Unverified50% confidenceBenchmarkExact time
Averaged over 3 random seeds at the same learning rate, good views (two different descriptions of the same photo) reached 0.320, bad views (a single description with some words masked) reached 0.192, and next-word prediction reached 0.148.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
11/2/2026
First Seen
Valid until: 1/31/2027
Sources
Related Entities
Related Claims
UnverifiedThe authors performed a small learning-rate search for each method, taking the best rate with single-seed runs, yielding 0.328 / 0.196 / 0.168, with the ranking unchanged, showing that the lead is not an accidental product of learning-rate tuning.74% similarUnverifiedRandom accuracy is about 0.33 for three-class classification and about 0.25 for four-class classification; only the magnitude of improvement relative to the random level can reflect whether the model has truly learned anything.72% similarUnverifiedIn the fact-updating task, the success rate of the embedding retrieval method fluctuated between 0.30 and 0.95, with a maximum gap of 0.45 across different random seeds.71% similarUnverifiedOn a small CIFAR setup, typical inter-seed fluctuation is approximately ±0.3–0.5 percentage points, and GPU noise under the same seed falls within the same statistical range.65% similarUnverifiedBecause the experiment used temperature=0 and a fixed seed in a single deterministic run, any field flip moves the score by about 2.8 percentage points out of 12 tasks × 3 fields = 36 units, so the gap between 91.7% and 83.3% cannot be directly generalized into a statistically meaningful general advantage.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1127611API
curl https://kongchang.com/api/v1/knowledge/claims/1127611MCP
get_claim(id=1127611)