Researchers at the Universitat Politècnica de València, in Spain, have found that as Large Language Models (LLMs) grow larger and more sophisticated, they are more likely to span sensible yet wrong answers, making them less reliable. Researchers analyzed three popular LLMs; GPT by OpenAI, the LLaMA of Meta, and the BLOOM suite developed by BigScience, and noticed that with new versions accuracy increases, so do the hedging, refusal, or evasiveness. The study also found that LLMs rarely admit to a user that they do not know an answer.
Scaled-up LLMs are more prone to sensible yet wrong answers
Making it less reliable
- Advertisement -
Explore more ..
Study uncovered earliest evidence transport technology used by ancient settlers
Drag marks reveal how our ancestors moved heavy loads.
Researchers from Bournemouth University discovered early evidence of transport technology used by settlers in the USA over 20,000 years ago at White Sands National Park, New Mexico. They found drag marks from makeshift vehicles called “travois,” made from wooden poles, alongside ancient human footprints. The tracks included single lines, likely from dragging two poles joined at one end, and parallel lines, from poles crossed in the middle.
The quantum processor that redefines computing speed
It is one quadrillion faster than the fastest supercomputer.
Researchers from the University of Science and Technology of China (USTC) have created a superconducting quantum computing prototype called Zuchongzhi-3. This prototype operates with 105 qubits and 182 couplers, making it 1015 times faster than the fastest supercomputer and one million times faster than Google's latest results. This breakthrough marks a significant advancement in quantum computing performance.
MIT Researchers develop improved validation method for more accurate spatial predictions
MIT Improves Spatial Prediction Validation
Researchers at MIT have tackled a perplexing issue in spatial prediction validation, which is critical for fields like weather forecasting and environmental health. Traditional validation methods often fail because they don’t account for the unique characteristics of spatial data, leading to inaccurate forecasts. The research will be presented at the International Artificial Intelligence and Statistics Conference.
Breakthrough algorithm lets you do speech recognition offline
Researchers develop offline speech recognition algorithm
Researchers at the University of Copenhagen have developed a new algorithm that eliminates the need for an internet connection to use speech recognition. This algorithm is proficient at decoding speech without requiring substantial memory allocation. It forgets what it doesn't need in real-time by employing a clever strategy.
TRENDING
- Advertisement -