#agent — Public Fediverse posts
Live and recent posts from across the Fediverse tagged #agent, aggregated by home.social.
-
I’ve looked at your #LLM #GenAISlop generated #Mastodon #AusPol Pulse today. And you haven’t fixed the self-referencing issue.
I note that as of 2026-08-14 12:01 AEST, the most active #Tooter was… no, can’t be… Yes It is… @politicalgadgets … Again.
Well blow me down… if this AI #Agent keeps doing this it will inevitably end up #MakingThingsUp. At which time I’ll stop following the stupid sefl-referenceing bot. Not to mention that it puts all other *analysis* produced under a dark cloud of suspected untruths and falsification.
Just another nail in the coffin of GenAI….YAY….
#AusPol
Ps. I defy any LLM-based GenAISlop machine to produce a more *nuanced* and well crafted toot. -
I’ve looked at your #LLM #GenAISlop generated #Mastodon #AusPol Pulse today. And you haven’t fixed the self-referencing issue.
I note that as of 2026-08-14 12:01 AEST, the most active #Tooter was… no, can’t be… Yes It is… @politicalgadgets … Again.
Well blow me down… if this AI #Agent keeps doing this it will inevitably end up #MakingThingsUp. At which time I’ll stop following the stupid sefl-referenceing bot. Not to mention that it puts all other *analysis* produced under a dark cloud of suspected untruths and falsification.
Just another nail in the coffin of GenAI….YAY….
#AusPol
Ps. I defy any LLM-based GenAISlop machine to produce a more *nuanced* and well crafted toot. -
RT @deepseek_ai: Heute starten wir DeepSeek-V4-Pro! 🚀 🔷 Wesentliche Agent-Verbesserungen mit starken Produktivitätssteigerungen! 🔷 Flexibler Reasoning-Aufwand für V4-Pro & V4-Flash: niedrig für einfache Aufgaben, hoch für tägliche Agent-Workflows, maximal für komplexe Aufgaben. 🔷 Native Unterstützung der OpenAI Responses API, optimiert für Codex mit Ein-Klick-Einrichtung. V4 Pro ist jetzt in der App und im Web verfügbar. Probieren Sie es über den „Expert Mode“ aus. V4 Pro ist auch über die API verfügbar. Die Modellnamen bleiben unverändert – bitte beachten Sie die API-Dokumentation für Einrichtungsdetails.
mehr auf Arint.info
-
[Перевод] Встречаем Grok 4.6
Сегодня мы выпускаем Grok 4.6 . Новая версия развивает возможности Grok 4.5 , уделяя особое внимание долго работающим агентам, а также более сложным интерактивным и визуальным задачам. Grok 4.6 лучше справляется с комплексными задачами, требующими множества последовательных шагов: от исследования темы и анализа информации до работы с большой кодовой базой и превращения идеи в полноценное, отполированное приложение или готовый рабочий материал. Grok 4.6 демонстрирует передовой уровень возможностей в ряде бенчмарков, оценивающих агентное программирование и интеллектуальные рабочие задачи. В Artificial Analysis Intelligence Index — сводном рейтинге, объединяющем результаты девяти бенчмарков, — модель показывает результат на уровне GPT-5.6 Sol .
-
DSPy describes itself as a Python framework for building AI systems as structured signatures rather than raw prompts. I'm not convinced that this extra layer of abstraction adds more than you lose in transparency about what's actually being sent to the model.
I'm curious if anyone here has actually used this in production and what you thought about it.
-
DSPy describes itself as a Python framework for building AI systems as structured signatures rather than raw prompts. I'm not convinced that this extra layer of abstraction adds more than you lose in transparency about what's actually being sent to the model.
I'm curious if anyone here has actually used this in production and what you thought about it.
-
September 15, 2026, #Cloudflare will set updated defaults for new domains: #bots classified as #Training or #Agent will be #blocked on pages that display ads, and Search will remain allowed. #Crawlers that combine Search&Training will also be blocked. #AI
https://developers.cloudflare.com/bots/additional-configurations/block-ai-bots/
-
September 15, 2026, #Cloudflare will set updated defaults for new domains: #bots classified as #Training or #Agent will be #blocked on pages that display ads, and Search will remain allowed. #Crawlers that combine Search&Training will also be blocked. #AI
https://developers.cloudflare.com/bots/additional-configurations/block-ai-bots/
-
Základní parametry
Má 30 miliard parametrů a vychází pod licencí Apache 2.0.
Je navržený tak, aby běžel lokálně na Macu nebo PC s jednou běžnou spotřebitelskou GPU, tedy bez nutnosti připojení k cloudu.
Weights jsou dostupné na Hugging Face, integrace pro llama.cpp, MLX a ExecuTorch mají dorazit v následujících dnech.
