#llama

7 posts · Last used 17d

Back to Timeline
OpenUK @openuk@hachyderm.io · Jul 28, 2026
AI Sovereignty discussions have shifted as we discuss a Stack and "layered sovereignty" following the NVIDIA 5-layered cake 🍰. Shift from frontier models to individuals building the infrastructure layer, open source, via GitHub. 🛠️ Read the full report here: https://openuk.uk/report/aiopenness2026/ #OpenModels #OpenWeights #AIOpenness #OpenSourceAI #OpenSource #AIInfrastructure #AIStack #Llama #NVIDIA #OpenUK #OpenHQ
0
0
0
Rost Glukhov @ros@techhub.social · Jul 23, 2026
Install llama.cpp, run GGUF models with llama-cli, and serve OpenAI-compatible APIs using llama-server. Key flags, examples, and tuning tips with a short commands cheatsheet #Cheatsheet #AI #LLM #DevOps #OpenAI #API #SelfHosting #Prometheus #llama.cpp https://www.glukhov.org/llm-hosting/llama-cpp/
1
0
4
Eudaimon ꙮ @eudaimon@fe.disroot.org · Jul 19, 2026

«Un estudi de la Universitat d'Oxford revela que la intel·ligència artificial actua com un "veí invisible" que ens canvia la ideologia»

https://www.ara.cat/media/manipula-ia-li-demanem-millori-text_1_5788377.html

Una investigació matemàtica demostra que eines com Grok distorsionen els debats sobre temes sensibles, com l'avortament

Bàsicament: proposa entendre un LLM com una persona de personalitat X, amb valors X, tendències X, que té un pes social equivalent al número de persones que hi parlen regularment. En el cas dels grans models, això vol dir un pes de centenars de milions globalment, i si pensem en Espanya o Catalunya, menys, però diria que a aquestes alçades, entre uns i altres, pesen algun valor de dues xifres, posa un 30%, respecte el total. O sigui: cada LLM és un actor social de molt pes, amb la seva visió i biaixos (com els tenim cadascú individualment). El que em fa gràcia és després la preocupació per la influència electoral d'això. Una democràcia real (entenent per democràcia quelcom molt més profund que els models actuals) requereixen, i molt importantment, un periodisme (aka circulació d'informació) molt conscient del paper que juga, de la necessitat que cobreix d'ajudar a entendre i desentranyar la tremenda complexitat de la societat i els seus processos, perquè després la gent prenem decisions en funció de la informació. OK, però la cosa és que el periodisme porta contaminat molt de temps. Podem començar per la tinença dels mitjans, molt concentrada globalment, amb uns interessos molt llunyans d'això que he dit del periodisme. Podem continuar pels propis governs, que fan tot el possible per influir en els mitjans (públics o privats). I també per les corporacions propietàries de les grans "xarxes socials", amb algoritmes que premien i castiguen també de forma interessada certes opinions. Tot això per dir que sí, que els LLM influencien (i no poc), però és que ja fa molts anys (dècades) que aquest element clau de la democràcia està fortament degradat. Segurament des del principi ja ho estava i, digues-me malpensat, és per això que es va acceptar aquest sistema polític des de les instàncies del poder real (com deia noséqui, parafrasejant, "si votar pogués servir per canviar realment les coses, estaria prohibit") I amb això aprofito per deixar caure aquesta hipòcrita preocupació per la influència de potències estrangeres en les eleccions locals. No per descartar-la (segurament passi), sinó per dir que si preocupa que la gent sigui influïda en les seves decisions per pocs actors amb molt poder, fa dècades que això està passant. I aquestes institucions aparentment preocupades, justament fan el mateix en altres països. Exagerat (i alhora el que passa), és com criticar i atacar "els seus genocides" i en canvi defensar i justificar "els nostres genocides".

