Meta Superintelligence Labs vs Facebook AI Research (Jan 30, 2026)
AI by Hand ✍️ Seminars
Library › Seminar Series 2026
Manifold-Constrained Hyper Connections (mHC) from DeepSeek (Jan 9, 2026)
Meta Superintelligence Labs vs Facebook AI Research (Jan 30, 2026)
After delivering influential models like LLaMA and SAM, Meta’s AI research had been the subject of ongoing debate and speculation.
Instead of engaging with the headlines, the AI by Hand Frontier Seminar took a quieter approach, letting the papers speak for themselves, limited to what Meta was willing to openly publish.
I walked through two of the most recent publications from FAIR and two from MSL, using them to understand what Meta was actually investing in and where its AI research direction appeared to be heading.
Our guest industry expert was Yichen Wang, a Research Scientist at Meta working on GenAI safety.
The four papers covered in the seminar were as follows:
Facebook AI Research (FAIR)
Text-Guided Semantic Image Encoder
https://arxiv.org/abs/2511.20770 (11.25.2025)
Safety Alignment of LLMs via Non-cooperative Games
https://arxiv.org/abs/2512.20806 (12.23.2025)
Meta Superintelligence Labs (MSL)
REFRAG: Rethinking RAG based Decoding
https://arxiv.org/abs/2509.01092 (10.12.2025)
Dr. Zero: Self-Evolving Search Agents without Training Data
https://arxiv.org/abs/2601.07055 (1.11.2026)
Outline
Transformer
LLM (Decoder-Only)
RAG
📄 MSL Paper #1 : REFRAG
Vision Transformer
Vision-Language Model (VLM)
📄 FAIR Paper #1: Text-Guided Semantic Image Encoder
Industry Expert Interview
📄 FAIR Paper #2: Safety Alignment of LMs via Non-cooperative Games
GRPO
📄 MSL Paper #2: Dr. Zero- Self-Evolving Search Agents without Training Data
Feedback
Recording & Workbook
Become a paid subscriber to access the full recording and the associated Excel workbook.