🤖 DLI2026 Conference - Poster Judging
This article details the poster judging role at the DLI2026 conference, highlighting the involvement of a PhD candidate in evaluating emerging researchers.
Key Points:
• PhD Candidate Moyahabo Rabothata will serve as a Poster Judge.
• The role involves evaluating researchers from across the continent.
🚀 DLI2026 Conference - Engagement Opportunities
This article informs attendees about opportunities to engage with DSFSI_Research activities at DLI2026 in Lagos, including talks, workshops, and poster sessions.
Key Points:
• Attendees can catch talks and join workshops presented by DSFSI_Research.
• Visiting their posters is another opportunity for engagement.
🔗 Resources:
• DSFSI Blog ↗ - Read the full lab announcement
🤖 DLI2026 Conference - Research Poster Showcase
This article highlights specific research posters presented at DLI2026, focusing on natural language processing applications in African contexts.
Key Points:
• Dr Seani Rananga will present research on detecting misinformation in African languages using LLMs.
• Nontokozo Manukuza's poster covers interpreting isiZulu idiomatic expressions with LLMs.
🤖 Qwen3.6-35B-A3B-Escha-W2 - Quantized MoE Model
This article describes the Qwen3.6-35B-A3B-Escha-W2 model, a Mixture-of-Experts (MoE) text generation model that uses 2-bit quantization for efficient operation.
Key Points:
• The model operates with 35 billion total parameters but only 3 billion active parameters due to its MoE architecture.
• It employs 2-bit quantization to achieve a smaller footprint without sacrificing quality.
• Built for sglang and zml, the model is saved in safetensors format.
• This design makes it suitable for edge deployment or environments with tight VRAM constraints.
🔗 Resources:
• Qwen3.6-35B-A3B-Escha-W2 Model ↗ - Explore detailed information about the model
Image
🤖 Urban Flood Prediction - Physics-Informed CNN-LSTM
This article presents a method for street-scale urban flood prediction using a hybrid Physics-Informed CNN-LSTM model. The approach aims to combine aggregate accuracy with localized plausibility.
Key Points:
• The method uses a Physics-Informed CNN-LSTM for urban flood prediction.
• It focuses on achieving both high aggregate accuracy and street-level plausibility.
🔗 Resources:
• arXiv Paper ↗ - Physics-Informed CNN-LSTM for Street-Scale Urban Flood Prediction
Image
💡 GPT-5.6 Sol - API Settings for ARC-AGI Performance
This article discusses how modifying API settings can impact language model benchmark performance, specifically for GPT-5.6 Sol on ARC-AGI 3.
Key Points:
• Two API settings adjustments tripled GPT-5.6 Sol's score on ARC-AGI 3.
• The underlying GPT-5.6 Sol model itself was not modified; only its API configuration changed.
• Related topics include Kimi K3, Opus 5, autonomous agent attacks, and the open-weights AI debate.
🔗 Resources:
Image
⭐️ Support
If you liked reading this report, please star ⭐️ this repository and follow me on Github ↗, 𝕏 (previously known as Twitter) ↗ to help others discover these resources and regular updates.
