AIIndex pretraining’s comparison baseline draws criticism
One participant argues that the comparison is against no pretraining—not vision-language model pretraining—and calls that a weak baseline.
TLDR
One participant said their criticism targeted a video’s marketing claim of “doing real useful work,” not the engineering behind it. In reply, another argued that comparing an index-pretrained model with one trained from scratch sets a weak baseline. They said the claim is about outperforming no pretraining—not vision-language model pretraining.
Combined views
214
1 Source, first seen 3h ago
Index pretraining’s comparison baseline draws criticism
One participant argues that the comparison is against no pretraining—not vision-language model pretraining—and calls that a weak baseline.
TLDR
One participant said their criticism targeted a video’s marketing claim of “doing real useful work,” not the engineering behind it. In reply, another argued that comparing an index-pretrained model with one trained from scratch sets a weak baseline. They said the claim is about outperforming no pretraining—not vision-language model pretraining.