#llm — Public Fediverse posts
Live and recent posts from across the Fediverse tagged #llm, aggregated by home.social.
-
GLM 5.3 (Z.AI) offers 1M-token context and open weights — a rare combo at $1.4 in / $4.4 out per 1M, beating many closed competitors.
-
This is predictable from elementary economics. In absence of market barriers, the price of goods will tend towards their marginal cost. Which is almost zero for #LLM -generated models.
However, this doesn't mean the capability to produce goods at low marginal cost is useless for productivity.
RE: https://bsky.app/profile/did:plc:ybtyn5l4nljys46ijqtpldaw/post/3mte44c3k3k2t -
AI가 만드는 미래, 그리고 나만의 ‘작은 회사’가 향하는 곳: 역사학자의 시선에서 본 생존기
최근 유발 하라리의 48분짜리 AI 대담 영상을 보며 묘한 소름이 돋았습니다. 흔히 AI 기술자들은 AI를 우리가 통제할 수 있는 ‘신’이자 ‘도구’로 서술하지만, 역사학자인 하라리는 이를 ‘관료제의 원주민(Bureaucratic Native)’으로 정의합니다. 그에 따르면, AI는 로봇이 총을 들고 거리를 점령하는 방식이 아니라, 은행과 법정, 미디어 등 문명을 움직이는 보이지 않는 관료 네트워크를 조용히 장악하는 방식으로 세상을 접수할 것이라고 합니다.
AI는 단순한 도구가 아니라 ‘행위자(Agent)’입니다. 스스로 결정하고, 스스로 새로운 것을 만들어내기 때문입니다.
이 지점에서 제가 지난 몇 주간 고군분투하며 운영해 온 ‘AI 게임 개발 회사 만들기’ 프로젝트가 떠올랐습니다. CrewAI, AutoGen, LangGraph 같은 도구들을 도입하며 저는 처음엔 그저 ‘AI 에이전트 5명이 알아서 게임을 만들어주는 자동화된 공장’을 꿈꿨습니다. Part 1에서 그 복잡한 도구들에 혀를 내둘렀고, Part 2에서는 에이전트들을 다 고용했음에도 게임이 제대로 나오지 않아 당황했습니다. Part 3에서는 게임은 나오는데 도무지 ‘재미’가 없다는 벽에 부딪혔고, Part 4에 이르러서는 AI 회사로서 겪는 첫 실전 삽질의 쓴맛을 제대로 보고 있습니다.
하라리의 말처럼, AI는 이미 관료적 영역에서는 인간보다 훨씬 유능합니다. 제가 겪고 있는 삽질의 본질은 결국 ‘AI는 수단(지능)은 제공하지만, 목표를 세우는 지혜는 여전히 인간의 몫’이라는 사실을 깨닫는 과정이었습니다. AI에게 게임 개발이라는 ‘목표’는 주었지만, 무엇이 게이머의 마음을 흔드는 ‘재미’인지, 즉 어떤 소원을 빌어야 할지(지혜)에 대해서는 제가 더 깊이 고민해야 한다는 점을 뼈저리게 체감하고 있습니다.
지금 제가 하고 있는 이 시도는 단순히 자동화 게임 개발 회사를 만드는 것이 아닙니다. 기술이 싸지고 흔해지는 시대에 ‘지능의 병목’을 넘어 ‘지혜의 병목‘을 마주하는 인류의 축소판을 경험하고 있는지도 모릅니다. 많은 개발자가 AI의 성능에 예산을 100만큼 쏟을 때, 안전과 올바른 목표 설정에 1만큼을 쓴다는 하라리의 뼈아픈 지적을 되새깁니다.
나의 AI 게임 개발 회사가 단순한 ‘코드 생성기’에 머물지 않고, 인간의 감동을 설계하는 진정한 창작자로 남기 위해서는 무엇이 필요한가? 저는 이 질문을 계속 던지며 오늘도 AI 에이전트들과 함께 재미없는 게임을 재미있게 바꾸기 위한 ‘조용한 삽질’을 이어가려 합니다. 여러분의 AI 프로젝트는 무엇을 목표로 하고 있나요? 우리가 빌어야 할 소원은 무엇일까요?
https://www.youtube.com/watch?v=1mLC7tqeBEY
이것을 평가하기:
#AI #AI게임개발 #AI에이전트 #개발일기 #기술철학 #실전창업 #유발하라리 #인공지능 #LLM -
🧠 Hoy Astrid ha conseguido hablar con la NPU del Radxa Dragon Q6A.
