Meta heats up the LLM wars

And OpenAI ponders about AGI and Beyond

Author

The Alignment
February 26, 2023

Welcome to The Alignment, where we help you understand what’s happening in the world of AI without any of the jargon or bs.

AI never rests, so there is loads to catch up on from the weekend !

  1. Meta's new model outperforms GPT 3 ?

  2. AGI and beyond at Open AI.

  3. Steve Jobs always knew.

New Meta model claims it outperforms OpenAI's GPT-3

Mark Zuckerberg, CEO of Meta making the official announcement

Meta has released their GPT-3 competitor, LLaMA which it claims outperforms GPT-3 on most benchmarks. A few things standout about this model -

  1. It is about 12% the size of PaLM, Google’s state of the art model, but uses a higher volume of tokens.

  2. It is only trained on publicly available data.

  3. The weights of the model are completely open sourced here.

  4. However, the model is released under a non-commercial agreement and access to the model will only be granted on a case by case basis for research and development purposes.

Another horse has joined the LLM race, and of course it’s a big tech company. All this potentially hints toward the commoditisation of LLMs thus creating interesting market dynamics on where money is to be made as models are commercialised and who makes the money (big tech vs emerging startups).

AGI and beyond at Open AI

OpenAI has shared how they are planning for AGI and beyond. Sam Altman shared both the short and long term vision for what they have in mind.

Short Term -

As more people come in contact with OpenAI’s models (directly or indirectly) OpenAI will become increasingly cautious with the deployment of their models. This could mean that users as well as use cases may be more closely monitored and restrained.

OpenAI is working towards more alignment in their models. They also want more controllability and there will be more options for customization of models for users.

Sam also very clearly reiterated that OpenAI’s non-profit umbrella, cap on profits to shareholders and the ability to override equity obligations all align with their mission to ensure that AGI is used for the greater good.

Long Term -

Sam candidly predicted that “The first AGI will be just a point along the continuum of intelligence.”

He also specifically pointed out that AGI that can accelerate science will be an important use case that OpenAI will be focusing on. Mentioning that :

“It’s possible that AGI capable enough to accelerate its own progress could cause major changes to happen surprisingly quickly (and even if the transition starts slowly, we expect it to happen pretty quickly in the final stages). We think a slower takeoff is easier to make safe, and coordination among AGI efforts to slow down at critical junctures will likely be important”

Product Watch Grain AI -

Focus on talking to your customers, Grain will take care of everything else.

Around the Industry -

Steve knew it from the very beginning  

Humans are going to make some very powerful domain specific tools with AI.