Hi @thsottiaux
, Sometimes GPT-5.4 in Codex outputs Hindi characters like this while speaking French. It also occasionally drops accents. For example, “l’intitulé” becomes “l intitule”. Perhaps this is a tokenizer issue? That’s a weird behavior. It would be great to get it
LLMS
-
GPT-5.4 outputs Hindi characters and drops French accents
By
–
-
Domain Experts Will Build AI Killer Apps
By
–
Claude Code was developed by a coder. The AI killer apps for each field will be developed by people in it.
-
Grok 4.20 Heavy Beta 2 fast, Beta 3 with fixes and gains
By
–
Grok 4.20 Heavy (Beta 2) is extremely fast for deep analysis. Beta 3 will have many fixes and functionality gains.
-
Claude builds interactive Cessna 172 instrument panel in chat, potential for education
By
–
I asked Claude to build an interactive instrument panel from a Cessna 172 directly in the chat…
— AI Breakfast (@AiBreakfast) 12 mars 2026
Pretty cool! Not perfect by any means but when this gets refined, it has major potential for education. https://t.co/6C87CkhQuM pic.twitter.com/DhLCWQCq5uI asked Claude to build an interactive instrument panel from a Cessna 172 directly in the chat… Pretty cool! Not perfect by any means but when this gets refined, it has major potential for education.
-
Frontier Models and Router Systems for AI Optimization
By
–
The most effective AI systems don't rely on a single model. Frontier models provide state-of-the-art performance for complex tasks, while routers automatically select lightweight, open-source models for simpler jobs to optimize accuracy, latency, and cost. Learn more:
-
Claude Consistency Issues With Detailed AI Skills Configuration
By
–
That’s for sure! What are you looking into these days that is failing for you? On my end, still struggling with consistency even if I have super detailed Claude skills, some stuff are quite random and can’t manage to make them the same every time.
-
Open-Weight Local AI Models as Alternative to Centralized AI Services
By
–
He's making a great case for people using open-weight models on their own local devices. If the OpenAI is the new electric company, having on-prem AI machines is like having your own solar panels.
-
Codex Spark Best Practices Guide and Comprehensive Tips
By
–
for more comprehensive tips, check out our guide to Codex Spark: cerebras.ai/blog/codex-spark…
-
Cerebras Codex Spark Revolutionizes Coding with 1,200 Tokens Per Second
By
–
Our coding workflows were designed to accommodate slow inference. @OpenAI's Codex Spark powered by @cerebras changes the game.
— Cerebras (@cerebras) 12 mars 2026
Here's how we make the most out of 1,200 tokens per second, with @MilksandMatcha. pic.twitter.com/vv4a80wfFAOur coding workflows were designed to accommodate slow inference. @OpenAI's Codex Spark powered by @cerebras changes the game. Here's how we make the most out of 1,200 tokens per second, with @MilksandMatcha.