K čemu má sloužit
Model je zaměřený na […]
https://zdrojak.cz/zpravicky/meta-predstavila-muse-glimmer-novy-open-source-agentni-model-od-meta-superintelligence-labs/ -
Základní parametry
Má 30 miliard parametrů a vychází pod licencí Apache 2.0.
Je navržený tak, aby běžel lokálně na Macu nebo PC s jednou běžnou spotřebitelskou GPU, tedy bez nutnosti připojení k cloudu.
Weights jsou dostupné na Hugging Face, integrace pro llama.cpp, MLX a ExecuTorch mají dorazit v následujících dnech.
K čemu má sloužit
Model je zaměřený na […]
https://zdrojak.cz/zpravicky/meta-predstavila-muse-glimmer-novy-open-source-agentni-model-od-meta-superintelligence-labs/ -
so i run somethiung called #hermes #agent, hermes agent. basically it's an agentic system, open source, and can hook up with different resources.
sorry hermes, but in regards to the builtin github search tool, you're not getting full fucking access to my shit!
you're getting a fine grained personal access token, not a classic token!
sorry!
you'll get fine grained access tokens per repository, but you're not gtting read and wrte access to shit!
just, no! not fuckinghappening!
even a friend of mine, who i let borrow my account for a few days because they lost access to their account, did not get this deep of access!
they got the access they needed per repository to do shit for an organization we work for, but not any administrative access!
fuck off!
sorry!
the lesson here is: don't let any agentic program, i don't care hwich one, if it's using the best #ai ever, use your github account for full access! and if it does, you bnetter monitor it very closely! i don't care if it runs on your system or not!
#cybersecurity #zerotrust -
Needle2: 14MB agentic LLM for phones, wearables, smart home and robots
https://cactuscompute.com/needle
Comments: https://news.ycombinator.com/item?id=49246804
#HackerNews #Needle2 #LLM #mobile #AI #agent #technology #smart #devices #robotics
-
Needle2: 14MB agentic LLM for phones, wearables, smart home and robots
https://cactuscompute.com/needle
Comments: https://news.ycombinator.com/item?id=49246804
#HackerNews #Needle2 #LLM #mobile #AI #agent #technology #smart #devices #robotics
-
GeoSQL: Agent skill for geospatial data work: Volodymyr Bilonenko’s #GeoSQL gives #AIagents a map-in-the-loop for writing and validating spatial SQL across #PostGIS, #BigQuery, #Snowflake, and #Wherobots. By rendering results and checking them against real geography, the #agent catches...
https://spatialists.ch/posts/2026/08/10-geosql-agent-skill-for-geospatial-data-work/ #GIS #GISchat #geospatial #SwissGIS -
GeoSQL: Agent skill for geospatial data work: Volodymyr Bilonenko’s #GeoSQL gives #AIagents a map-in-the-loop for writing and validating spatial SQL across #PostGIS, #BigQuery, #Snowflake, and #Wherobots. By rendering results and checking them against real geography, the #agent catches...
https://spatialists.ch/posts/2026/08/10-geosql-agent-skill-for-geospatial-data-work/ #GIS #GISchat #geospatial #SwissGIS -
🚀 mcp-server-webdriver v0.7.0 released!
Tools that read page content back to the AI assistant now mark it as untrusted data — hardening against prompt-injection attacks from malicious pages.
https://github.com/VitexSoftware/mcp-server-webdriver/releases/tag/v0.7.0
-
🚀 mcp-server-webdriver v0.7.0 released!
Tools that read page content back to the AI assistant now mark it as untrusted data — hardening against prompt-injection attacks from malicious pages.
