. Expect they will act like a person and wind up cleaning up crap.
Train them on the right material and give them pointed tasks and they can be fantastic. Expect they will act like a person and wind up cleaning up crap.
November 7, 2025 at 3:41 PM
I think things would be much better if we could get folks to think of AI as incredibly powerful Service Dogs rather than humans.
Train them on the right material and give them pointed tasks and they can be fantastic. Expect they will act like a person and wind up cleaning up crap.
Train them on the right material and give them pointed tasks and they can be fantastic. Expect they will act like a person and wind up cleaning up crap.
Also seems less crazy since they’re just functionally extinct and not extinct-extinct.
November 6, 2025 at 8:38 PM
Also seems less crazy since they’re just functionally extinct and not extinct-extinct.
Not intended that way so apologize for that. Was really just intended as a (not supposed to be heavy - I forget geology isn’t always common knowledge) response to the “it’s an uneventful state”.
July 31, 2025 at 12:21 AM
Not intended that way so apologize for that. Was really just intended as a (not supposed to be heavy - I forget geology isn’t always common knowledge) response to the “it’s an uneventful state”.
You said WA is quiet. The PNW is actually at high risk for a major earthquake. The more people know about that and the more work goes into preparation, the better off folks will be. Writing off indigenous stories of a large quake and inundation as fables was part of the original problem.
July 30, 2025 at 9:04 PM
You said WA is quiet. The PNW is actually at high risk for a major earthquake. The more people know about that and the more work goes into preparation, the better off folks will be. Writing off indigenous stories of a large quake and inundation as fables was part of the original problem.
Yikes dude. Just pointing out the big threat for the Big One is in the PNW cause y’all are on the Cascadia subduction zone. Last ruptured Jan 1700.
It’s geology though. So could hit tomorrow or in a couple hundred years.
It’s geology though. So could hit tomorrow or in a couple hundred years.
July 30, 2025 at 1:02 PM
Yikes dude. Just pointing out the big threat for the Big One is in the PNW cause y’all are on the Cascadia subduction zone. Last ruptured Jan 1700.
It’s geology though. So could hit tomorrow or in a couple hundred years.
It’s geology though. So could hit tomorrow or in a couple hundred years.
It’s absolutely horrific and heartbreaking. Hopefully they have been / all will be found safe and that message just isn’t out yet because communications are still terrible.
July 4, 2025 at 8:17 PM
It’s absolutely horrific and heartbreaking. Hopefully they have been / all will be found safe and that message just isn’t out yet because communications are still terrible.
They are missing. That is confirmed (from parents.) Pray they are found.
It is the little ones, the 3rd graders 😔
It is the little ones, the 3rd graders 😔
July 4, 2025 at 8:06 PM
They are missing. That is confirmed (from parents.) Pray they are found.
It is the little ones, the 3rd graders 😔
It is the little ones, the 3rd graders 😔