Meta’s Open Source Llama 3 Is Already Nipping at OpenAI’s Heels

[AlexB23]

Hello folks. Open source technology is slowly catching up to ChatGPT. For myself, I run a free, open-weights AI model on my laptop, and have since March 22, 2024. AI can be used for good, or for bad, just as nuclear fission energy could be used to power a city with clean electricity, or sadly destroy a city with a nuke. Article link: Meta’s Open Source Llama 3 Is Already Nipping at OpenAI’s Heels


Short summary of the open-access article by WIRED (<20% in length, for fair usage reasons):

In an article published on April 25, 2024 in WIRED by Will Knight, Jerome Pesenti expresses his excitement over Meta's decision to release Llama 3, a powerful open source large language model that can be downloaded and used by anyone. As the former vice president of AI at Meta, Pesenti had advocated for the company to make its technology more accessible. Sizzle, his new startup, plans to evaluate Llama 3 as a potential replacement for OpenAI's GPT-4 model in many cases due to its power, affordability, and openness. The release of Llama 3 could shift the balance of power in AI by encouraging developers and entrepreneurs to use more affordable open source models instead of closed ones from companies like OpenAI and Google.

Meta made Llama 3 more powerful by feeding it larger amounts of higher-quality training data and new techniques to filter out redundant content. The cost of running Llama 3 on a cloud platform like Fireworks.ai is just a 20th of the cost of accessing GPT-4 through an API, making it an attractive option for companies like Sizzle. Other tech giants, such as Microsoft and Apple, have also released open source AI models recently. The coming months will reveal whether these open models can truly displace premium ones for some developers, and even more powerful open source AI is on the horizon. Meta's CEO Mark Zuckerberg believes that opening up its AI models will ultimately benefit the company by lowering costs and preventing competitors from dominating the field.

 

[Belk]

LLama is what we will be using to create a LLM for our solutions desk.

 

[durangodawood]

Why is it called Llama?

 

[Hans Blaster]

Why is it called Llama?

Insertion of vowels in to LLM (large language model, I think. I almost typed LLVM). They must think it was written in Hebrew or Sumerian.

 

[durangodawood]

Insertion of vowels in to LLM (large language model, I think. I almost typed LLVM). They must think it was written in Hebrew or Sumerian.

Ah.. I should have picked up on that.

Maybe Elon will make Llxmx.

 

[AlexB23]

Why is it called Llama?

LLaMA = Large Language Model Meta AI

 

[AlexB23]

Ah.. I should have picked up on that.

Maybe Elon will make Llxmx.

Elon made an abomination named Grok.

Elon Musk brings controversial AI chatbot Grok to more X users in bid to halt exodus

Musk is expanding access to Grok, possibly to revitalize Premium user base

www.techradar.com

 

[AlexB23]

LLama is what we will be using to create a LLM for our solutions desk.

What is a solutions desk? Is that the IT term for a help desk? It is cool that you guys will be using open-source AI. By the way, the article from my OP, I summarized (fair usage) with Mistral 7B running locally on my laptop.

 

[Belk]

What is a solutions desk? Is that the IT term for a help desk? It is cool that you guys will be using open-source AI. By the way, the article from my OP, I summarized (fair usage) with Mistral 7B running locally on my laptop.

Yes, it is a help desk.

 

[AlexB23]

Yes, it is a help desk.

Cool. So, what are your thoughts on running AI locally? I feel it is more private and secure that way, and would rather wait a minute for a locally run AI to respond compared to a server/cloud-based AI such as ChatGPT, unless I host my own server (have no idea how to do that, plus my laptop is too slow to be a server).

 

[Belk]

Cool. So, what are your thoughts on running AI locally? I feel it is more private and secure that way, and would rather wait a minute for a locally run AI to respond compared to a server/cloud-based AI such as ChatGPT, unless I host my own server (have no idea how to do that, plus my laptop is too slow to be a server).

Totally agree on the privacy and security. I work for a city government and we are being very careful with AI. The last thing we want to do is train a LLM and have some of the training data be something that contains citizen data. We are working with Microsoft to see what controls we can put in place before copilot affects our systems.

 

[Chesterton]

LLaMA = Large Language Model Meta AI

I hate to nitpick about the math, but LLaMA ≠ Large Language Model Meta AI, unless maybe you're dyslexic.

 

[AlexB23]

I hate to nitpick about the math, but LLaMA ≠ Large Language Model Meta AI, unless maybe you're dyslexic.

From Meta themselves: Introducing LLaMA: A foundational, 65-billion-parameter language model

 

[AlexB23]

Totally agree on the privacy and security. I work for a city government and we are being very careful with AI. The last thing we want to do is train a LLM and have some of the training data be something that contains citizen data. We are working with Microsoft to see what controls we can put in place before copilot affects our systems.

Agreed. Honestly, it is best to stay away from co-pilot, as it is not run locally on your laptop or data systems. If you have to build your own server using NVIDIA GPUs, then build your own server.

 

[Belk]

Agreed. Honestly, it is best to stay away from co-pilot, as it is not run locally on your laptop or data systems. If you have to build your own server using NVIDIA GPUs, then build your own server.

There is no real way to stay away from co-pilot. I HAVE to work in the Azure cloud and there is no way to completely block co-pilot. Best we can hope for is to limit it and have controls.

 

[Chesterton]

From Meta themselves: Introducing LLaMA: A foundational, 65-billion-parameter language model

Nope. Not there.

Forgive me, I'm old. I remember a simpler time when things made sense. AT&T stood for American Telephone & Telegraph, and IBM stood for International Business Machines. See, the equal sign belongs between those.

I saw an ad for a business. It's called Credit Human. Credit Human. I'm guessing it's some kind of church?

 

[AlexB23]

Nope. Not there.

Forgive me, I'm old. I remember a simpler time when things made sense. AT&T stood for American Telephone & Telegraph, and IBM stood for International Business Machines. See, the equal sign belongs between those.

I saw an ad for a business. It's called Credit Human. Credit Human. I'm guessing it's some kind of church?

Yeah. No idea what Credit Human is. Sounds humanist to me. Large LAnguage model Meta AI. Yeah, a terrible acronym. Companies are coming up with vague name for things nowadays it seems. Oh wait, Credit Human is a credit union. Looked it up for you: Checking, Savings and Loans | Credit Human

 

[AlexB23]

There is no real way to stay away from co-pilot. I HAVE to work in the Azure cloud and there is no way to completely block co-pilot. Best we can hope for is to limit it and have controls.

Agreed. I wish you luck with this.

 

[Belk]

Agreed. I wish you luck with this.

Thank you. It is certainly interesting times.

 

[AlexB23]

Thank you. It is certainly interesting times.

Yep. AI's cool, except for an AI priest, though some of my Bible analysis are partially done with an open source AI named Mistral 7B on my laptop. Have you heard of the defrocked AI priest?

This is Fr. Justin, before being digitally defrocked. He was run on a server, and was not open source.

Catholic Group Defrocks AI Priest After It Gave Strange Answers

A Catholic group has defrocked an AI-generated priest chatbot after it stirred controversy among the faithful.

futurism.com