Advertisement
Canada Markets close in 5 hrs 43 mins
  • S&P/TSX

    22,077.50
    +179.52 (+0.82%)
     
  • S&P 500

    5,246.50
    -20.45 (-0.39%)
     
  • DOW

    38,034.17
    -407.37 (-1.06%)
     
  • CAD/USD

    0.7311
    +0.0019 (+0.2668%)
     
  • CRUDE OIL

    79.07
    -0.16 (-0.20%)
     
  • BTC-CAD

    93,736.81
    +901.12 (+0.97%)
     
  • CMC Crypto 200

    1,470.41
    +14.54 (+1.00%)
     
  • GOLD FUTURES

    2,370.50
    +6.40 (+0.27%)
     
  • RUSSELL 2000

    2,050.68
    +14.49 (+0.71%)
     
  • 10-Yr Bond

    4.5680
    -0.0560 (-1.21%)
     
  • NASDAQ

    16,845.23
    -75.35 (-0.45%)
     
  • VOLATILITY

    13.83
    -0.45 (-3.16%)
     
  • FTSE

    8,223.12
    +40.05 (+0.49%)
     
  • NIKKEI 225

    38,054.13
    -502.74 (-1.30%)
     
  • CAD/EUR

    0.6745
    -0.0001 (-0.01%)
     

Meta's alternative to ChatGPT leaked online, what we know so far

Facebook Chairman and CEO Mark Zuckerberg attends the annual Munich Security Conference in Germany, February 15, 2020. REUTERS/Andreas Gebert
Facebook and Meta chairman and CEO Mark Zuckerberg: Meta released LLaMa to a limited number of researchers in February but it was leaked. Photo: Andreas Gebert/Reuters (Andreas Gebert / reuters)

Meta's rival to ChatGPT has been leaked online and is now available for download. Yahoo Finance investigates how much it differs from Microsoft and Google's AI alternatives.

In February, Meta (META) unveiled LLaMA, their memorably-named Large Language Model competitor to Microsoft's ChatGPT and Google's Bard.

At the time Mark Zuckerberg's tech-firm stated that they would only give researchers access to the artificial intelligence application, so as to "maintain integrity and prevent misuse".

Meta's initial release statement said: "To maintain integrity and prevent misuse, we are releasing our model under a noncommercial license focused on research use cases, and access to the model will be granted on a case-by-case basis to academic researchers."

ADVERTISEMENT

Read more: How ChatGPT could lead to 'mass unemployment' – The Crypto Mile

However, someone has now leaked Facebook/Meta's Large Language Model alternative to the public via fringe online messaging platform 4Chan.

After the leak, developers have taken the new chat-bot app out for a test drive, so what do we know so far?

So far Meta's offering in the AI gold-rush to create the ultimate 'automated companion' for our journeys throughout cyber-space is similar in style to ChatGPT and Bard, with an ability to generate human-like responses to prompts and questions.

According to popular AI newsletter The Neuron, "early testers are saying that LLaMA isn't as good as the models powering ChatGPT.

"That's mostly because ChatGPT has anchored people to chat, and LLaMA wasn't taught how."

Read more: Microsoft’s Bing is the first threat to Google’s search dominance in decades

On machine learning platform Hugging Face, one developer inquired, "anyone know how to get going with LLaMA?"

Another platform released the code for LLaMA, on the popular website Github. Where developers had quickly used Meta's code to introduce 'ChatLLaMA', made by Nebuly, which was described as having a training process that was 15 times faster than ChatGPT.

After the leak was detected Meta sent a takedown notice to Hugging Face, citing legal issues.

Meta's Chief AI Scientist Yann LeCun has also claimed that although LLaMA is a smaller AI model that uses less parameters than rivals it will be just as powerful, if not more, than other leading language models.

Rivals ChatGPT and Google's Bard are trained following the scaling law, which attests that the more parameters the model has access to, the better the result.

However, a paper released last month by Meta artificial intelligence researchers claimed that "although it may be cheaper to train a large model to reach a certain level of performance, a smaller one trained longer will ultimately be cheaper at inference".

Read more: Microsoft's ChatGPT investment could create 'game-changer' AI search engine

Within artificial intelligence technology, 'inference' explains how an AI model can use the data that it has been trained on to understand new information that it has never been exposed to before. In application, this gives the AI the ability to make decisions and evolve its understanding when fed fresh data.

According to AI news outlet The De-Coder, Meta's alternative to ChatGPT "is more compact and faster and is even said to be on par with Google’s PaLM-E model".

Google's PaLM-E combines language, speech and robotics, but has not yet been released.

What are Large Language Models?

Large Language Models (LLMs) have captured mass media headlines across the globe since the release of OpenAI's ChatGPT in November, which has received investment from Microsoft (MSFT) to the tune of $10bn (£8.3bn).

ChatGPT now has over 100 million users, the fastest application to mass adoption on record, and Microsoft has stated that the new technology will operate as a user's 'co-pilot' as they navigate the web.

Read more: How to master using the new AI tool ChatGPT

Large Language Models allow 'artificial intelligence' to generate content on request via a trawl of large data sets, in a matter of seconds.

These generative-AI tools can produce haiku, jokes, corporate emails, business plans, fix computer code, write academic essays and even pieces of original fiction in the style of any writer you want, from Joyce and Dostoyevsky, to Shakespeare.

Microsoft and Google (GOOG) are competing to build the best models to boost their search engines.

Watch: How ChatGPT could lead to 'mass technological unemployment' - The Crypto Mile

Download the Yahoo Finance app, available for Apple and Android.