what makes you say that? given a fixed amount of time it seems very difficult to tell who's AI these days
@jxmnop
-
GPT-2 Negation Understanding: From 2019 Struggles to Modern Progress
By
–
i fondly recall arguing in 2019 with another researcher about how whether models understand negation. this really was a major sticking point at the time, GPT-2 couldn't reliably distinguish between "I did like that" and "I didn't like that". at some point we fully blew by the
-
Per-User Context Prompt Optimization for AI Personalization
By
–
btw i suspect you could simulate this level of personalization with effective per-user context-specific prompt optimization
-
Zero-Cost Model Serving: SFT vs RLHF Strategy
By
–
hypothetical situation – i am an AI company that's reduced the cost of transferring and storing models to zero. i can serve each user their own model with no overhead what do i do? directly SFT user-specific models on their data? or RLHF on the chat ratings? something else?
-
Brain Vision Processing Gap in Modern LLMs Architecture
By
–
the human brain reserves 40% of its processing exclusively for vision. modern LLMs somehow evolved without this entirely
-
Image pretraining impact on text-based AGI evaluation benchmarks
By
–
maybe our crux here is that most of the 'AGI evals' are text-based. so my point is that adding image pretraining doesn't help on any of the (text-based) evals.
-
MLPs and Transformers: Key Components for AI Intelligence
By
–
ah i guess you could argue CV gave us MLPs which are a part of transformer which is important for intelligence. that's a fair point
-
Linguistics Approach Limitations in AI Development
By
–
that's actually a good point. linguistics turned out not to help either.
-
Computer Vision Research Failed to Advance AGI Progress
By
–
very surprising that fifteen years of hardcore computer vision research contributed ~nothing toward AGI except better optimizers we still don't have models that get smarter when we give them eyes
-
Test-Time Compute: The Evolution of AI Model Optimization
By
–
because this was before anyone cared about ‘test-time compute’