Depends. If you are serving millions of customers this type of inference scaling is going to cost you more than training with RL.
MACHINE LEARNING
-
Image Format Complexity Requires More Training Data
By
–
exactly, that’s the messiness of working with the image format I mentioned. I think you can make to generalize well on all these but since there are more degrees of freedom it will require more data to train (luckily this can be done with automatic data augmentation but still)
-
How Constraints Drive Innovation in Neural Network Architecture
By
–
Yeah, I think that’s what brought us CNNS (as opposed to fully connected neural nets), LoRA, and many more
-
Reconsidering VLLM terminology at 3B parameters
By
–
Actually I was thinking about it when typing, and I don't know. I don't want to be that person who goes against the common terminology (like softargmax haha) but it really is a V*L*LM at 3B parameters.
-
DeepSeek OCR Paper and Multimodal LLM Article Resources
By
–
Link to the paper: https://
github.com/deepseek-ai/De
epSeek-OCR/blob/main/DeepSeek_OCR_paper.pdf
… My "Understanding Multimodal LLMs" article with more info on how images are fed to LLMs, how cross-attention works, etc: https://
magazine.sebastianraschka.com/p/understandin
g-multimodal-llms?utm_source=publication-search
… -
Visual Tokenization Challenges: Aspect Ratios and Image Preprocessing Complexity
By
–
I know it’s popular to hate tokenizers, but visual representations (which are also tokenized) bring a lot of messiness as well. Aspect ratios, cropping, resolution, brightness, etc. Sure, models learn to deal with that but it requires lots of data to make them robust wrt these.
-
World simulation models like Genie 3 move beyond pixel prediction
By
–
the solution? it’s not better video models. it’s world simulation models like google’s genie 3. genie doesn’t just predict pixels – it simulates physics rules. gravity, collision, momentum. it understands that when you climb a ladder, your hand grips, your foot lifts, your
-
Karpathy’s exposition highly recommended for scientists and engineers
By
–
I just listened to it and wholeheartedly recommend it. @karpathy
’s exposition is very informative and aligns well with what most hands-on scientists and engineers believe. -
Prioritizing Knowledge Reduction Techniques Over Terminology
By
–
I’m more interested in reductions than what one calls it. Knowledge of reducing, eg of dagger to distillation, can be very helpful to decide what to do with a finite compute budget.
-
AI 2025: From Prompts to Purpose with Autonomous Agents
By
–
AI in 2025 won’t be about prompts, it’ll be about purpose. We’re moving from reactive systems to autonomous agents that sense, reason, and act independently. Mastering AI now means thinking like an Agentic Architect, someone who designs systems that learn, adapt, and