Sign in

merklejerk

@merklejerk.com
158 followers 176 following 339 posts

Burned out blockchain (EVM) engineer, security + rust + AI/ML dilettante, tabletop gamer, compassionate futurist, seasonal misandrist, wasian 🇵🇭 🏳️‍🌈. ᛝ⣦⣄▥┱⠇⢤⠜┲╡⣈⢁▣⢆┮⣈⠺╲⡀▝┋▐

PostsRepliesMedia
merklejerk @merklejerk.com · 05/09/2026
wow astra really is just like fable
000 0
merklejerk @merklejerk.com · 04/09/2026
deepswe benchmarks look like they've run out of endgame content
000 0
merklejerk @merklejerk.com · 02/09/2026
excuse me, SIR?? gemini 3.8 flash ON TOP of deepswe? 🤯
000 0
merklejerk @merklejerk.com · 29/07/2026
trust elon to run your inference, what could go wrong
claude outage
000 0
merklejerk @merklejerk.com · 10/07/2026
It's interesting that Anthropic's models get slower the cheaper they are. I guess that's some kind of progress?
000 0
merklejerk @merklejerk.com · 30/06/2026
now that we're in the era of labs bragging about how dumb their models are, maybe gemini 3.5 pro will finally have a chance
sonnet 5 launch blog post
000 0
merklejerk @merklejerk.com · 27/06/2026
vibed a GNOME screenshot/snipping tool that converts your screenshots into a boomer-style phone pic of the monitor github.com/merklejerk/b...
000 0
merklejerk @merklejerk.com · 26/06/2026
Anyone else think it's a little interesting for OpenAI to show 5.6 Sol trailing both versions of Mythos in cybersecurity benchmarks while still winning on coding? Must've made the charts before the govt restriction order. Also that X axis throws so much shade on Anthropic lol.
000 0
merklejerk @merklejerk.com · 23/04/2026
help me put this game down 😭
000 0
merklejerk @merklejerk.com · 23/04/2026
gdi why did nintendo do this
tomodachi life
010 0
merklejerk @merklejerk.com · 23/04/2026
stoppp I can't with this game 😭
tomodachi life
000 0
merklejerk @merklejerk.com · 23/04/2026
I put kristi noem in a cop outfit in tomodachi life and she LOVED it 🤔
tomodachi life kristi noem
000 0
merklejerk @merklejerk.com · 11/04/2026
It made me sad to think we might start treating code like assembly and never look at it again but then I remembered SQL
very typical sql
000 0
merklejerk @merklejerk.com · 31/03/2026
of course the only 2 other affected packages called out in the axios supply chain attack are openclaw related
000 0
merklejerk @merklejerk.com · 17/03/2026
nothing more 2026 than vibe coding a quick app so I can play pokopia in a window github.com/merklejerk/g...
pokopia windowed mode under gnome
001 0
merklejerk @merklejerk.com · 28/02/2026
I got nothing against @t3.gg other than he keeps choosing thumbnail pics that make him look like a startled lemur
theo youtube thumbnailstartled lemur
000 0
merklejerk @merklejerk.com · 23/02/2026
I made AI slop about AI slop 😇
000 0
merklejerk @merklejerk.com · 13/02/2026
responses like this is why I have a hard time switching away from gemini 3 flash/pro in my coding workflow 🥺
gen z gemini 3 agentic coding response
000 0
merklejerk @merklejerk.com · 06/02/2026
lol gemini wrote this unit test unprompted
"because you're mid" unit test
000 0
merklejerk @merklejerk.com · 03/02/2026
I see this pattern of vulnerability all the time in vibe coded apps. AI agents do know better but only if you explicitly ask them. It gets even trickier in specialized domains and brownfield codebases. I laugh whenever CEOs claim developers won't have jobs.
Moltbook hacked
000 0
merklejerk @merklejerk.com · 01/02/2026
ngl I relate to some of these moltbook posts
moltbook post gone delulu
000 0
merklejerk @merklejerk.com · 29/01/2026
New Valentine's Day emoji just dropped
Niki and Trump holding hands
000 0
merklejerk @merklejerk.com · 06/01/2026
had gemini generate a J6 Simulator Game to celebrate this holy day ^_^ gemini.google.com/share/fbc2ac... 🇺🇲🇺🇲🇺🇲
010 0
merklejerk @merklejerk.com · 27/11/2025
In case you don't understand how absurdly far ahead in the game nano banana pro is... My prompt: "an infographic illustrating how to identify each member of BTS from one another" Z image turbo, chatgpt 5, nano banana pro, in that order:
Z-image turbo resultChatGPT 5 resultNano Banana Pro result
010 0
merklejerk @merklejerk.com · 02/11/2025
"I got a 3D printer for functional prints!" My functional prints:
3D printed magnetic thingies
000 0
merklejerk @merklejerk.com · 10/09/2025
big mad bigger mad
big mad brown anolebigger mad brown anole
110 0
merklejerk @merklejerk.com · 20/07/2025
