Je découvre #LMArena c'est un site qui permet d'utiliser différents modèles IA (LLMs , génération d'images et recherche wb) sans avoir besoin de compte.
On ne peut pas choisir quel modèle on veut utiliser, il nous propose à chaque fois 2 réponses et nous invite a voter pour la meilleure d'entre-elles.
C'est crée par des chercheurs d'UC Berkeley. Ils proposent différents classements basés sur les retours utilisateurs.
Çà ne semble pas limité dans l'usage.
Pre-LLM assisted debugging tools.
'Will LLMs eliminate the need for junior engineers? Should senior engineers get out of the profession before it’s too late? My answer to all these questions is “I haven’t the foggiest”. '
Falls auch ihr aufgrund unternehmerischer Entscheidungen eures Arbeitgebers voraussichtlich nicht umhinkönnt, zukünftig KI-Tools zur Effizienzsteigerung einzusetzen:
Überlegt euch schon jetzt, mit welchen Kennzahlen sich euer Output heute messbar machen lässt. Führt Buch darüber. Mit einem handfesten Vorher-Nachher-Vergleich habt ihr eine bessere Verhandlungsgrundlage für das nächste Gehaltsgespräch.
Produktivitätssteigerung darf keine Einbahnstraße sein. 
(Bitte keine Grundsatzdiskussion – denkt dran, dass nicht alle Menschen das Privileg haben, aus moralischen Gründen "mal eben" den Job wechseln zu können.)
Truly wild reading.
"If such a prompt injection is included in a submission and it consequently results in a positive LLM-generated review, we consider this a form of collusion (which, as per past precedent, is a Code of Ethics violation) that both the paper authors and the reviewer would be held accountable for, because it involves the author explicitly requesting and receiving a positive review.
https://blog.iclr.cc/2025/08/26/policies-on-large-language-model-usage-at-iclr-2026/
Why are they not just banning the use of LLMs to write a review from scratch? That is the real problem here.
I don't know this conference but I'll make sure never to submit there.
Falls auch ihr aufgrund unternehmerischer Entscheidungen eures Arbeitgebers voraussichtlich nicht umhinkönnt, zukünftig KI-Tools zur Effizienzsteigerung einzusetzen:
Überlegt euch schon jetzt, mit welchen Kennzahlen sich euer Output heute messbar machen lässt. Führt Buch darüber. Mit einem handfesten Vorher-Nachher-Vergleich habt ihr eine bessere Verhandlungsgrundlage für das nächste Gehaltsgespräch.
Produktivitätssteigerung darf keine Einbahnstraße sein.
(Bitte keine Grundsatzdiskussion – denkt dran, dass nicht alle Menschen das Privileg haben, aus moralischen Gründen "mal eben" den Job wechseln zu können.)
PostNAS starts with a pre-trained model and freezing its MLP weights, enabling efficient, cost-effective exploration of optimal attention designs without costly retraining. This approach delivers up to 53.6× generation throughput speedup and 6.1× prefilling speedup. It also achieves higher accuracy on MMLU and MMLU-Pro than recent advanced MoE full-attention models, despite their larger scale.
There's magic trick that can make any #AI project successful. When the LLM's success rate is too low, just say "it's OK, we will have a human in the loop" and ship it anyway.
There's just one problem with this magic trick: most #LLM implementations don't have the thoughtful and intentional guard rails that would make this a real human-in-the-loop system.
Instead, the loop becomes a vicious cycle of deception and deskilling, losing all benefits of oversight.
https://productpicnic.beehiiv.com/p/human-in-the-loop-is-a-thought-terminating-cliche
Can anyone put the case against LLMs more elegantly?
Microsoft launches Copilot AI function in Excel, but warns not to use it in 'any task requiring accuracy or reproducibility'
There's magic trick that can make any #AI project successful. When the LLM's success rate is too low, just say "it's OK, we will have a human in the loop" and ship it anyway.
