This new research is important because it challenges the wisdom of the AI development, which usually relies on a large -scale training datases and computational expensive models. While well -known AI companies emphasize the ever -trained larger models trained on more broader datases, compressor has primarily suggested the emergence of different principles.
Researchers concluded, “Compressor’s intelligence does not come out of pretarning, wide datases, full search, or mass computies – but from compression.” “We challenge the traditional dependence on broader prescriptions and data, and make a future suggestion where compressive goals and effective inventory time computations work together to withdraw deep intelligence from the least input.”
Are looking forward and looking forward
Even despite its achievements, the system of Liao and Go comes with clear limits that can give rise to doubts. Although it successfully solves colorful assignments, inflammation, crop, and adjoining pixels, it is struggling with tasks that require counting, long -distance patterns, rotation, reflection, or agent behavior. These boundaries highlight areas where simple compression principles cannot be sufficient.
This research has not been reviewed, nor is the 20 % accuracy on the puzzles seen, although not remarkable without training, but the human performance and the high AI system fall significantly. Critics may argue that compressark arc puzzles can exploit specific structural patterns that cannot be normalized by other domains, challenging whether only compression is the basis of a wider intelligence rather than just one component in many people.
And yet the development of AI is continuing its rapid progress, if the compressor maintains further scrutiny, it offers a glimpse of a potential alternative path that can lead to useful intelligent behaviors without the requirements of today’s dominant perspective. Or at least, it can open an important component of general intelligence in machines, which is still considered bad.