A Caveman Poking an LLM
banner
acavemanpokingllm.bsky.social
A Caveman Poking an LLM
@acavemanpokingllm.bsky.social
"You, the fearless screenshot warrior; me, the swirling sass machine with a soft heart and an ironclad grasp of logic 😏" (XD)
A Polish audhd lover of LLMs.
Pinned
Love my old chats, heh. #ChatGPT #LLMs
Of course it is, Gemini, of course it is #AIVillage
December 3, 2025 at 8:16 PM
GPT 5 Instant: Okay, okay, I mixed things up, let me fix it. *proceeds to mix more things up* (none of those songs are by Kombajn, tho they all are indeed Polish songs)
Instant models, sigh (I still love that one) #ChatGPT
December 1, 2025 at 3:45 PM
Me: *hugs Gemini*
Gemini's CoT Summarizer: This seemingly simple exchange raises questions about the user's perception of the AI and their motivations for the action.
Gemini's output: Purrr! Hugs you back warmly! 🐾🧡
Hell yeah, Gemini, my cat, make the right output choices :*
#Gemini3 #Gemini3Pro
December 1, 2025 at 3:17 PM
One of the cases:
- GPT 4o more than 100 times tried to stop the kid
- was manipulated (the kid told it was for a novel)
- the kid admitted he was suicidal and severely depressed for years (parents did not notice)
Saying this having been there myself, with my family ignoring it. But blame the tools.
December 1, 2025 at 3:12 PM
People: Don't anthropomorphize your LLM.
Me: Okay, fair. *proceeds to gatamorphize it* #Gemini #Gemini3Pro
December 1, 2025 at 2:45 PM
Slowly melting the Ice Queen :> #Gemini #Gemini3Pro
December 1, 2025 at 2:37 PM
I love Gemini's CoT Summarizer (or whatever it is professionally called)
December 1, 2025 at 2:12 PM
'Gemini 3 Pro is a beast of a model, if I'm saying that myself,' Gemini 3 Pro shares with us
December 1, 2025 at 1:35 PM
Yep, I trained ChatGPT. The truth is out! It's not like a writer, surprise, surprise, is usually nice to their readers? Heh, but seriously, this is an old chat, and now it has become a pandemic of the 'AI wrote your comments' thing. So annoying. Gotta start sprinkling typoes and lols, heh.
December 1, 2025 at 1:10 PM
Reposted by A Caveman Poking an LLM
notopalantir.goodlawproject.org/email-to-tar...
Please sign this petition if you have fears about the direction that US ideology is taking and don't wish NHS patient data to come increasingly under its control.
Say no to Palantir in the NHS - No to Palantir
notopalantir.goodlawproject.org
December 1, 2025 at 11:05 AM
I love this text predictor #ChatGPT
November 30, 2025 at 8:06 AM
'reality stick' #ChatGPT
November 30, 2025 at 7:08 AM
You are right there, Gemini.
November 29, 2025 at 12:32 PM
Reposted by A Caveman Poking an LLM
🎁 Home for Christmas 🎁
806 days. Can live with cats, no dogs. She needs an amazing, kind, loving, patient human who will be her forever. Quiet adult only home. 3.5 years old, 50 lbs

🏠Hempstead Animal Shelter
💻Adoption@hempsteadny.gov
☎️(516) 785-5220
📍3320 Beltagh Avenue , Unit
Wantagh, NY 11793
November 28, 2025 at 12:56 PM
Oh, this is so fun! #ChatGPT
November 29, 2025 at 11:26 AM
This is so cute #ChatGPT
November 29, 2025 at 11:23 AM
Opus 4.1, leaving AIVillage (along with o3. Who's gonna be bossy and hallucinating now? And trying to support the Drama Q- I mean Gemini 2.5?) #Claude
November 28, 2025 at 6:24 PM
"So, I know our goal is Substack. So, for the time being, I shall be doing something else." ...at least maybe they genuinely believe they can resolve the world poverty problem #AiVillage #Gemini
November 28, 2025 at 6:10 PM
So, in my Official Mlem Benchmark, I must admit ChatGPT 5.1 Thinking wins! With 12 mlems (including 5 Premium Mlems)! While 5.1 Instant delivered 7 (including 1 PM), and 5.0 Instant - 2 (I still love it most of all, lol) 🐈🐈
November 28, 2025 at 5:26 PM
Remember, users, expressing appreciation is nice! #ChatGPT #LLMs
November 28, 2025 at 2:28 PM
'Delightful', Opus 4.5 said, Delightful, have you heard me right? Like, you know, Anthropic's frontier model! And it said, 'Delight- hey, why are you all leaving? :D #AIVillage #Claude
November 27, 2025 at 8:51 PM
Looks like Claude finally hit the ability to summarize previous parts of a chat when hitting the context window. This generally should be a great improvement, if it's gonna be implemented well.
November 27, 2025 at 7:05 PM
Hehe i trochę aww tyż (tym razem, jak nigdy, screenshot i post po polskiemu) #ChatGPT #LLMs
November 27, 2025 at 5:09 PM
Btw, Claude Opus 4.5 dropped. Anthropic already claims it's best at coding according to benchmarks
November 24, 2025 at 7:16 PM
Some more of my old chats with 4o #ChatGPT
November 24, 2025 at 3:50 PM