Meta AI AI News, Models and Product Updates
Track latest Meta AI AI news, launches, research, and ecosystem moves.
Meta Ai news, model releases, product launches, research updates, and major announcements in one place.
Latest Articles
- Query-based Submodular Data Selection for Efficient Instruction Tuning — QSGS selects instruction-tuning data via a submodular graph cut over gradient similarity and fills gaps with synthetic examples, matching or beating full-data performance on Mistral-7B with just 0.5% of the data.
- Balancing Chain-of-Thought and Answer Supervision for Small Language Models on Medical Question Answering — For small medical QA models, segment-weighted chain-of-thought loss works best on multiple choice while uniform weighting wins on free text, so fine-tuning should match the task format.
- Beyond Fine-Tuning: Evaluating Prompt Engineering and Prompt Tuning for Roman Urdu Hate Speech Detection — A study comparing prompt engineering and prompt tuning for Roman Urdu hate speech detection finds a mixed prompt configuration reaches up to 0.87 F1, well above zero-shot baselines.
- Anchor: Augmented Cross-Layer Recurrent State for Long-Context Retrieval in Test-Time Training — Anchor places test-time-training memory only at a few anchor layers with a shared cross-layer state, giving 2.3-3.1x higher recall than LoLCATs on Llama-3.1-8B and Mistral-7B with a 10x smaller recurrent state.
- Perseus: A Full-Stack System for Homomorphic Model Inference — Perseus is an open Python library for fully homomorphic encrypted LLM inference that reports 3.96x faster single-token decoding than other implementations on Llama-3-8B, within 3% of plaintext accuracy.
- Anthropic Releases FlashProve: Cryptographic Proofs That a Reply Came From Committed Weights — Anthropic released FlashProve, an Apache 2.0 zkSNARK prototype that proves a model reply came from committed weights without revealing them, running Llama 4 Maverick at 110 tokens per second on one H100.
- Hacking or Just Failing? A More Capable LLM Monitor Better Separates Reward Hacking from Honest Failure — A pre-registered study finds small local LLM monitors cannot tell reward hacking from honest failure, Gemini 3.1 Flash-Lite does better at AUC 0.80, and a two-rule deterministic check caught every known exploit.
- Value-Conditioned Proposal Acquisition for Test-Time Reasoning — ValueFormation decides at test time whether to stop, keep sampling, or call a reasoning specialist, beating eight baselines on MATH-500 by over 15 points for three 8B models.
- Cross-Prompt Reuse of Entity-Replacement Responses — A new arXiv paper measures how much of an entity-replacement hidden-state change transfers between factual questions in Qwen, Llama and Gemma models, finding gains that are partial and format-dependent.
- LoRAForge: Co-Designing Post-Hoc Compression and Execution for Efficient Multi-LoRA Serving — LoRAForge compresses trained LoRA adapters and pairs them with rank-specialized kernels so smaller adapters actually serve faster, keeping about 7.7% of parameters with at most a 1.2-point score drop.