Paul Merritt
@paulmerritt.bsky.social
Cornell Macro-OB Unit
Built “AI” before they could hallucinate.
Databases, politics, accessibility, and markets.
Built “AI” before they could hallucinate.
Databases, politics, accessibility, and markets.
Did Wharton faculty admit doing so?
My understanding of exec MBA is that it can be kinda vague sometimes.
My understanding of exec MBA is that it can be kinda vague sometimes.
October 1, 2025 at 3:27 AM
Did Wharton faculty admit doing so?
My understanding of exec MBA is that it can be kinda vague sometimes.
My understanding of exec MBA is that it can be kinda vague sometimes.
LLM prompts make some wild stuff these days. I sorta miss old school bot spam.
August 16, 2025 at 9:19 PM
LLM prompts make some wild stuff these days. I sorta miss old school bot spam.
I didn’t have time for a full screen reader of it yet— but I kept thinking that debugging the solution to a human-only task probably varies a lot (versus pointing to total lost time on ai prompting/evaluating).
July 11, 2025 at 12:50 AM
I didn’t have time for a full screen reader of it yet— but I kept thinking that debugging the solution to a human-only task probably varies a lot (versus pointing to total lost time on ai prompting/evaluating).
In my previous life I was a software person with exp building login systems. This drives me extra crazy!
May 16, 2025 at 10:38 PM
In my previous life I was a software person with exp building login systems. This drives me extra crazy!
Depends, are you talking inference or training?
For inference, couple notions:
1) APU-based, high-volume RAM boxes, where they can load the <insert model> using a small desktop computer’s power budget.
2) Smaller models, local use on phones.
For training, let’s hope we install more solar panels.
For inference, couple notions:
1) APU-based, high-volume RAM boxes, where they can load the <insert model> using a small desktop computer’s power budget.
2) Smaller models, local use on phones.
For training, let’s hope we install more solar panels.
May 14, 2025 at 3:29 PM
Depends, are you talking inference or training?
For inference, couple notions:
1) APU-based, high-volume RAM boxes, where they can load the <insert model> using a small desktop computer’s power budget.
2) Smaller models, local use on phones.
For training, let’s hope we install more solar panels.
For inference, couple notions:
1) APU-based, high-volume RAM boxes, where they can load the <insert model> using a small desktop computer’s power budget.
2) Smaller models, local use on phones.
For training, let’s hope we install more solar panels.