What I mean here is that traditional LLMs are trained on tasks irrelevant to what they will do for the user. It’s like training a plane to efficiently operate on the runway, but not to fly. In short, it is almost impossible to train an LLM, and evaluating is just as challenging. Then, training is not even necessary. In this article, I dive on all these topics.
➡️ Training LLMs for the wrong tasks
Since the beginnings with Bert, training an LLM typically consists of predicting the next tokens in a sentence, or removing some tokens and then have your algorithm fill the blanks. You optimize the underlying deep neural networks to perform these supervised learning tasks as well as possible. Typically, it involves growing the list of tokens in the training set to billions or trillions, increasing the cost and time to train. However, recently, there is a tendency to work with smaller datasets, by distilling the input sources and token lists. After all, out of one trillion tokens, 99% are noise and do not contribute to improving the results for the end-user; they may even contribute to hallucinations. Keep in mind that human beings have a vocabulary of about 30,000 keywords, and that the number of potential standardized prompts on a specialized corpus (and thus the number of potential answers) is less than a million.
➡️ Read the full articles at https://mltblog.com/3CEJ9Pt, also featuring issues with evaluation metrics and the benefits of untrained LLMs.
In this article, you will find my PowerPoint presentation describing the most recent features of xLLM, a CPU-based, full context, secure multi-LLM with real-time fine-tuning & explainable AI. It includes several new diagrams describing the innovative architecture, upcoming developments, new features and different use cases.
Content
➡️Enterprise use case: corporate corpus of a Fortune 100 company. ➡️Original version dealing with large websites such as Wolfram and Wikipedia. Comparison with OpenAI. ➡️xLLM for clustering and predictive analytics. Use case: unstructured text (articles) from a media company. ➡️Integration of our game-changing NoGAN tabular data synthesizer, and state-of-the-art model evaluation technology. ➡️Integration of external tools, for instance to solve math problems. ➡️Upcoming version for auto-indexing and cataloging large repositories. ➡️Demo: enterprise xLLM in action, featuring the modern user interface (full web API, not just a prompt box) with command menu and numerous options not found in other LLMs, including debugging, suggested prompts, choice of agents, and fine-tuning in real time. ➡️Relevancy score displayed to the user, for each returned item. I call it the new PageRank for RAG/LLM, using a technology radically different from Google search. See picture.
New startup coming soon!
We will be launching soon (January) a new startup focusing on GenAI at scale for Enterprises; xLLM will be part of the offer with exclusive features. We are looking for early adopters to partner with us on the Journey. The co-founder and CEO, to be announced soon, is Senior Director of GenAI at a Fortune 100 company, where the first version of Enterprise xLLM was implemented. More to come!
