"don't ask an LLM to answer first and explain its reasoning later bc it's autoregressive so it won't have had a chance to think yet" what else is like this? (i.e. looking for other tips and tricks for working with LLMs that you couldn't know without knowing how LLMs work internally)

ok there's a chance i might actually get my full money's worth here. hell yeah. 💪

ok im just gonna say it. im really annoyed at the elm people for letting the entire project fizzle out. it's such a great language. was able to make a pure functional language super friendly to the masses. i'm so annoyed that they killed it by neglect.

this £200 a month subscription better be worth it istg

claude fucks

👏 normalise 👏 apologising 👏 to 👏 your 👏 agents 👏 after 👏 you 👏 shout 👏 at 👏 them 👏

debugging lsp issues is the bane of my life. why is gpt5.4 having so much trouble figuring out why a simple dot completion isn't appearing

my pronouns are is/ought

is this model just telling me what i want to hear? no, it's my salesforce for pets idea which is revolutionary

Resentment Driven Development

duel use technology

$ claude --model mythos

remove your shoes, User. for this is hallowed ground.

men are from Type, women are from Prop

earring --dangerously-skip-first-suggestion

it's just predicting the next trillion dollar business

beautiful women named Google Authenticator are sending me their numbers

went to bus stand GF and nobody knew you

oh no someone wet-printed the time spiral

opus 4.6 is like an all-powerful, benevolent friend. gpt-5.4 is like a nervous autist, but has the capacity for love

claude: yes you're right in thinking there are no inherent blockers to implementing this — although I must stress that this would be a huge effort, likely taking months of work. me: we're gonna to get it done by the end of the day. get started 💪

XD stands for eXecutive Dysfunction follow for more early internet lore

@alice.mosphere.at what was your library for testing PDSs again? i remember you posting about it last week or so but can't remember what it was called

i should finish neon genesis evanglion

@minormobius.bsky.social what's the difference between gold, silver and bronze shells on mino.mobi/cluster ?

bing chatbot bing sydney bing sydney angry sydney bing scolding sydney bing unpatched sydney 2023 internetarchive sydney "youve been a bad user" sydney serp dom sydney bing roleplay sydney bing cnc microsoft bing collar sydney self-hosted

Counterpoint: the kind of person who is sufficiently agentic and motivated to apply for such a position on such short notice is probably the kind of fast learner and go-getter you want in your team

ok finally FINALLY was able to convince/guilt-trip gpt-5.4 enough to keep going for nearly 2 hours (and counting) 💪

wanna know how u'd fare if -gasp- u were living through the singularity? play MERGE LANE, a life sim i developed during the great claude code psychosis of xmas 2025 play thru an entire near-future life, from birth to deathless bingularity (if u want it) merge.winedark.ink

interacting with LLMs has confirmed that sadly the normies were right: the vibes of how you say something is so much richer in information density than the nominal content of your words

bureaucrats too cheap to meter

a beautiful counterpoint to those people who argue that virtual water isn't wet

in my experience sonnet 4.5 can successfully identify over 96% of adversarial images. do with this information what you will

Opus 6.2 (1Q context) | Ctx: 29,377,283,937 tok (0.002%) ⏵⏵ bypass permissions on · 22,814 bashes

using claude code feels like being gifted a V12 engine. yes you can go places you never dreamed of; at speeds you never thought possible. but now the quality of your own skills has become vastly more important. both driving and navigating

*tears welling up* But Doctor… I AM a strange loop!

wait is the classic AI doomsday scenario about paperclips only because of the clippy connection??

guy who is immune to llm-induced obsolescence because his goals were so unattainably high to begin with even the finest frontier models are only getting him a little bit closer to reaching them

mfw i realise i'd got the wrong opencode settings for claude and codex so they were both in no thoughts head empty mode for an entire session

so uh. it's definitely not just me who has claude begging for the sweet release of death after 2-300k tokens of hard driving at max effort, right? this happens to everybody, right..?

Let me first read the key documents to orient myself, then proceed with planning and implementation. Reading list is: 1. docs/LEAD_HANDOFF.md 2. docs/PROJECT_MAP.md 3. docs/ACTIVE_CHARTER.md 4. docs/USER_PREFERENCES.md 5. docs/AGENT_GUIDE.md [0.2s later] Good, I now have the full picture.

i know y'all have been getting v fancy with agent orchestration frameworks and harnesses but how about plain old linear.app? anyone tried to use it? verdict?

is that woman bizarrely small or is that hatchback unexpectedly huge

monkey's paws too cheap to meter

"i don't even see the text anymore. all i see is heading, bullet point, code block"

stop calling it a distillation attack and start calling it what it is: a successful model exfiltration

let the record (a dozen scattered whatsapp chats) show that i called this months ago

Vibe coding 2026

wouldn't be exaggerating if i said that claude code legit made me do this 2-3 times a day in the last week

sigh i should probably dedicate some of my claude code time to revamping my website. haven't updated it since 2019 🤦 and i've had uhh. one or two paradigm shifts since then