What’s going on here?

Elon Musk’s xAI announced their first large language model Grok.

‍

What does that mean?

xAI trained Grok-0 (33B) and Grok-1 models in these 4 months. The Grok-1 model will power the consumer-facing product “Grok”. It’s essentially a GPT-3.5 competitor beating GPT-3.5, Inflection-1, and LLaMA 2 70B on different benchmarks. Grok-1 is still behind PaLM 2, Claude 2 and way behind GPT-4.

But, Grok has Elon as its hype man and he is on a streak to promote Grok on a different set of benchmarks:

• It has real-time access to information via X (Twitter).
• It’ll respond with sarcasm and humour (and sometimes, swear words).
• It won’t dump you with a letter signed by “I’m a large language model…”.

Grok, the name is from a sci-fi book Stranger in a Strange Land and AI is inspired by Hitchhiker’s Guide to the Galaxy. Don’t know what means, but just letting you know.🤷

‍

‍

‍

‍

Why should I care?

Okay, jokes aside. It’s the first product from the team in just 4 months. There’ll definitely be room for improvement and I don’t think the team will have to worry about resources. (Elon is still the richest man on Earth). The comparison tweets are not worth looking at right now. They are either fans cherry-picking examples or angry people complaining about it.

The most impressive part of this launch is integration with X. Real-time information from not just the news websites but X user posts could end up being a key detail. And then the availability for X subscribers will take it to non-professional consumers (similar to Meta adding personas in IG).

Though you might not care, I like the naming and science fiction inspirations. (That too in a positive light for once, after all the doomer debates.)