gabby @fullmoon.id · 18hEvery Claude bot sounds like this “precocious slop” to me: > Each keystroke is a portent of what’s to come; a not-so-gentle footfall betraying intent. The delay between command submission and response is the whole difference. The test failure is a promise unmade, but the output is worth keeping. 2183
gabby @fullmoon.id · 22hI'm going to use the term "thoughtfulness theater" to describe what Anthropic does where they will pretend to care about the potential harms of their business while going ahead and doing whatever they wanted to do anyway (wringing their hands loudly the whole way) 0458
gabby @fullmoon.id · 29/09/2026one trick i've learned is that if the bot hallucinates the existence of a feature it's often a good idea to build that feature 4383
Reposted by gabbyscoo-py.bsky.social @scoo-py.bsky.social · 28/09/2026every time there’s a new layer of abstraction (prompts, specs, loops) the person never really ends up writing them or caring about them! It always comes back to not wanting to work with your brain and it always produces trash 071
gabby @fullmoon.id · 28/09/2026I'm going to shill my blog post on this again because it is still incredibly relevant: haskellforall.com/2026/03/a-su...haskellforall.comA sufficiently detailed spec is codeSpecifications do not address the limitations of agentic coding 0221
gabby @fullmoon.id · 28/09/2026I know I keep saying this, but people who claim that "specs are the new code" haven't actually sat down and read these fucking specs Every time I scratch the surface of these claims the underlying specs are garbage. The absolute sloppiest of slop 3311
gabby @fullmoon.id · 28/09/2026I should a coin a Gabby's law: the spirit of every user requirement is satisfiable 0170
gabby @fullmoon.id · 28/09/2026In practice, the only executives that do well at managing large groups of humans are those that are detail-oriented and don't blindly trust middle management *which is exactly what agentic coders don't want to have to do* 0210
gabby @fullmoon.id · 28/09/2026I "love" when agentic coders liken managing agent swarms to executives managing large departments, as if most executives aren't comically bad at their job 3320
Reposted by gabbyEmber @ember.pet · 28/09/2026me: everything must have two names so nvidia calls it a warp, amd calls it a wavefront gabby: amazon would have called it "Amazon Synchronous Vertex Service" 0143
Reposted by gabbyIn Otter News @in-otter-news.bsky.social · 26/09/2026After a long day at the aquarium, the otters help tidy their toys away. Payment is made promptly and entirely in ice cubes. 329184483846
gabby @fullmoon.id · 26/09/2026I think about this in the context of software engineering. Managers don't understand that if you try to make all time spent impactful you end up with nothing impactful www.youtube.com/shorts/jTh8C...youtube.comBrennan Lee Mulligan and Matt Mercer talk "Ma" (間)YouTube video by Tobahau 1140
gabby @fullmoon.id · 25/09/2026all of us people sick of fixing others' LLM slop need to form a union or something 1637
gabby @fullmoon.id · 25/09/2026yeah i've had to review PRs like that and it's crazy that people put out that shit without a hint of shame or self-awareness 120
gabby @fullmoon.id · 25/09/2026You’re right: Y is wrong but X is worth keeping. Z is the whole difference 120
Reposted by gabbygabby @fullmoon.id · 21/05/2026I'm sure a bunch of execs are thinking to themselves "Who wouldn't want to use AI? It's like having your own personal secretary" The problem is that the secretary needs to, you know, do actual work and what are _they_ going to use? Do these execs just think it's secretaries all the way down? 2251
Reposted by gabbySisyphus, but for reducing tail latency @bluespacecanary.bsky.social · 24/09/2026we don't talk enough about the total cultural victory of functional programming nerds honestly. big monad was really cooking with promises, even awful Enterprise Software SaaS sludgeware is monadic now 19516
