In my HRI lab, the main issue that we have been trying to tackle is precision over many categories and contextual emotion recognition. Gpt-4o needs to be benchmarked properly to compare it to current research (paper forthcoming).
RESEARCH
-
Embodiment and Acoustic Context in Human-Robot Interaction Systems
By
–
I would need to try it out also to know more, to see how it handles other examples. They only demo'd a smile. And HRI thinks a lot about embodiment. For example we should take into account acoustic context eg. https://
arxiv.org/abs/2205.04952 -
Multimodal Emotion Analysis: Early Fusion Techniques and Historical Context
By
–
I agree that the integration is impressive, esp. early fusion of 3 modalities at scale. But multimodal emotion analysis and other kinds of fusion have been well studied, check out work by @lpmorency I also worked on this ~10 years ago using late fusion
-
Who Overhypes AI Capabilities When Convenient?
By
–
Lo que pongo en el hilo. Ha pasado un año y los grandes males que iban a llegar con sistemas como AutoGPT aún a día de hoy no se han cumplido (y AutoGPT es un ejemplo de los muchos que podría poner) Entonces… ¿quién overhypea las capacidades de la IA cuando le conviene?
-
AutoGPT Fears: Evidence After One Year
By
–
"We’ll see what actually happens over the next 12 months" A year has already passed. Can you provide evidence that supports your fears about systems like AutoGPT?
-
GPT-4 Performance Gains: 30 Points Benchmark Jump Since March
By
–
Tomarse enserio a Marcus es creerse que desde Marzo del año pasado (ni año y medio) GPT-4 ha seguido igual… Cuando la realidad es que en algunos benchmarks el salto es de 30pp (rosa vs amarillo) y en el LLM Arena supera en 100 puntos de ELO al GPT-4 original.
-
Discovering Hidden Tokens Inside AI Tokenizers Easter Egg
By
–
Finding weird tokens inside tokenizers is the new Easter egg tech hunt
-
Prompt technique: hiding chain-of-thought in code comments
By
–
Teaching ChatGPT to “think silently” by hiding planning, chain-of-thought, or other intermediate writing in interpreted Python code blocks as comments or strings. Text remains visible to the model (and the user, by clicking to view code) without cluttering the chat dialog.
-
Prototype-First Learning vs Endless Definition Discussions
By
–
Yeah I agree. But the sentiment of my comment was based on my experience with people:
1) some will come up with some simple definition and then start building prototype and then learning from it and iterating
2) some will have never ending discussions about the definition and -
Tensor Algebra: Scaling AI Architectures Efficiently with TensorLy
By
–
Tensor algebra is the way to scale AI architectures in an efficient way. We have been working on it for more than a decade. Check out http://
tensorly.org led by @JeanKossaifi with @PyTorch framework