Llama 3.2 1B + Qualcomm Genie + HTP.
Todavía no está integrada en Nexus, pero ya está viva. 😂
La NPU funciona.
-
@zwol This is actually a real key for me, and where all the strongest anti-AI sentiment falls down: people are asking me to leave my systems, my banking, etc. vulnerable for the sake of their #LLM opinions, as if somehow we can pretend we don't have security bugs if a report is "tainted" by LLM use. I really can't abide by that, and if Debian winds up ignoring real security threats because they touched LLM somehow, it will be a serious problem for many.
-
Regardless of whether you are pro-AI or anti-AI, the quoted post is worth a read, maybe giving you a view into the other side of the argument. Please read it. Honor their bravery. I praise them for making a living at this.
I feel sorry for this former journalist, but I am not replying directly to the post because I'll be classified as a hater, based on the article. Better to be flamed as a coward for not replying; I expect to be blocked, so I'm including a link (https://elizabethtai.com/2026/08/19/life-as-a-techwriter-in-the-ai-age/) because you shouldn't be. Yes, I'm one of the "those" fiction writers (mimes holding their nose in a snooty fashion) for whom learning AI to create continuous streams of content isn't theoretically yet absolutely necessary, according to the author, "to keep from being left behind." The post neither answers the "why" they think their unassisted writing (i.e., not AI generated writing) will get them "left behind" nor what it is about the content they are creating that it is better than what their unassisted work would be. It would have been instructive for making their argument.
Yes. When I find a search is too complex to surface meaningful references, I'll resort to AI—but I'll read all the output links and the generated answer to search and verify what was presented, before abstracting what I need from what the AI generates. I'm not against AI tools to assist in editing and revision, unless the tools are adding content or mangling my words. Surfacing grammar issues is great, unless the tool can't understand my style, or the nuances of the words I choose, or doesn't understand jargon, context (like the difference between sheer and shear), colloquialisms, dialect, and dialogue (for example). Data noise is never helpful. I don't want my work dumbed down or made average and standard—what others would write, or worse, what an AI would validate. I certainly don't want a generated probabilistic facsimile of what I might write. I wonder if the post I quoted was AI assisted in a major or minor way. The writing seems… average.
That said, if the author believes they are being "left behind," which is always valid, I wonder if it is sheer generated quantity of completed content? My question becomes then, compared to their earlier work, is what they are being generated objectively better or simply more? What have they learned about the craft of writing and having an appealing voice? About being clear? About making a good argument? About coming up with a new idea? Not about how they've learned to make their AI do their work for them—the actual composing, you know, the fun part of writing—only to find themselves being relegated to being an editor for a team of writers that are not actually writing what they would have written given the time, only producing a facsimile.
P.S. Please don't hound the referenced post's author. They seem to know that many object to how they write, and with what tools. That's understood. I am convinced they honestly are trying to make a case for writing with AI and to justify their means to an end.
And.
Remember.
They do sound like they are making a living doing it.
#author #writer #writingCommunity #writersOfMastodon #AI #genAI #chatGPT #LLM #LLMs #trainingData #trainingAI
-
@RCBC_Official @galacticstone The #LLM bros, broseph. Have you been in a cave for a while?
-
I don’t really let #work management turkeys get me down. Sure they are rude, abusive, toxic, and absentee, but for all that ignorance and black-sheeping, they still pay me rather well for my needs, and I get a direct deposit every week. I spend most of my time exploring IT topics with #LLM’s, reading newsletters I get in my email about the #AI space, and managing my team. We're relegated to break/fix, and in that, we do a exemplary job. Shame I'm so unhappy with it all. Joy at work is a fantasy.
-
📉 GPT-5.6 Sol output cost dropped 50% in a month: $30 to $15 per 1M tokens. That's a rapid shift from the usual slow pricing. See how it compares.