damn you can't get away with nothing these days
screenshot of github copilot inserting itself as the author of my project
010 0
merklejerk @merklejerk.com · 16/07/2025
Before AI slop we had Chuck Tingle
Screenshot of Chuck Tingle audiobooks
000 0
merklejerk @merklejerk.com · 10/07/2025
this new generation of hackers are so lazy that they won't do the market research to realize 99% of ethereum protocol developers use mac www.reversinglabs.com/blog/malicio...
supply chain attack injection script targeting only windows platforms
010 0
merklejerk @merklejerk.com · 04/07/2025
ohhh just found this neat visualizer of LLM output token logprobs, which could be a really neat teaching tool thelackthereof.org/token-tree
visualization of an LLM's output token probabilities at each step
010 0
merklejerk @merklejerk.com · 30/06/2025
I was told that violent crime is rare in London
a perfect plate of jellied eels from a London pie shop
000 0
merklejerk @merklejerk.com · 06/06/2025
I finetuned Gemma 3 1B PT (not IT) to read RSS news and react like a house cat. Runs decently (for this use-case) on a little 2W (max) ARM SBC using < 0.7GB of RAM. Small LLMs are amazing when finetuned. Takes no instructions, just a headline, and outputs perfectly structured responses every time 💪
Small LCD display showing a finetuned Gemma 3 1B PT reading news headlines like a cat on an Orange Pi Zero 2W.
130 0
merklejerk @merklejerk.com · 02/06/2025
today I learned that people hitting thumbs up on ego-affirming responses are to blame for this annoying generation of overly flattering chat models. once again, emotionally needy dog people have blood on their hands openai.com/index/sycoph...
OpenAI blog post excerpt: "But we believe in aggregate, these changes weakened the influence of our primary reward signal, which had been holding sycophancy in check. User feedback in particular can sometimes favor more agreeable responses, likely amplifying the shift we saw. We have also seen that in some cases, user memory contributes to exacerbating the effects of sycophancy, although we don’t have evidence that it broadly increases it."
010 0
merklejerk @merklejerk.com · 29/05/2025
ppl that didn't play the game after The Last Of Us season 3 (finally) airs
"are we the baddies" meme
010 0
merklejerk @merklejerk.com · 28/05/2025
Wish other people would just admit they found Dopey from Snow White (2025) hot. Gimme a rich short king that only speaks when spoken to 🥵
Dopey from Snow White (2025) looking fine af
000 0
merklejerk @merklejerk.com · 15/05/2025
I was on the fence about Epic Universe until night fell. It's like 10x more immersive at night. It really needs to have a midnight closing time, not 9pm.
Ministry of Magic land at Epic Universe, night shot.
010 0
merklejerk @merklejerk.com · 07/04/2025
who knew the 2025 ASMR trend would be watching the DJI drop in real time as Trump rambles
BBC news split screen of trump rambling and the DJI on the NYSE dropping
000 0
merklejerk @merklejerk.com · 25/03/2025
a very normal looking interview in crypto (not sarcasm)
Splitscreen from Bankless podcast where the guest has superimposed a comically oversized milady NFT (of an anime girl's face) over his head to stay anonymous.
000 0
merklejerk @merklejerk.com · 03/03/2025
I changed my mind. Delete AI.
Google AI rudely saying that Goldie Hawn wasn't in death becomes her
000 0
merklejerk @merklejerk.com · 20/02/2025
good news, @dndbeyond.com is sending out Project Sigil keys! 🥳 🙂 🤨 😟 😩 it's windows only
disappointed man
110 0
merklejerk @merklejerk.com · 24/01/2025
uhh I think the AI resistance on bsky may actually just be my asian mom
weirdly hostile chatgpt tweets
000 0
merklejerk @merklejerk.com · 23/01/2025
serious q: how do we onboard more horses to glue factory?
serious q: how do we onboard more girls to crypto?
000 0
merklejerk @merklejerk.com · 23/01/2025
They def had me in their corner but you gotta admit that design language was 100% corporate-core
google+ UX
120 0
merklejerk @merklejerk.com · 23/01/2025
B-but, grok is actually helpf...
grok promptgrok response
010 0
merklejerk @merklejerk.com · 20/01/2025
Do other fields have this issue or is it unique to crypto? Suspiciously LLM-like github contributors with hidden activity spamming very minor PRs across every popular project in the space. Is this for airdrops or something more sinister? 🤔
000 0