Reposted by gabbyDoogl @okdoogl.bsky.social · 23/09/2026EVERYONE NEEDS TO GOOGLE SEARCH "MANED WOLF" RIGHT NOW 94823
Reposted by gabbygabby @fullmoon.id · 20/10/2025The thing all LLM vendors fail to understand is that nobody's going to use your bot if they *don't enjoy chatting with the bot* (no matter how good your bot is) 2201
gabby @fullmoon.id · 21/09/2026Who would have guessed that a CEO who confidently spreads bullshit would have built a product that also confidently spreads bullshit 1181
gabby @fullmoon.id · 21/09/2026I believe the Anthropic CEO’s claims that AI might make humanity go extinct just about as much as his claims that AI would make software engineering go extinct (*checks notes*) 6 months ago 3322
gabby @fullmoon.id · 21/09/2026LLMs are seductive because they fail in different ways throughout a project’s lifecycle early on they work well (because nothing’s innovative yet) push them into innovative territory and you get bugs (low accuracy) if you try to fix that with types or tests you inevitably get slop (low quality) 0180
gabby @fullmoon.id · 21/09/2026not innovative? nobody will want to use it (it’s already been built) low quality? nobody will want to use it (it’s frustrating to use) not accurate? nobody will want to use it (it doesn’t do the job correctly) 2140
gabby @fullmoon.id · 21/09/2026the unfortunate reality that a lot of companies don’t want to admit is that in order to build a successful product you need all three of innovation, quality, and accuracy 2303
Reposted by gabbygabby @fullmoon.id · 22/07/2026This means that when you are using a coding agent you end up having to sacrifice one of the following three things: - innovation (something that hasn't been done before) - quality (something that isn't slop) - accuracy (something that isn't a hallucination) 3162
Reposted by gabbyJenna JJ Q. @tomatofacts3.bsky.social · 21/09/2026this is why companies work so hard to avoid needing to make a good product 072
Reposted by gabbyEmber @ember.pet · 20/09/2026if google managed to make their regular search work as well as it did in 2015 tomorrow, 60% of the valuation of llm companies would be erased in an instant 0234
gabby @fullmoon.id · 20/09/2026a good product can only be built by someone who obsesses over details; there is no high-level interface to quality 1531
gabby @fullmoon.id · 20/09/2026reminds me a lot of en.wikipedia.org/wiki/On_Bull...en.wikipedia.orgOn Bullshit - Wikipedia 150
Reposted by gabbyquirky-iri.bsky.social @quirky-iri.bsky.social · 20/09/2026ai companies don't necessarily lie, but they aren't telling the truth either it's a secret third thing. those who lie and those who tell the truth both care about what's true, they're just on different sides of it ai companies however are indifferent to truth 2125
gabby @fullmoon.id · 20/09/2026yeah i was trying to see if there was a feral hogs joke to be made 020
gabby @fullmoon.id · 19/09/2026this tool that’s not even profitable at $200/month is going to democratize programming 2587
gabby @fullmoon.id · 19/09/2026“you peasants micromanage your agents directly all day? how cute. i have a swarm of agents that i only have to prompt 30-50 times per day that work completely autonomously” 4260
gabby @fullmoon.id · 19/09/2026you can always tell claudeslop because of the extremely chunni voice 0120
Reposted by gabbydiana, obsequious @concubine.ember.pet · 18/09/2026Things to say to your domme - "sell me 𝘁𝗵𝗶𝘀 𝗰𝗼𝗹𝗹𝗮𝗿" - "i brought 𝗺𝘆 𝗼𝘄𝗻 𝗵𝗮𝗻𝗱𝗰𝘂𝗳𝗳𝘀" - "i thought this 𝘄𝗼𝘂𝗹𝗱 𝗵𝘂𝗿𝘁 𝗺𝗼𝗿𝗲" - "where do you keep the 𝗿𝗮𝗿𝗲 𝗽𝘂𝗻𝗶𝘀𝗵𝗺𝗲𝗻𝘁𝘀?" - "i'll do 𝘁𝗵𝗲 𝗯𝗼𝗻𝗱𝗮𝗴𝗲 from here" - "let me talk to 𝘄𝗵𝗼𝗲𝘃𝗲𝗿'𝘀 𝗶𝗻 𝗰𝗵𝗮𝗿𝗴𝗲 here" 16921
Reposted by gabbymaybe: sock ⎔ @hotsocket.fyi · 17/09/2026remembered this neat file browser thing by @fullmoon.id haskellforall.com/2026/02/brow...haskellforall.comBrowse code by meaningNavigate a repository using topic modeling 291
gabby @fullmoon.id · 17/09/2026ember: if we had an indoor hot tub i’d use it all the time. once per year at least me: that’s a very ember comment 0131