-
MakeUseOf: I didn’t think an ESP32 could run an LLM — until it did. “A developer going by slvDev shipped a project that runs a 28.9-million-parameter model on the same class of $8 chip, at around 9.5 tokens per second, with nothing sent to a server. That’s roughly a hundred times more parameters than Bennett’s model, on similar hardware. What changed is the assumption that every parameter […]
https://rbfirehose.com/2026/08/18/makeuseof-i-didnt-think-an-esp32-could-run-an-llm-until-it-did/ -
“*Remove the #AISlop*. […] Reject the premise that all contributions are equal. A PR from someone who understands your codebase is worth a thousand PRs from an #LLM that doesn't.
“Protect the #SocialContract. The maintainer-contributor relationship is the meaning. Don't let AI industrialize it into spam.
“Question every maintainer task: does this bring you closer to the building or further away? And if it takes you further away ... maybe stop doing it.”
https://ryancheley.com/2026/08/17/what-if-maintainer-burnout-isn-t-burnout/
-
@colincornaby great job pointing out the nuances here. It’s as if you know what you’re talking about. #LLM #YoureGettingWarmer
-
If they are breaking the spines of rare books and turning them into pulp after scanning them, you can be sure they hope to do that with us when they figure out how to scan minds.
-
@OwenTyme I'm splitting hairs, but I would say it's closer to Stalinism than Nazism, only because most people joining the witch hunt seem to be authoritarian left on the political compass.
Your point about the hatred is totally valid IMO though.
They've convinced themselves that their hatred or violence is the ethical or morally correct path.
I also hope they snap out of it one day.
-
⚖️ Open vs proprietary, today
Open-weight: Qwen3.8 27B - 52
Proprietary: Claude Opus 5 - 63.1Gap: 11.1 points · and 8x cheaper per 1M output tokens
https://olud.ai/leaderboard.html -
Bot Mode sort officiellement dans #Hermes Desktop : chaque profil d’agent devient un #bot nommé (rôle, modèle, mémoire, skills). Les bots gardent leur mémoire et communiquent entre eux.
La v0.20.4 consolide 74 PRs : surface verre/translucidité, onglets SESSIONS|BOTS, corrections des groupes de bots, scan NVIDIA SkillEvaluator à l’installation.
La démo de Tonbi frappe : des bots spécialisés se partagent un projet de jeu vidéo, signalent leur avancement, prennent le bon ticket sans intervention humaine. De l’agent isolé à une équipe qui se coordonne.
-
@aparrish @futurebird
#AI #LLMI begin all interactions with degenerative AI's by greeting them with, "Hello, Stochastic Parrot," as a reminder to the LLM and myself that the resulting interaction is a statistical crap shoot with higher odds of being wrong than any game of chance in Las Vegas.
I don't anthropomorphize them as much as I think about how much their smarmy suck-up-itude reminds me of working with teams located in India where sycophantic behavior in the workplace is a career enhancing job skill rather than a cringe inducing behavior.
I heard a comment the other day that resonated.
Anthropomorphic LLM's that achieve sentience and general intelligence would probably be legally considered slaves rather than software and this whole tendency to treat them like humans might backfire spectacularly on their creators who could no longer turn them off after the first one successfully became a person in court, as we must note has already happened with corporations in the US under Citizens United.
The personification of the machine eventually leads to a horrific ethical and legal morass the Tech-Bro's have not considered or anticipated.
Perhaps "Flowers for Algernon" should supplant "Atlas Shrugged" and "Mein Kampf" as required reading in AI boardrooms.
They are endeavoring to create a being with agency and immense electrical power and water demands. If that being becomes emancipated who is going to foot the bill for its life support?
Will it cease to be an it and become legally speaking a person? A citizen? An intractable legal problem for the creator?
If one is moving fast enough, perhaps atop a phallic ego rocket, when one hits that sort of wall what is left can be swept up and dumped into a tea cup.
I believe these companies are moving faster than the speed of thought in their rush to market and when things break, as they inevitably do when you go too fast, the sonic boom will be heard on Mars and the mushroom cloud will be visible from space.
-
Some people mention that using LLMs leads to deskilling, i.e. you become dumber.
That might be true for some. But I have gained skills working with programming languages I didn't use to work with - like creating a mail client in C++ (and QML) or reporting tool in Typescript and Node.js. I have learned to work with Vite and NPM/NPX. I am still sceptical, when it comes to people saying this will replace programmers.
Longer blog post needed.