#LLM #IA #Grok #ChatGPT #Llama #Mistral #Gemini #AI

0
0
0
troed @troed@fedia.io · Jun 14, 2026

Opencode llama-server prefill/generation stats plugin

Opencode llama-server prefill/generation stats plugin #opencode #llama Hover or focus to reveal Sensitive
I've just published an Opencode plugin I made since I couldn't find one that fulfilled the exact use case I had myself. Publishing and posting in case it's useful for someone else too. When working with local models I have the need to know why there's suddenly nothing happening (besides the blue cylon-bar) but the plugins I found only showed token generation data when something was passed into Opencode. That meant that during >1 minute prefills there was no output at all. This plugin uses the /slots endpoint (enabled by default) in llama-server to deduce whether it's currently generating tokens or doing prompt processing, and also the current tps for that activity. Now I can just run llama-server as a daemon and I no longer feel the need to go inspect its output just to see what's up. It's likely only useful in a single-user scenario, but it has been tested with both single and multiple parallel slots. Installation: opencode plugin @troed/oc-ls-stats@latest --global
0
1
0
The Bad Place @TheBadPlace@mastodon.ozioso.online · Apr 09, 2026
undefined | Meta's New AI Model Gives Mark Zuckerberg a Seat at the Big Kid's Table by Will Knight Meta announced its first major AI model since the company’s 2025 AI reboot, unveiling Muse Spark as the flagship of its new Meta Intelligence Labs division. Described by CEO Mark Zuckerberg as a step toward “personal superintelligence,” the model is being offered through meta.ai and the Meta AI app, but unlike previous Llama releases it remains closed‑source for now. Zuckerberg emphasized that the goal is to create agents that not only answer questions but actively perform tasks, hoping the technology will spark a wave of creativity, entrepreneurship, growth, and health‑related advances. Muse Spark is positioned as a significant upgrade over Meta’s earlier Llama 4, delivering multimodal capabilities that handle images, audio, video, and text, as well as advanced reasoning and strong coding skills. Independent benchmarking by Artificial Analysis placed the model in the top five of all tested systems, claiming it outperforms contemporary offerings from OpenAI, Anthropic, Google, and xAI on several tasks. Meta also highlighted the model’s medical‑advice proficiency, noting a collaboration with more than 1,000 physicians to curate training data for more factual and comprehensive health responses. The launch follows a massive overhaul of Meta’s AI strategy, including high‑profile hires, billions invested in AI startups, and a $14.3 billion acquisition of Scale, whose CEO now leads the effort. To guide the rapid scaling of ever more capable systems, Meta released its Advanced AI Scaling Framework, outlining safety checks and responsible practices as the models move toward superhuman performance. While Muse Spark itself remains proprietary, Meta has signaled plans to open‑source future iterations as part of its broader push to reclaim a leadership role in the AI landscape. Read more: undefined #meta #markzuckerberg #llama #meta-ai #meta-ai
0
0
0
The Bad Place @TheBadPlace@mastodon.ozioso.online · Apr 09, 2026

bing news | Meta's New AI Model Gives Mark Zuckerberg a Seat at the Big Kid's Table

Meta unveiled its first major AI model since the company’s 2025‑2026 AI reboot, naming it Muse Spark. Presented as a step toward Mark Zuckerberg’s goal of “personal superintelligence,” the closed‑source model is being offered through meta.ai and the Meta AI app, though it will not be downloadable like earlier Llama releases. Zuckerberg highlighted that Muse Spark is designed to act as an “agent” that does tasks for users, aiming to spur creativity, entrepreneurship, growth, and health improvements.

According to Meta’s self‑reported benchmarks, Muse Spark outperforms several contemporary models from OpenAI, Anthropic, Google, and xAI on a range of tasks, scoring 52 on the Artificial Analysis Intelligence Index and landing in the top five models tested by the firm. The model is natively multimodal—supporting text, images, audio, and video—and boasts advanced reasoning and strong coding abilities. A standout feature is its enhanced medical‑advice capability, built with input from over 1,000 physicians to deliver more factual and comprehensive health responses.

The release follows an aggressive overhaul of Meta’s AI division, including hiring top talent with hefty compensation packages and investing billions in AI startups such as Scale, whose CEO Alexandr Wang now leads the effort. Meta also published an “Advanced AI Scaling Framework” outlining safety checks for future, more powerful models, signaling the company’s intent to continue pushing the frontier of AI while addressing safety as it scales toward superhuman performance.

Read more: https://www.wired.com/story/muse-spark-meta-open-source-closed-source/

#meta #markzuckerberg #musespark #meta-ai #meta-ai #llama

0
0
0

You've seen all posts