How To Use Metas New LLaMa 2 Generative A.I. Tool As A Chatbot, And How It Compares To ChatGPT And Google Bard

As the name suggests, this is the second version of the Meta Tool (LLaMA stands for Large Language Model Meta AI). According to Meta, the new LlaMa is trained on 40% more data than its predecessor and doubles the context length.

But how does it compare to some other text generators like ChatGPT, Bing Chat or Google Bard?

I've been playing around with LLaMa 2 to see how it handles some common tasks that generative AI tools come in handy for. What I've found is a powerful, open-source template with a lot of potential for customization and customization for different experiences. But as a consumer-grade AI assistant for functions like typing or researching, LLaMA 2 is a serviceable tool, but no better than some bots out there.

A different kind of AI bot

One thing to understand about LLaMa 2 is that its main purpose is not to be a chatbot. LLaMa 2 is a public LLM available for developers to download and customize as part of Meta CEO Mark Zuckerberg's model refinement and development program.

That said, if you want to use LLaMa 2 as a chatbot, you'll need to take advantage of the private betas available on platforms like Hug Face. The version we are using, HuggingChat, was created by the developer community by installing LLaMa 2 on Hugging Face. There are other places where you can try different chatbots based on LLaMa 2, but HuggingChat is a specialized chatbot developed as an open-source alternative to ChatGPT.

Click here to try HuggingChat.

Philip Schmid, Hugging Face's chief technology officer, told Fortune that while the chatbot is comparable to other AI bots, it's not a perfect comparison. A unique feature of LLaMa 2 is that it can be cost-effectively tailored to specific needs. The template is not designed for a specific purpose like products like Bing Chat.

LLaMA 2 is also offline. This means there is a “knowledge threshold” of December 2022. This is newer than ChatGPT's September 2021 threshold. The creators of the HuggingChat chatbot added a web search option, but this is still in its infancy and doesn't offer LLaMa 2 the same features as other web search chatbots. If you need up-to-date information on the Internet, it is best to use a tool such as Bing Chat or Google Bird.

In a post announcing the release of LLaMa 2, Facebook researchers wrote that LLaMa 2 forms generally perform better than existing open-source models and are closer to closed-source models like ChatGPT, like based on human reviews in the article. The paper acknowledges that it is not fully comparable to GPT4, OpenAI's most advanced LLM.

Try LLaMa 2!

I asked the bot to email my co-workers telling them I was leaving town. Make a concise note that fits the fragile formality of the business area.

He can write letters, but can he handle difficult things like turning down a job offer? I asked the bot to write an email response saying I couldn't accept the job offer. He wrote three short, impersonal paragraphs that could pass as a person, but as an email. Email certainly won't take away the frustration of a rejection.

So I asked again, more specifically, more personally, and more apologetically. He responded with a play on words, perhaps too formal, but email was an option this time. It looks like the LLaMa 2 demo will evoke false regrets when asked.

LLaMa can handle such tasks, especially if you want her to know the details. He can write decent summaries and can easily jot down a memo if someone needs help. There is an advantage to politely declining meetings: just give them specific names, times, and reasons, or write specific, formal letters.

Compared to ChatGPT, I found LLaMa 2's font to be decent, but too formal. I would use chatgpt because they often have a greater tendency to keep people in their language for degrees. LLaMa was a little rough and generic for the job.

For more creative or "literary" writing tasks, there was also LLaMa 2 There are problems following the mixed word count instructions. If I want a 150-word story, I'll give it 190. He can write a 16-line haiku or poem about anything, but it's hard to tell if he's any good. Do you think "Images come alive, sorcerers enter digital conflict, mutual symphony" is a powerful haiku?

I asked him to write about “The Destiny of Journalism in 2020” and he wrote a horrible 16 line poem. While chatbots aren't known for their literary finesse (and I'm probably not qualified to judge a poem), this one seemed half-baked. It didn't rhyme, and while it sprang up funny lines like "Tattooed guy, once the pride of fourth grade" and had a coherent theme, I wouldn't say it was well written.

When it comes to research, LLaMA 2 falls short

I asked the bot for some hard facts and told myself about the real estate crisis in China. In order to summarize the Chinese market, social issues and infrastructure, a few points are compiled. If you click on it for more information, you can explain property prices and the impact of the COVID-19 pandemic .

Then I asked him to write a 50-word essay with quotation marks. This resulted in 71 words with the publication name in brackets at the end. I turned on the Web Search feature, which lets you search the web, and asked again: He gave me 50 words this time, but each link led to a page that didn't exist.

When I was asked what is happening with the July 2023 crisis, I was again fed misleading excuses for misinformation and other broken links.

December 2022 Between knowledge and troubleshooting features, it's probably best not to use it for critical research. It still adorns, but it needs polishing. The same rules apply to all generative AI tools; Always research what you create. However, with this tool, it is particularly important to do so. He has hallucinatory quotes and his knowledge is borderline. If you ask him to summarize and summarize the information or change the text, his response is rather shockingly confusing.

Should you swim at LLaMA?

LLaMA 2 offers hugging face unlike other chat programs like ChatGPT, Google Bard and Bing Chat. It's promising for an early version of a chatbot, but it's still pretty raw. Not ideal for exploring and had some "confusing" moments (sorry for the anthropomorphism).

If I want to use the demo for anything other than testing and note-taking, I have some long, sometimes unfinished work to do.

However, there are countless reasons to use an AI chatbot, and tools like LLama 2-based HuggingChat are constantly being changed and updated. Therefore, I recommend you try this bot for yourself and see if it suits your needs better. Just be aware of its limitations.

You can try HuggingChat here.

Here is Fortune's 3-step guide to using chat AI tools

LocalGPT and Lama-2. Added chat history and custom prompt templates