Grok-2 is a beast

Howdy, wizards.

I’ve only been on holiday for a week yet it’s like several months of tech development happened meanwhile.

Let’s dive right into it.

Dario’s Picks

The most important news stories in AI this week

‍

1. xAI launches Grok-2, an instant hit. Grok-2 is Elon Musk's latest AI model, and is now available in beta to X users. The new model surpasses GPT-4 and Claude 3.5 Sonnet on key benchmarks in the LMSYS chatbot arena, but is still behind GPT-4o and Gemini 1.5 Pro. In addition to the main model, xAI is also launching Grok-2 mini – a very capable sibling model in a smaller size. Grok-2 is vastly improved over the company's previous model which lagged far behind competitors.

The new model can also generate much better images than its predecessor and, judging by some of the stuff people are generating with it, apparently doesn't have a lot of filters.

‎ Why it matters‎ ‎ Grok-2 is up there with the current state-of-the-art models: GPT-4o, Claude 3.5, Gemini 1.5, Llama 3.1. This puts xAI in the pool of legitimate competitors in the AI race. At the same time, it keeps showing that new models aren't yet leaping above GPT-4o. Elon has a lot cooking, including a supercluster of GPUs for Grok-3, deep pockets and real-time data from his X platform – which makes the next Grok series much anticipated.

‍

2. Google demoed Gemini's new voice mode, Gemini Live. It's similar to ChatGPT's advanced voice mode, where it's close to real-time and you can talk to interrupt the model. You can also choose between 10 different voices, which sound very realistic. The feature is rolling out to Gemini Advanced subscribers this week, starting with the Gemini mobile app on Android, and coming to iOS over the next weeks.

New Extensions are also coming to Gemini, including Keep, Tasks, Utilities and Calendar which will give Gemini context about what's happening on your screen and make for a more integrated experience between Gemini and your other Google apps – similar to what Apple Intelligence is doing.

‎ Why it matters‎ ‎ The combination of Gemini's new voice mode and upcoming Extensions (which will also give Gemini context-awareness of what's happening on your screen) makes it a strong contender to both ChatGPT and Apple Intelligence. While Gemini Live is impressively fast and sounds very human-like, it's reported to be a bit too chatty and to hallucinate sometimes.

‍

3. Google hires key Character.ai staff. Google acquired key staff, including the founders, of Character.ai. In case you missed it, it's a platform for roleplaying with AI characters, either fictional or real; it's one of the most trending AI products among younger people (just their website has 200m+ visitors a month). C.ai's co-founder Noam Shazeer is actually an ex-Googler who was a co-author on the groundbreaking Transformers paper, so is returning to his roots. The deal lets Google use Character.ai's models (non-exclusively) and lets Character.ai use Google's models along their own LLM.

‍

‎ Why it matters‎ ‎ After two years of AI startups popping up left and right, an implosion seems to be happening where the "small" fish gets bought up by Big Tech. But not in the traditional way, as tech companies opt for quasi-acquistions and licensing deals to keep regulators at bay. First Microsoft with Inflection, then Amazon with Adept, and now Google with Character AI.

Ps if you're wondering what Character.ai is all about, I suggest checking out my categorised list of the top 1,000 characters.

‍

4. What people ask chatbots, based on data. Washington posts has dug into some 40,000 conversations from two GPT-based chatbots, which researchers say are representative for how the population use chatbots. Some highlighted popular use cases include creative writing and storytelling, dirty talk (there's censorship but people still try), homework help, advice and personal questions and of course, coding.

‍

‎Why it matters‎ ‎ Not any huge surprises from the results of this study, but what's more illuminating is looking at the specific chat examples given in the article. It's clear that chatbots are way more than just a productivity tool – they're often used for things like roleplay, emotional connection, and as a confidante and advisor for our personal problems.