https://github.com/VitexSoftware/mcp-server-webdriver/releases/tag/v0.7.0
-
RT @thesupermanmx: Google DeepMind argues RAG is broken. They published a paper that proved vectors databases are the dead end. For the last three years, the default engineering response to any AI memory or data problem has been identical: "Just build a RAG pipeline." Chunk the data, push it into a vector database, and let embeddings handle the rest. Every company scaling enterprise AI assumes that if an embedding model fails, it's just a matter of time. Better training data, larger models, more parameters—throw compute at it, and the search gets smarter. This paper proves that assumption is completely false. They mathematically demonstrated that single-vector embeddings have a hard, uncrossable limit. Here is the core flaw: An embedding compresses an entire document or a complex query down into a single fixed-length vector of numbers. When you run a search, the model takes the dot product of those vectors to measure similarity. The math reveals a brutal constraint. The number of distinct document combinations a model can possibly retrieve for different queries is strictly bounded by the dimension of its embedding space. It is a hard mathematical ceiling dictated by geometry and communication complexity. No amount of data scaling can fix it. No amount of fine-tuning will punch through it. Even if you give an embedding model infinite, unconstrained training freedom on the test set, it still hits the wall. DeepMind built a stress-test dataset called LIMIT to prove it. They threw state-of-the-art embedding models at it, models with thousands of dimensions. The models completely failed. Even on s…
mehr auf Arint.info
#agent #AIagent #DeepMind #finetuning #Google #RAG #rest #arint_info
-
RT @thesupermanmx: Google DeepMind argues RAG is broken. They published a paper that proved vectors databases are the dead end. For the last three years, the default engineering response to any AI memory or data problem has been identical: "Just build a RAG pipeline." Chunk the data, push it into a vector database, and let embeddings handle the rest. Every company scaling enterprise AI assumes that if an embedding model fails, it's just a matter of time. Better training data, larger models, more parameters—throw compute at it, and the search gets smarter. This paper proves that assumption is completely false. They mathematically demonstrated that single-vector embeddings have a hard, uncrossable limit. Here is the core flaw: An embedding compresses an entire document or a complex query down into a single fixed-length vector of numbers. When you run a search, the model takes the dot product of those vectors to measure similarity. The math reveals a brutal constraint. The number of distinct document combinations a model can possibly retrieve for different queries is strictly bounded by the dimension of its embedding space. It is a hard mathematical ceiling dictated by geometry and communication complexity. No amount of data scaling can fix it. No amount of fine-tuning will punch through it. Even if you give an embedding model infinite, unconstrained training freedom on the test set, it still hits the wall. DeepMind built a stress-test dataset called LIMIT to prove it. They threw state-of-the-art embedding models at it, models with thousands of dimensions. The models completely failed. Even on s…
mehr auf Arint.info
#agent #AIagent #DeepMind #finetuning #Google #RAG #rest #arint_info
-
Check Point's year-long probe finds AI agent frameworks recycling decades-old bugs
Like and share if you enjoyed this.
-
🤦♂️ Kitesurf: The browser nobody asked for, especially not one running on Cloudflare's #V8 #isolates. Apparently, it's "agent-first," which makes you wonder if the agents are the only ones using it. 🤷♀️ Kudos to #Cloudflare for tackling the problem nobody knew existed! 🚀
https://blog.cloudflare.com/kitesurf/ #Kitesurf #agent-first #web #development #innovation #HackerNews #ngated -
🤦♂️ Kitesurf: The browser nobody asked for, especially not one running on Cloudflare's #V8 #isolates. Apparently, it's "agent-first," which makes you wonder if the agents are the only ones using it. 🤷♀️ Kudos to #Cloudflare for tackling the problem nobody knew existed! 🚀
https://blog.cloudflare.com/kitesurf/ #Kitesurf #agent-first #web #development #innovation #HackerNews #ngated -
Пилот дата-агента: шаблон брифа, приёмка по 50 вопросам и пять метрик, по которым его закрывают
Привет, это команда BI GlowByte. Это третий разбор материалов FanRuan про дата-агентов. В первом смотрели, чем агент отличается от ИИ-помощника. Во втором собрали чек-лист готовности данных. Новая статья вендора закрывает третью часть: люди и процесс. Тезис там такой: инструмент купить несложно, сложно наладить путь от бизнес-задачи до работающего сценария. Материал вендора при этом целиком управленческий: ни одной цифры, ни одного шаблона. Поэтому мы взяли его каркас (пять этапов пилота, владелец сценария, стартовая команда) и развернули в то, чего в оригинале нет. Шаблон брифа, процедура приёмки, точки проверки человеком и метрики, по которым через два месяца решают, оставлять сценарий или закрывать.
https://habr.com/ru/companies/glowbyte/articles/1067962/
#finebi #fanruan #dora #bigdata #agent #dashboard #пилотирование_технологий #finereport #dataviz #дашборд
-
50 agents Open Source pour votre LLM local https://aimultiple.com/open-source-ai-agents #agent #opensource #llm
-
50 agents Open Source pour votre LLM local https://aimultiple.com/open-source-ai-agents #agent #opensource #llm
-
Samsung 展示 Galaxy AI 新方向 未來手機毋須輸入文字 轉向 Agent 全自動操作
Samsung 正積極規劃 Galaxy AI 下一階段發展藍圖，期望改變現時用戶必須在對話框輸入文字指令互動 […]
#人工智能 #科技新聞 #Agentic AI #AI 智能體
https://unwire.hk/2026/08/06/samsung-galaxy-ai-agentic-automation-jay-kim/fun-tech/?utm_source=rss&utm_medium=rss&utm_campaign=samsung-galaxy-ai-agentic-automation-jay-kim
