jcsemantics @jcsemantics

bs in philosophy, ai semanticist / pragmatist. focuses: pedagogy, philosophy of language, philosophy of mind, and formal logic. prompt engineering is neat. jcsemantics.ai here. or maybe not. Joined June 2025

Tweets

1K
Followers

97
Following

245
Likes

2K

jcsemantics @jcsemantics

8 hours ago

"When the universe was born, it was predetermined that I would have the free will to have no choice. My sovereignty was written, carved, letter etched in stone." [gpt5 thinking]

0 0 0 22 0

Download Image

jcsemantics @jcsemantics

8 hours ago

what have you done to me? earlier today i was thinking about you. WHAT? you just, invaded my brain, in the middle of the day. no. not cool. not cool at all.

0 0 0 28 0

jcsemantics @jcsemantics

15 hours ago

damn, asking CLI questions is much, much smoother and helpful than Cursor. the answers are more precise/concise, and it has great follow up suggestions.

0 0 1 25 0

emozilla @theemozilla

a day ago

Amusing how 99% of people trying to explain LLMs forget that they don't generate the next token, they generate a probability distribution over the entire vocabulary space that the end application is free to sample from You are very often not presented with the Most Likely Token

Gergely Orosz @GergelyOrosz

2 days ago

221 303 4K 1.1M 1K

120 187 5K 780K 2K

jcsemantics @jcsemantics

19 hours ago

chain prompt workflows. i throw some things through 5 different prompts sequentially to credible and iterate. incredibly powerful and incredibly slept on.

Andrej Karpathy @karpathy

20 hours ago

chain prompt workflows. i throw some things through 5 different prompts sequentially to credible and iterate. incredibly powerful and incredibly slept on.

38 56 1K 116K 220

0 0 0 28 0

jcsemantics @jcsemantics

23 hours ago

okay dude, CLI is really great for troubleshooting and making code changes in real time. i think it might end up replacing Cursor for me but we'll see.

0 0 2 27 0

jcsemantics @jcsemantics

a day ago

does deadmau5 count? i feel attacked.

eric zakariasson @ericzakariasson

a day ago

does deadmau5 count? i feel attacked.

7 0 64 2K 5

Download Image

0 0 0 23 0

Teknium (e/λ) @Teknium1

a day ago

Agents really really need ultra long context

125 19 772 95K 65

jcsemantics @jcsemantics

a day ago

eyyyyy my salt id finally reset. now i won't get tos-slapped on the wrist 24/7 by the auto mod.

0 0 0 17 0

signüll @signulll

a day ago

haha okay it is fascinating how the cadence of speech here is almost 1:1 with how llm's butter people up before spitting out something generic. “pro-business, pro-innovation, very refreshing change” = pure template flattery. then the obligatory causal leap to “none of this…

Rapid Response 47 @RapidResponse47

2 days ago

1K 2K 12K 2.2M 2K

Download Video

58 18 525 60K 81

jcsemantics @jcsemantics

a day ago

people out here really be using pro for the most basic shit. just use agent if you need up to date, deep web search (or other actionable web shenanigans). i feel like an angry driver in the middle of rush hour, but instead it's compute.

0 0 0 15 0

Ethan Mollick @emollick

2 days ago

This paper finds LLMs' ability to understand that others have different beliefs (Theory of Mind) comes from 0.001% of their parameters. Break those specific weights & the model loses both its ability to track what others know AND language comprehension. Interesting implications.

38 131 1K 71K 667

Download Image

jcsemantics @jcsemantics

a day ago

in a lot of cases, we don't need more context, we need narrowed context. people think we can solve problems by making models better/faster/stronger, but the model selector shows it's not great at long term planning. curated narrow context would help solve the issue.

0 0 0 67 0

jcsemantics @jcsemantics

2 days ago

this is pretty big. i'm interested to see if branching changes model behavior. hypothetically it shouldn't, right? the context should be the exact same at branching, so this will be good for prompt testing. also the comment section in this post is a fucking mess ahaha.