The Five Most Key Takeaways from This Blog
- Meta recently released Llama 3.1, an open-source A.I. large language model (LLM). An LLM is a generative A.I. model that has a huge amount of data in its training-data set. So, the “large” part refers to the data set, in addition to the potentially large amount of responses that it can create as a result of that wide-ranging training.
- “Open source” means it is free to use and, importantly, customize, by anyone. Well, anyone with a network connection at least.
- Meta itself made the claims that this LLM is more capable than the offerings from OpenAI, the creator of ChatGPT, and Google, which offers Gemini (née Bard).
- Part of the motivation of the open-source approach is that it could encourage start-ups and other companies to both develop their own models that they would otherwise pay OpenAI or Google to do. Heck, some of those companies may use Llama 3.1 to create A.I. that rivals OpenAI and Google. In other words, it can draw users away from the competition.
- Meta does not have a large cloud-computing service as its rivals Microsoft and Google do (such services allow those companies to sell its A.I. services), so Meta’s gambit with open source allows it to find its way into influencing A.I. in business.
So, Is It Really Better Than Those Models?
Probably not, if media reports are to be trusted.
As a contributor to ZDNET reports, he broke this very model with a question that also threw Gemini for a loop: “What is the conjugation of the Georgian verb ყოფნა?”
So, as it turns out, OpenAI’s ChatGPT-4o was able to give a correct answer to this question. That answer was in the form of a verb chart (think back to those box charts in your text books for foreign-language classes). He was even able to get a verb chart for future-tense forms of the Georgian word for “to be” when he asked a follow-up requesting those forms.
And what was Llama 3.1 able to spit out for that reporter? Perhaps a fancier-looking verb chart with even more flourishy customer-service speak (e.g., something like a “Let me know what else I can help you with!”) surrounding the verb-forms chart?
Well, as that reporter reports, Llama 3.1 was unfortunately only able to offer back “gibberish” as an answer. That is not to say that the whole answer resembled some failed modernist literary experiment but rather offered back gibberish Georgian words in different verb tenses, which was not even in the first answer. Instead, Llama 3.1 acquiesced to the demand only after some prodding.
The first answer from Llama 3.1 to the ZDNET experiment has a tail-between-legs quality to it, in that it recommends that the writer check out some Georgian textbooks (good luck finding that in your local public library branch, unless you live in Georgia (and this writer is not referring to the U.S.A. state of Georgia, either)) or another “trustworthy source”. What is even more disappointing, the chatbot did not even offer such sources, instead basically just telling that ZDNET contributor to do his own research.
If It Is Not Better Than ChatGPT, Then Why Use It?
Fair question, but business owners can actually get something out of using Llama 3.1, assuming your business does not need to produce content in the Georgian language. (Do note that the company may be working to improve its model’s “multilinguality” or, depending on when you are reading this, may have already done just that.)
The major advantage that the open-source model offers is something that the Key Takeaways section opening this blog mentions: users (which can include a business’s tech-savvy employees) can customize these models to specific purposes.
This means that you can fine-tune this model to generate answers that are specific to your company’s needs. For instance, if you need to generate internal reports based on data that belongs to your company, then you can use that data to train the model on, which in turn can mean your company has a custom-fit LLM at its disposal.
However, do be wary about what data you give this model, as sensitive data should still be closely guarded. If you were to share such data, there is no guarantee that the LLM may not somehow reveal it in conversation. Or, the data could become more readily accessible to cybercriminals.
Plus, even just using the Meta AI platform, a small window pops up saying your conversations may be used by the company. So, be careful about what you tell it.
Overall, Llama 3.1 is a serviceable free chatbot that business owners with the proper resources could fine-tune to specific, beneficial ends.
Recent Comments