There's just one problem with this magic trick: most #LLM implementations don't have the thoughtful and intentional guard rails that would make this a real human-in-the-loop system.
Instead, the loop becomes a vicious cycle of deception and deskilling, losing all benefits of oversight.
https://productpicnic.beehiiv.com/p/human-in-the-loop-is-a-thought-terminating-cliche
Can anyone put the case against LLMs more elegantly?
Microsoft launches Copilot AI function in Excel, but warns not to use it in 'any task requiring accuracy or reproducibility'
Folks, @micr0 is the creator and maintainer of altbot@fuzzies.wtf and tldr@fuzzies.wtf, arguably two of the most useful #Fediverse bots I've ever seen, utilizing a locally run, low-power #LLM server operated with #privacy and #transparency primarily in mind.
If I wasn't already living off of charity, I would gladly donate to this cause. So instead, I'm asking everyone who is willing and able, please visit the post below and consider donating to help micr0 continue to provide two immensely powerful and helpful services for #inclusion and #accessibility. They deserve to at least break even on their bills for such (imho) important work.
Additionally, I encourage everyone to follow one or both bots (I'm not tagging them in the post to avoid triggering another loop like I did the last time) if you are overly-verbose like myself and need a concise, abridged version of my babblings, or don't have the spoons to write alt text for images AND videos.
Folks, @micr0 is the creator and maintainer of altbot@fuzzies.wtf and tldr@fuzzies.wtf, arguably two of the most useful #Fediverse bots I've ever seen, utilizing a locally run, low-power #LLM server operated with #privacy and #transparency primarily in mind.
If I wasn't already living off of charity, I would gladly donate to this cause. So instead, I'm asking everyone who is willing and able, please visit the post below and consider donating to help micr0 continue to provide two immensely powerful and helpful services for #inclusion and #accessibility. They deserve to at least break even on their bills for such (imho) important work.
Additionally, I encourage everyone to follow one or both bots (I'm not tagging them in the post to avoid triggering another loop like I did the last time) if you are overly-verbose like myself and need a concise, abridged version of my babblings, or don't have the spoons to write alt text for images AND videos.
@MissConstrue 🧵ChatGPT-5 as reference link fixer #LLM
I asked: “Please give quotes and summary of notes after Agena undocking, including recognition, diagnosis, and recovery from stuck thruster problem. Include direct quotes and for each block of quotes, the debriefing report page number. I want a short narrative using direct quotes including attribution of the astronaut speaking.”
One issue: it used PDF rather than Report page numbers for citations.
4/4
@MissConstrue 🧵ChatGPT-5 as reference link fixer #LLM
I agree. I have however been able to easily prompt and poke ChatGPT-5 to review a 13 year old blog post about Neil Armstrong (I gave it a URL) and suggest fixes for any broken links it found iin the post’s reference list. It found the broken links and made helpful suggestions, but I wanted more.
1/N
AI simply fills the void created by capitalist relations that erode human connection, breeding alienation that leaves people isolated, adrift, and emotionally starved.
It’s a hollowed-out shell of communal life, where AI chatbots have become digital opioids providing addictive facsimiles of human relationships. These are symptoms of societal collapse, where atomization drives the masses to seek solace in algorithms.
https://www.theatlantic.com/technology/archive/2025/08/ai-mass-delusion-event/683909/
Why SearchFOSS?
1) We believe that access to web information should be community-based and not owned by big corporations. By using SearchFOSS or any other @pears instance you contribute to this!
2) It's curated: we rely on wisdom of the crowd to have an index that's as useful as possible with a minimum of use. We are exploring limited automated indexing (e.g. following RSS feeds of trusted sources) to help grow the index faster, but won't do indiscriminate scraping.
2/
4) It's federated: similar to how #mastodon & #fediverse work, your instance is your local community but you're part of a larger and growing network. If you search on SearchFOSS, you get both local information and results from the @pears network. Some cool related remote instances are ai-pears.org and retronlp.org
3/
A space for Bonfire maintainers and contributors to communicate
