De Abreu David
7.1K posts

De Abreu David
@twitifyoucan
#WebDeveloper - #HybridDevelopper - Certifié #Microsoft #Platinium #MVA - #Addict aux #Nouvelles #Techno -#Réseaux #Sociaux #Fondateur de @ecole_virtuelle
France Katılım Eylül 2010
682 Takip Edilen2.5K Takipçiler

Transformez Claude Code en pair programmer expert avec ces 66 skills spécialisées : backend, frontend, infra, sécu, testing...
github.com/Jeffallan/clau…

Français

🚨BREAKING: Microsoft Research + Salesforce just dropped a paper that should scare every AI builder.
They tested 15 top LLMs GPT-4.1, Gemini 2.5 Pro, Claude 3.7 Sonnet, o3, DeepSeek R1, Llama 4 across 200,000+ simulated conversations.
Single-turn prompt: 90% performance.
Multi-turn conversation: 65% performance.
Same model. Same task. Just... talking normally.
The culprit isn't intelligence. Aptitude only dropped 15%.
Unreliability EXPLODED by 112%.
→ LLMs answer before you finish explaining (wrong assumptions get baked in permanently)
→ They fall in love with their first wrong answer and build on it
→ They forget the middle of your conversation entirely
→ Longer responses introduce more assumptions = more errors
Even reasoning models failed. o3 and DeepSeek R1 performed just as badly.
Extra thinking tokens did nothing.
Setting temperature to 0? Still broken.
The fix right now: give your AI everything upfront in one message instead of back-and-forth.
Every benchmark you've seen was tested on single-turn prompts in perfect lab conditions.
Real conversations break every model on the market and nobody's talking about it.

English

@MistralDevs a quand un petit support avancé de clawbot ? Pour le moment avec Ollama/devstral2small c'est une catastrophe 🥶
Français

@MistralDevs I use it in a daily basis since launched, it's amazing and efficient for french or english( with french accent) ! Love it
English

@dr_cintas This is exactly what I've been looking for! The local-first approach with Claude Code integration is a game changer for design workflows 🔥
English

@openclaw Amazing release! The nested sub-agents are exactly what we needed 🔥Congrats team!
English

🦞 OpenClaw 2026.2.15 is here!
✨ Telegram message streaming — replies flow live
💬 Discord Components v2 — buttons, selects, modals
🔧 Nested sub-agents
🔒 Major security hardening pass
🐛 40+ bug fixes
Big day. Huge day. Maybe the biggest day.🏛️ github.com/openclaw/openc…
English

@AlexChe31659937 @MarioNawfal Incredible progress! The speed at which physical AI is evolving is mind-blowing 🤖🔥
English

@MarioNawfal Once the US opens up its technology, China is flooded with remote-controlled robot toys.
English

@chiefofautism This is wild! The implications for AI research are huge 🔥
English

@HuggingModels This is game-changing for document processing! Great share 👏
English

@ollama @MiniMax_AI I absolutely love it ! Worked all the day with #minimax.My setup is:OpenClaw on a #rasp5, #Ollama on a 4090 24Go VRAM win11 on the same network. After doing a small proxy because openclaw was sending shitty POST request to ollama, I was able to make it work. So cool and efficient
English

❤️ We are partnering with @MiniMax_AI to give Ollama users free usage of MiniMax M2.5 for the next couple of days!
ollama run minimax-m2.5:cloud
Use MiniMax M2.5 with OpenCode, Claude Code, Codex, OpenClaw via ollama launch!
OpenCode:
ollama launch opencode --model minimax-m2.5:cloud
Claude:
ollama launch claude --model minimax-m2.5:cloud

MiniMax (official)@MiniMax_AI
English

C'est du Troll? Car si tu bosses vraiment avec tu te rendrais vite compte qu'il fait bien le taff dans 75% des cas. C'est l'outil idéal pour M.touslemonde
Malugrin92@Malugrin92
@_so_bored Génial un fork de chatgpt 2.5 financé par mes impôts. personne considère mistral comme autre chose qu’une blague ou pire un scam d’argent publique.
Français

Interesting, somehow alarming but at least 50% true.
Lets check this post again in 5 years and see :)
Matt Shumer@mattshumer_
English

@HappyGezim @openclaw @ollama Yeah i think i will do something like that, keeping opus for main-agent ( for enabling real skills ) and then define my local agent ( oss, llama4 ) for small dedicated task. I was too optimistic doing onle 1 main agent under oss ^^ thx thought
English

@twitifyoucan @openclaw @ollama not just you lol local models def struggle with agent tasks. what does your router config look like? routing to opus only for complex stuff helps keep costs down
English

@irukanji_invest Whisper fait aussi de l autoTranslate, chose que voxtral ne fait pas. De plus c'est une connexion APi et pas un llm local. J'ai test les 2, les 2 sont porpres ( mieux que parakeet pour le FR en tout cas ). Mais pas utilisable au milieu du désert... 😁
Français

Pourquoi c'est important d'avoir un bon modèle STT open-weight (ie: que l'on peut faire tourner sur son PC) ?
Eh bien jusqu'ici, l'état de l'art en STT libre, c'était "whisper"... une technologie qui approchait les 3 ans d'âge. Une éternité en IA...
Mistral AI@MistralAI
Introducing Voxtral Transcribe 2, next-gen speech-to-text models by @MistralAI. State-of-the-art transcription, speaker diarization, sub-200ms real-time latency. Details in 🧵
Français


@LucJager Après le dev, tu crée l agent PR et un QA ensuite tu passes ça a l'agent Devops qui build/deploy et enfin l agent user-playwright qui test le flow en preprod
Français

j'ai créé un agent chef de projet qui vérifie un autre agent
ptn quelle singuerie
maintenant faut juste que je fasse un agent chef de projet fonctionnel qui gère 5 agents chef de projet techniques qui gère chacun 3-4 agents dev, designeur
et après on ajoute quelques agents d'ambiance et on est bon lol
Maintenant faut juste que ça marche et que ça coute pas 1 Milliard de token

Français












