AI research | The Decoder

Matthias Bastian

Jul 18, 2024

LLMs hit a wall when processing complex information from lengthy texts

Researchers have discovered that special AI models can solve complicated tasks just as well if they only use points such as "......" instead of whole sentences. This could make it more difficult to control what is going on in these models.

Maximilian Schreiner

Jul 18, 2024

Google Deepmind develops open-source AI to tackle biases in evaluating language models

Matthias Bastian

Jul 17, 2024

OpenAI pits AI models against each other to produce clearer, more verifiable text

OpenAI has created a new method to enhance the clarity and verifiability of AI-generated text. This approach could boost confidence in AI systems and expand their use in critical fields.

Maximilian Schreiner

Jul 17, 2024

Google Deepmind's new PEER architecture uses a million tiny experts to boost AI efficiency

Maximilian Schreiner

Jul 16, 2024

Nvidia researchers develop DoRA, a smarter way to fine-tune AI models for specific tasks

Maximilian Schreiner

Jul 16, 2024

Wild Gaussians: New AI method enables 3D reconstruction from user-captured web photos

Matthias Bastian

Jul 15, 2024

Update

OpenAI's Strawberry project reportedly cracks over 90% of tough math contest problems

Illustration of a strawberry being generated in an artificial neural network.

Matthias Bastian

Jul 14, 2024

Language models like GPT-4 memorize more than they reason, study finds

Clean, sleek illustration of an upside-down lemonade glass.

Matthias Bastian

Jul 12, 2024

Researchers combine two language models and a database for more accurate LLMs

The "Speculative RAG" framework combines two language models to make Retrieval Augmented Generation (RAG) systems more efficient and accurate.

Matthias Bastian

Jul 12, 2024

OpenAI unveils five-level AI scale, aims to reach level 2 soon