Gen AI
Generative AI for your business
Generative AI is poised to rapidly change how businesses operate, offering enhanced efficiency and smarter decision-making.
The pace and complexity of developments can be bewildering. Reap the benefits of AI, without the learning curve, with JetRock as your AI concierge.
Below is JetRock's view on the current state of the art and practical applications of this rapidly evolving technology.
Reach out to start a conversation.
The basics: Your one minute AI cheat sheet
Sound smart when AI comes up in conversation
LLMs
Large Language Models - neural networks that power the new wave of generative AI models.
Tokens
A token represents part of a word, usually about 5 letters. LLMs process text one token at a time, rather than one letter or one word at a time.
Context Length
Context length refers to how many tokens an LLM can keep in working memory at once. Context lengths have exploded from ~8K to over 1 million tokens in the last year.
LLM Rankings
LMSys' Chatbot Arena Leaderboard is the go to place to see which models are currently considered to be the best:
Frontier
Frontier models refer to AIs that are at the outer limits of capablity. Since there are many axes to compare models on, there are several on the frontier rather than one dominant model.
Multimodal
Multimodal models can consume and produce more than just text, including images, video, and audio. The frontier models are all multimodal now.
Hallucinations
LLMs tend to make up answers rather than say they don't know. This is less of a problem with the frontier models, but it requires a change in mindset from traditional software.
Thought Leaders
If you want a single source of solid information, follow Ethan Mollick on LinkedIn and read his (short) book: Co-Intelligence
Leading AI models
OpenAI GPT-4
OpenAI's GPT-4o, which powers ChatGPT, is the gold-standard all others are compared to. It's the only one with a built-in code interpreter, which is key for data-focused applications.
Anthropic Claude
Anthropic's Claude 3.5 Sonnet model is another highly capable generative AI model that has potentially eclipsed the older GPT-4 in general intelligence. It appears to be on par with GPT-4o, including the ability to run code.
Google Gemini
Google's Gemini 1.5 Pro and 1.0 Ultra models lead on context length and video processing, but lag in general intelligence.
Open Source Models
There are several like Meta's Llama 3 and Mistral's Large model. These models lag modestly in capability, so JetRock doesn't focus on them, but they have a vocal community of supporters.
Current AI use cases
1. Interactive AI applications
When the end user needs to interact with the results iteratively, an AI chatbot can be employed. Example use cases include data acquisition, generating ad hoc charts and reports, and creating GIS maps.
2. Non-interactive AI-enabled solutions
AI can also provide new capabilities to traditional reporting deliverables, such as data entry from non-standardized documents, analyzing images and video, and generating concise summaries from large bodies of text.
3. Automating routine processes
AI can be utilized to speed the development of python code to automate routine processes that require the same output to be refreshed repeatedly. JetRock uses this capability extensively.
1. Interactive AI applications

1

Data acquisition
AI chatbots can replace SQL queries, allowing for more intuitive and flexible data acquisition for everyone at the organization, not just SQL gurus.

2

Ad hoc charts and reports
Generative AI can be used to create custom charts and reports on the fly, tailored to the user's needs, using plain English, using example documents to apply the users intended tone and style.

3

GIS mapping
AI-powered tools can generate interactive GIS maps, opening easy spatial analysis to non-GIS professionals.
2. Non-interactive AI-enabled solutions
Data entry from non-standardized documents
AI can extract and process data from a variety of non-standardized sources, such as PDF, Excel, and handwritten documents, streamlining data entry tasks.
Concise summaries from large bodies of text
Generative AI models excel at generating concise summaries and answers from large bodies of text, providing valuable insights and reducing the time required to review extensive documentation and reporting.
3. Automating routine processes

1

GitHub Copilot
Copilot, an AI-powered coding assistant, can accelerate the development of Python scripts to automate routine processes. JetRock currently uses this extensively.

2

OpenAI ChatGPT
ChatGPT, a large language model, can also be leveraged to generate and refine Python code for automation tasks.
Current state of the art: OpenAI's GPTs

1

Versatility
OpenAI Plus allows subscribers to build their own GPTs: customized versions of GPT-4 with custom instructions and pre-loaded documents.

2

Limitations
While powerful, the consumer version of ChatGPT has limitations, such as a low (1-2 GB) memory capacity and the need to manually load data, which can hinder its use on large datasets. It is expected this constraints will be lifted in the near future.

3

Current State
GPTs capabilities are impressive, but generally fall short on being able to completely replicate multi-step workflows, even when used by an experienced prompter.

4

What's Next
GPT-4o is being rolled out and it's improved voice real-time voice and vision will open additional use cases.
GPT-4.5/5 is expected in late 2024 and is widely expected to be another significant step up in capability. Google and Anthropic are also nipping at OpenAI's heels.

5

Take Off
Around 2027-2028, there's a significant chance we reach AGI (Artificial General Intelligence). At this point, things get very hard to predict. Read this essay from a former OpenAI researcher to understand this scenario:
What's the latest?
For the latest AI updates, go to our LinkedIn group page:
Click here
The AI business starter-kit
1. A robust data foundation
  • Clean, organized datasets are crucial for running reliable AI models.
  • Combining the best features of data warehouses and lakes, platforms like Databricks and Azure Fabric offer scalable, flexible solutions for handling diverse data types.
2. Precision-crafted AI models
  • Tailor the base AI models to your unique business needs, leveraging your data.
  • An easy-to-use user interface is essential for adoption.
JetRock expertly handles both of these AI essentials for you.
Why Partner with JetRock
Your AI concierge, at your service
For many companies we talk to, their data system consists of emailing excel files back and forth to each other. That's slow and chaotic. They aren't using modern analytics tools and they are in no position to take advantage of the new AI-based tools that are rolling out.
That's where we come in. We understand both how to build effective data systems and integrate them with AI.
We're not pushing a mass-market tool. We are your data and analytics team.
We work with you to understand your specific goals and workflows across your whole organization: business development, accounting, operations, engineering, regulatory, etc.
Then we deploy our Azure based data integration system to pull all of your data into a common data warehouse. We pair that with interactive dashboards, automated email reporting, and custom data analysis, all tailored to your unique approach.
Bottom line, your team gets the relevant information you need to make better informed decisions faster because we've put that information right in front of you in the format that you want, with no legwork required. It accelerates the whole company.
- Kevin Thuot
Managing Partner
JetRock Analytics

Let's Talk
JetRock is committed to driving tangible results for business through generative AI.
Don't get left behind.
Reach out today to start the conversation.