Adevinta set out to quantify the impact of ChatGPT and GitHub Copilot on its thousands of developers. They measured adoption, accepted lines of code, and self-reported workflow acceleration.
Mistral is working to find a balance between openness and business success. It has a new license aimed at striking that balance. It will continue releasing other projects under Apache 2.0 as well as the new MNPL license.
A new model from the creators of SSMs, Mamba, and sub quadratic Transformer variants. Importantly, their new company Cartesia has trained a lightning-fast, realistic voice generation model. This points to their desire to move into the assistants' space.
OpenAI has signed licensing deals with The Atlantic and Vox Media, allowing their content to train its AI models and be shared in ChatGPT with proper attribution.
Mistral released a 22B code model with 32k context. It is a powerful and fast model with broad performance across many programming languages. It has open weights and is available via Mistral's platform.
PRISM is a foundation model trained on 1.2 billion small molecule mass spectra to enhance mass spectrometry analysis in drug discovery. It uses self-supervised learning to predict molecular properties from complex mixtures without prior annotations, significantly advancing the field of metabolomics. The model improves the identification and understanding of largely unknown natural molecules, accelerating the discovery of new therapeutics by leveraging massive, previously untapped spectral data.
DMPlug is a novel plug-in method that uses pretrained diffusion models (DMs) to solve inverse problems (IPs). Unlike existing interleaving methods, DMPlug views the reverse diffusion process as a function, addressing both manifold feasibility and measurement feasibility effectively.
OpenAI has signed PwC as its largest enterprise customer for ChatGPT, covering 100,000 users. PwC is the first partner to resell OpenAI's enterprise offerings.
Scale has made a private leaderboard for evaluating language models. The ordering isn't too surprising, but one interesting finding is the Llama 3 70B is extremely good at following instructions, often better than Claude Opus.
Come shape the future of AI with passionate engineers, researchers, and like-minded individuals. 2 days, 3 tracks, 60+ speakers — in-person in SF. Get $200 of with TLDR24 discount code!