So, VECS 2024 is coming to an end, what have we learnt?
I think that I’ve experience an exciting two days full of talks. The most inspiring one, for me, was the talk about Volvo Cars build processes:
$ bezel build //car
I think this summarizes the Software Defined Vehicle. We’re not there yet, since virtualization and containerization is not fully ready, but we are close.
Some food for thought, inspired by Zenseact’s talk (my own words):
— how do we create AI models that evolve over time and do not require so much updates – this is an interesting question for the future
— timing constraints are the most important ones – you must be able to track and identify objects 50 FPS or more, otherwise it gets difficult from the practical perspective
— Large Vision Language Models – multimodal llms
— use bulk data for pre-training, trillions of tokens, especially when you add videos to these models; not possible to annotate automatically
— https://arxiv.org/abs/2401.13601
— https://github.com/NExT-GPT/NExT-GPT
— https://github.com/BradyFU/Awesome-Multimodal-Large-Language-Models
— https://arxiv.org/abs/2403.09611
— https://arxiv.org/abs/2306.13549
I think that we’re now living in a super-exciting time in automotive software engineering. Plus, it was quite refreshing to listen to something more than LLMs in software development.