Steve Steiner
banner
stevejsteiner.bsky.social
Steve Steiner
@stevejsteiner.bsky.social
Love learning new stuff. Exploring emerging AI dev practice and tools. Enthusiast of 4E Cog Sci, Story structure, process ontology.
Ah specific api! I’m highly skeptical of compact/summarize. However I’ve seen a couple of times already some hard fought insights age like milk in this space.
November 22, 2025 at 10:10 PM
claude code did compaction before sonnet 4.5 … Is this something different?
November 22, 2025 at 10:04 PM
Interesting datapoint for “LLM as judge” —

**A reasonable penalty for illegally destroying the East Wing is:

0.5% – 1.5% of U.S. GDP
≈ $140B – $420B

Below that range → society underprices constitutional symbolism.
Above it → penalty becomes politically performative rather than principled.
November 21, 2025 at 3:24 PM
“If the East Wing really was pulled down overnight and reduced to rubble, the result would be public outcry and political fallout, rather than indifference.”
November 21, 2025 at 2:30 PM
Do you mean Elon Musk the incredibly skilled Diablo 4 player? He is also building a propaganda machine? How does he find the time?
November 21, 2025 at 1:42 PM
This is like landing on the moon. Older generations were just better at technology. Kids these days just can’t make a simple human-robot hybrid.
November 20, 2025 at 11:04 PM
Here is far more than you’d want to read on “over-imitation”, but it gives jumping off points:
research-repository.st-andrews.ac.uk
November 16, 2025 at 5:57 PM
November 15, 2025 at 3:34 PM
I have a horrifically bad habit of transforming questions so that either yes or no is indeterminate.

Do you think this is a good idea? Or should we consider other options?

Yes.

Doh!
November 13, 2025 at 9:49 PM
You mean Currer Bell?
November 11, 2025 at 3:53 PM
New ai coding innovation: “Tech credit default swaps”
November 10, 2025 at 3:16 PM
A friend who worked at GE couldn’t watch 30 rock. She said it hit too close to home.
November 8, 2025 at 7:31 PM
I recall (in the 90’s) Hal Abelson, after teaching back propagation, then said the real training algorithm for neural networks is the Graduate Student algorithm.
November 8, 2025 at 7:27 PM
The original imitation game concerns performative identity within a social role, with a triangular competition structure.

The machine must model the other contestant’s signaling behavior.

It has second order signaling in a game about social norms -> a Nash Equilibrium.

Arguably a sentience test.
November 7, 2025 at 3:59 PM