Recently Jensen from NVIDIA hit X (twitter) claiming that, once again, AGI is there and went on to congratulate OpenAI on the release of a new model. Of course, this is a 100% marketing play since OpenAI trained the new model on Nvidia hardware and also because there is no “hard” definition of what AGI (Artificial General Intelligence) really is. Some say it’s an AI model that can surpass human intelligence but, really that is just saying this new model has more data parameters and knows more stuff. That data has to come from somewhere and while some people think AI is magical, its really just predicting the next logical word based on the available data. Sometimes that word is correct and other times it just sounds correct.
This can then beg the question, “How do you attain AGI?”
One direction that Dennis believes is that AGI is not a single LLM but an entire system of LLMs which allows particular models to be experts on a certain subject allowing the orchestration model to route questions to the correct model to get the best answer. This doesn’t solve the issue of learning something it wasn’t trained on but would effectively allow the orchestration model appear to “learn” similar to picking up new skills by reading a book.
The remainder of this episode is a semi deep dive into using OpenClaw and how your model selection can make or break success when using AI Agents. Most users will hook agents into their cloud models which can quickly break the bank in token usage. For instance, the week of research used to prepare for this podcast could have cost almost 3000USD had it not been hosted on a Local LLM.
Related Links
HuggingFace
Balancing Model Quality and Hardware Demands in AI Workstations
Building a Multi-GPU AI Workstation on a Budget
Software Choices for a Multi-GPU AI Workstation
Fine-Tuning LLMs on a Local Multi-GPU AI Workstation
AGI has arrived according to the guy selling shovels
Episode 184 featured music:
Little People - Start Shootin' (http://www.littlepeoplemusic.com/)
