Running LLMs locally is possible and very useful as long as you manage expectations.
I'm excited to share that we are organizing the AI in Production Conference, on July 18 & 19, 2024, in Asheville, NC.
The GPT store is here but so is the new teams plan and how they interact is very interesting.
Can AI be used to create fun intelligent introductions at social, networking and business events?
The heart of a transformer is the attention module. Lately there's been a lot of work to address the drawbacks of large attention modules. If any of these approaches work out we could see much smaller and faster LLMs in the future.
In NLP entity linking is the entails labeling spans of text with pre-determined concepts. There is a new Driven Data competition for medical entity linking.
OpenAI recently came out with a paper title "Practices for Governing Agentic AI Systems" ... but there are more than three laws.
AI is going to change our world. What framework do you use to create and evaluate various scenarios of how that will happen?
LLMs are a powerful but slow and expensive way to do text classifications. N-Nearest neighbors with embeddings may be faster and easy.
In traditional software development the we have checklists for putting an app into production. What should be on that list of an AI based system?
Many of us are focused on the largest and "best" LLMs available but small models deserve a look also.
The mixture of experts (MOE) approach has gotten a lot of attention lately and will continue to be a research focus.
As professional software developers we want to deliver high quality software but how do we know for sure which tools and processes help us towards that goal.
A few days ago Apple quietly announced two Python based machine learning open source projects but what is the logic behind them and who are they for?
GenAI advancements are staggering and may affect programming jobs first.
Agents have been studied for many years and recently there has been increased interest in LLM based agents.
Synthetic data has become a hot topic in generative AI is there more than than first meets the eye.
Key points from a recent article discussing the changes that organizations may/will go through in response to AI.
The first weekly summary of a shorter more frequent writings on thoughts and impressions of AI. The single article is on Implementing a Retrieval Augmented Generation (RAG) System.
Many people consider running an open source LLM on their own hardware but are unsure where to start. In this article I highlight some of the issues that need consideration.
We've had a lot of interest in Low and No code AI tools so I've put together this basic introduction.
Artificial Intelligence (AI) and Machine Learning (ML) are making their way into our lives more and more. In this article I'll define some of the terms that I get asked about most frequently.
Vector databases have been getting a lot of attention but what are they and are they strictly necessary?
Many people make the assumption that they have to train an Large Language Model (LLM) on their data to get better results. Though that would definitely improve responses I suggest that prompt engineering may be sufficient and is faster, cheaper and easier to implement.
Recently I had the pleasure of speaking to local business and education leaders at the ChatGPT-AI Forum put on by Mountain Area Workforce Development Board of North Carolina. This is a summary of the speaker notes from that talk.
Recently I was working on analyzing some short texts and came up with an idea for extracting interesting themes from them. I thought the technique might be particularly useful for app/product reviews. Many businesses are interested in analyzing sentiment but this goes beyond that and tries to analyze recurring themes automatically.
I've been working with a client on analyzing some text documents and wanted to share a bit of what has been working for us. I can't share the data or the exact project details but it entails, finding similar text documents from a large collection of other documents a specific query example. Imagine searching a database of company statements, product descriptions, articles, contracts, emails, support/trouble tickets, etc. not by keyword but by 'meaning' and 'similarity'.
I'm a data scientist (machine learning engineer) and would like to use my skills to help others in my community. I'd also really like to help others help others. So I'm starting a project to bring together volunteer data scientists, engineers and students with non-profits and organizations to find opportunities to do good in their own communities.
Customer churn, the percentage of customers that stop using your product or service in a particular time period, can quickly become disastrous to your revenue. Acquiring new customers is more costly than keeping existing ones and even a reasonable sounding churn rate can result in a leaky bucket that is impossible to fill.
Automatically finding similar and duplicate images can be very useful as a quick way to show similar products or items from a collection of images. For example, I was shopping for a phone case and the online store had many many interesting designs but they were hard to navigate. Once I found a case that I _kind of_ liked I wanted to see other similar cases to find one that I _really_ liked. Unfortunately they only showed other popular cases that were not at all similar to the one I was considering.
Lately there has been a lot of interest in explainable AI/ML. Nobody wants to feel discriminated against by an algorithm and when we don't like its prediction or decision we want to know why it made that decision. Plus there is an added sense of security when we feel we understand (or could understand) how something works.
In previous articles we worked through basic approaches for text classification by presenting a simplified version of a problem posed by a client and examining the performance of several algorithms. In this article we improve (slightly) the performance of one of the algorithms with a grid and random hyper parameter optimization search.
In this article we talk about using the next simplest approach which TF-IDF with basic classifiers from Scikit-Learn (sklearn). We show that with minimal processing and no parameter tuning at all we get the impressive accuracy.
Recently I had a request from a client for help classifying short pieces of text. The exact nature of the text is confidential but the passages were similar to paragraphs from reviews and comments users write about products and services. We wanted a quick and easy way to establish a baseline we could use to compare various approaches. This would help us decide, based on performance and expected necessary investment, if further efforts in research, development and operations were necessary.
We're hosting an unconference March 24, 2018 in Portland, Oregon focused on finding ways to use AI to improve the lives of everyone in the Community.
The ACA Bot is a very early version of a chatbot that tries to answer some basic questions related to the Affordable Care Act.
There is a lot of hype around Artificial Intelligence (AI) and Machine Learning (ML). Its been called *'the new electricity'* and many believe it will fundamentally change our lives as much as the internet and the industrial revolution.
Working on a business or new project can be lonely. Your family and friends support your efforts, but they don’t really understand the details of what you are trying to do or what exactly you are going through. Projects are difficult and the long hours can make you feel isolated, frustrated or overwhelmed - but don’t let that kill your morale or progress.
These are slides and code from a workshop in Portland on using random forests at scale with Python, Apache Spark and H20.