Rule-based rewards (RBRs) use model to provide RL signals based on a set of safety rubrics, making it easier to adapt to changing safety policies wo/ heavy dependency on human data. It also enables us to look at safety and capability in a more unified lens as a more capable
RESEARCH
-
Large Models Generate Synthetic Data for Smaller Model Training
By
–
This is going to be the most impactful piece — enabling everyone to use a large model to create synthetic data to train performant smaller ones.
-
Harvard team maps human brain connectome in unprecedented detail
By
–
#ICYMI: Decade of brain decoding: Our Connectomics team, alongside @Harvard
, mapped every connection within a piece of human brain. One leading neuroscientist called the detailed map, which required 1.4 petabytes to encode, “a tour de force creation”. -
Emergency Paper Club Discusses Llama 3 Herd Models
By
–
EMERGENCY PAPER CLUB The @latentspacepod discord is meeting in 2hrs to talk thru @lvdmaaten et al's The Llama 3 Herd of Models, early contender to win the POTY* Awards! Join us (link below) with @swyx
, @vibhuuuus
, @picocreator
, @eugeneyan
, et al! *Paper of The Year, totally -
Mistral Large 2 Update Challenges GPT-4o and Sonnet
By
–
¡MISTRAL LARGE 2! Los franceses actualizan su modelo Large llevándolo a estar cerca de los mayores (GPT-4o, Sonnet 3.5). Es una mejora importante respecto al anterior modelo Mistral Large, pero claro… Muy a la par con Llama 3.1, que a diferencia de este modelo, es open
-
Mistral Large Improved Alignment and Instruction Capabilities Performance
By
–
Compared to the previous Mistral Large, much more effort was dedicated to alignment and instruction capabilities. On WildBench, ArenaHard, and MT Bench, it performs on par with the best models, while being significantly less verbose. (4/N)
-
Mistral Large 2 outperforms Llama 3.1 on Multilingual MMLU
By
–
On Multilingual MMLU, the performance of Mistral Large 2 significantly outperforms Llama 3.1 70B base (+6.3% average over 9 languages) and is on par with Llama 3 405B (-0.4% below). (3/N)
-
Mistral Large 2 outperforms Llama 3.1 405B on coding benchmarks
By
–
On HumanEval and on MultiPL-E, Mistral Large 2 outperforms Llama 3.1 405B instruct, and scores just below GPT-4o. On MATH (0-shot, without CoT) it only falls behind GPT-4o.
(2/N) -
Mistral Large 2: New 123B Model Outperforms Llama 3.1 405B
By
–
Today, we release Mistral Large 2, the new version of our largest model. Mistral Large 2 is a 123B-parameter model with a 128k context window. On many benchmarks (notably in code generation and math), it is superior or on par with Llama 3.1 405B. Like Mistral NeMo, it was trained
-
Show Me The Prompt: Advanced LLM Prompt Engineering Guide
By
–
– Show Me The Prompt. https://
bit.ly/4cT6N7B
#AI #MachineLearning #DeepLearning #LLMs #DataScience