- Tech for Finance
- ChatGPT-4 vs Llama 2 head-to-head, Max your tech 'time to value', AI powered debt collection
ChatGPT-4 vs Llama 2 head-to-head, Max your tech 'time to value', AI powered debt collection
1-2-3 Newsletter - Read Time - 4 Mins
Hello, and welcome to a another edition of the Tech for Finance 1-2-3 Newsletter where we have:
1. Tutorial - GPT-4 (Paid) vs Llama2 (Free)
2. Expert resources - How to max time to value with tech + data warehouses vs data lakehouses
3. Tech tools - AI debt collection, speech to text and AI team assistants
1/ GPT-4 vs Llama2
OpenAI aren't the only company to have developed a large language model. Meta have their own Llama model, and it's just had a facelift.
It's trained on 70 billion parameters (more info here) and is now a true contender in the generative AI space.
It's still opensource, so developers can download the model.
Unfortunately, Meta don't have a chat interface like with ChatGPT.
But, the good news is you can still chat to it via HuggingChat (provided by HuggingFace).
1. Go to this link.
2. Make sure your selected model is Llama 2
To put Llama 2 to the test, I pitted it against ChatGPT-4 using some of my most complex prompts.
1/ Scenario Planning
Note - I've not included the full prompt in the image below as it's too long, but you can find it in the link to chat further down the page.
1/ Chat GPT-4 (Paid) link to chat here.
2/ Llama 2 (Free) link to chat here.
The ChatGPT response has always been good with longer prompts. The result in this situation was pretty cut throat with succinct bullet points in response to my queries.
I was pleasantly surprised by the Llama response, as it seemed to add a little more meat to the bones. The formatting wasn't quite as clear as ChatGPT, and I had to prompt it to continue finishing it's response when it ran out of steam. But happy to do that considering it's free.
2/ Spreadsheet Merge Python Code
Now to test how the tools compare when generating code.
1/ Chat GPT-4 link to chat here.
2/ Llama 2 link to chat here.
Both gave the same library recommendations. But each gave different types of code.
The ChatGPT code is more elegant, and also gave instructions to ensure the script is located in the same file folder as the Excel files.
The Llama 2 code wasn't as elegant, and the code it generated would require you to specify what columns from each spreadsheet you wanted to merge.
From the limited time I've been using Llama 2 I'd say it's a fabulous free alternative to ChatGPT-4.
It's not quite as good for the more complex stuff compared to GPT-4, but using HuggingChat you can connect it to the internet which is another real plus.
ChatGPT-4 is still the only option if you want to make use of plugins and the code interpreter. But if you don't need all of that Llama 2 is potentially the next best thing.
2/ Expert Resources
1. This week we released my podcast with Dr Liran Edelist (Ph.D) strategic leader on all things business growth and EPM, and established author.
In our discussion we cover:
- KPIs and TTV (Time to Value)
- EPM and CPM Approaches
- Enterprise Performance Management and Customer Success
- Tech Transformation and Business Challenges
- Using Technology to Increase Productivity
- Analyzing Cost Structures for Software Projects
- AI's Future Impact on Financial Relationships
And much much more.
Check out the full podcast "How to maximise tech adoption using TTV (Time to Value)" on audio here. And on video here.
2. My friend Simon Devine released an article recently on the difference between a Data Warehouse and a Data Lakehouse.
You can find the full article here.
P.S - He's also working on an FP&A tool using PowerBI as well. You can find the recent webinar here.
3/ Tech Tools
1. Respaid - Collect small debts using AI + Respect - learn more here.
2. Cockatoo - Speech to text transcription - learn more here.
3. Sembly - AI team assistant - Learn more here
IMPORTANT - Always read the data policy for any tools you try. If you’re ever unsure. Make sure your data, and any customer or supplier data is anonymised.
NOTE - I am not affiliated with any of the tools in this e-mail.
So that’s it for this week.
Until Next Time
👨💻Connect with me on LinkedIn
🎧Subscribe to the Podcast
📺Watch on YouTube