Neural networks are not getting smarter, but they are learning to correct their mistakes very quickly. First of all, I was on the “Reflection” program on the neuroses of OPG-4. The name of the ideal natural essence – the compilers of the system, receiving the neural network, evaluate the frequent work and create recommendations for themselves to correct errors.
This system creates tests for self-testing the work of the neural network, but it does it faster, faster and more efficiently. On a non-universal basis, this time the claims proved that I pressed the button and proved it. But the result is stoiorn: on, in t “Humaneval” by p p p p p p ryaz that language After the application of “Reflexy” – 88%.
If a neural network can acquire a certain miniature, then it can also evaluate the quality of its implementation and give a construct. In the Alfworld multi-problem test, the application of the communication novelty increased the efficiency of GPT-4% from 9.7% from 9. In the HotPotQA test, analyzing data from Wikipedia, it ranked from 34% to 53%. The average estimate shows an increase in the efficiency of the neural network with a self-control system by 30%.
Unfortunately, GPT-4 still makes fundamental errors because system discovery in equs Adding a number of “Reflection” significant false positives, but since both executive and verifiers are the same system, architecture discovery cannot recognize its own blunders if they were not used on them from the outside. This makes this tool convenient and limited, provided that its p +
Source: Tech Cult
I am a professional journalist and content creator with extensive experience writing for news websites. I currently work as an author at Gadget Onus, where I specialize in covering hot news topics. My written pieces have been published on some of the biggest media outlets around the world, including The Guardian and BBC News.
