AI research | The Decoder

Jonathan Kemper

Jul 22, 2024

AI research

Graph RAG: Access to external data becomes much more accurate with Microsoft's approach

Abstract AI illustration of a colorful network graph.

Matthias Bastian

Jul 21, 2024

AI research

Researchers uncover an all-too-easy trick to bypass LLM safeguards

Editorial illustration of a burning Molotov cocktail

Maximilian Schreiner

Jul 21, 2024

AI research

Streetscapes AI generates uncannily realistic Street View scenes of entire cities from scratch

Matthias Bastian

Jul 20, 2024

AI research

AI models could surpass human skills by using weaker models for guidance

A cute widescreen low-poly illustration of a baby llama teaching a big llama. The baby llama is standing on a small mound with a confident expression, pointing to a chalkboard with its hoof. The big llama is sitting attentively with a curious look, slightly tilting its head. The background is a bright and colorful outdoor setting with green grass, blue skies, and a few fluffy clouds. There are small details like flowers, butterflies, and a sun wearing sunglasses to add to the playful and adorable atmosphere.

Matthias Bastian

Jul 19, 2024

AI research

AI models might need to scale down to scale up again

Ex-OpenAI researcher explains the backwards race in AI models

Jonathan Kemper

Jul 19, 2024

AI research

Mistral releases three new LLMs for math, code and general tasks

Maximilian Schreiner

Jul 19, 2024

AI research

Microsoft's SpreadsheetLLM can tackle huge scientific and financial spreadsheets

Matthias Bastian

Jul 18, 2024

AI research

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

AI research

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

Matthias Bastian

Jul 17, 2024

AI research

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.