Underrated as defined by:
To determine the most underrated AI lab of 2024, evaluate labs with strong research and innovation, including high performance models, but limited X/Twitter chatter, media coverage, or industry recognition by analyzing their contributions, funding, and expert opinions. Focus on those achieving significant results with less attention.
For example: Mistral Large 2 is 1/4 the size of Meta Llama 3.1 405b, with similar performance, but nearly no press or X/Twitter coverage, and when covered, is often mentioned as being superior to Llama 3.1
Market resolution:
Market will resolve based on your votes (aka yes/no orders). The top 1 resolves to YES.
To identify the most underrated AI lab of 2024, several candidates stand out based on their innovative contributions and relatively low media visibility. Here are some notable mentions:
Mistral: Known for its Mistral Large 2 model, which is significantly smaller than Meta's Llama 3.1 yet offers comparable performance. Despite its capabilities, Mistral has received minimal press coverage, making it a strong contender for being underrated.
DeepSeek AI: This lab focuses on advanced AI research but lacks the media attention that larger players receive. Its innovative approaches in AI development have not been widely recognized, which positions it as an underrated entity in the field.
Sakana AI: Founded by a co-inventor of the transformer architecture, Sakana AI is working on new AI architectures that aim to improve efficiency and scalability. Despite its promising innovations, it has not garnered significant media attention, which could classify it as underrated.
Liquid Neural Networks (LNNs): Developed at MIT, LNNs offer a novel approach to AI that adapts dynamically to data, enhancing efficiency. This innovative technology has the potential to reshape AI applications but remains under the radar in terms of public recognition.
These labs exemplify strong research and innovation while lacking the widespread recognition enjoyed by larger organizations like OpenAI or Google DeepMind.
Deepseek definitely fulfills this criteria, top 3 coding model overall, all models open source + the cheapest API on the market (cheaper than gpt 4o mini)
https://huggingface.co/spaces/bigcode/bigcodebench-leaderboard
@0xSMW Could you confirm what coproduct said? how will you count the votes and do you resolve YES to the one with the most votes and NO to all others or some other scheme