- Bot Eat Brain
- Posts
- xAI released Grok-1.5 with upgraded math, coding, and reasoning abilities.
xAI released Grok-1.5 with upgraded math, coding, and reasoning abilities.
PLUS: Clone your voice
Michael Parrish
April 02, 2024
Sponsored by
TOGETHER WITH
Good morning, human brains, and welcome back to your daily munch of AI news.
Here’s what’s on the menu today:
Grok-a-doodle-doo 🤖 🤘
xAI released Grok-1.5 with upgraded math, coding, and reasoning abilities.
Don’t trust humans. Blindly follow AI 🧑🔧 📟
Google DeepMind released a fact-checking method for LLMs.
OpenAI’s voice cloner 🗣️ 🫢
It unveiled the speech generator used in Read Aloud, ChatGPT, and more.
MAIN COURSE
Let your Grok flag fly 🤖 🤘
On Thursday, Elon Musk’s xAI announced Grok-1.5. It features improved reasoning capabilities, an extended context length, and more.
Is it… Grokked and loaded? 🤭
Nice. Grok-1.5 builds on its predecessor with enhancements in coding, math-related tasks, and more. It’s capable of processing up to 128K tokens.
What’s in Grok’s guts?
Grok 1-5’s custom framework utilizes JAX, Rust, and Kubernetes for efficient prototyping and training of new architectures.
So, it’s a Grok star?
The model scored 50.6% on MATH, 90% on GSM8K, and 74.1% on HumanEval benchmarks.
I want to Grok the Casbah.
xAI claims Grok-1.5 will soon be available to early testers and will introduce new features and roll it out to a broader audience “over the coming days.”
Who do I have to Grok for a job?
Want a career at xAI? Click here to apply.
Grok me Amadeus.
Yes, Grokfather. 🫡 In November, we covered xAI’s coding tool, PromptIDE. It allows you to create and test prompts while using Grok-1.
The same month, we reported on xAI’s AI model, Grok. It uses real-time world knowledge to wittily answer questions, suggest questions to ask, and more.
Two weeks ago, we covered xAI’s open-source release of Grok-1. You can download it now; it’s about 300 GB and contains 773 files.
SPONSORED BY AE STUDIO
AI brews beer and your big ideas
What’s your biggest business challenge? Don’t worry about wording it perfectly or describing it just right. Brain dump your description into AE Studio’s new tool and AI will help you solve that work puzzle.
Describe your challenge in three quick questions. Then AI churns out solutions customized to you.
AE Studio exists to solve business problems. They build great products and create custom software, AI and BCI solutions. And they once brewed beer by training AI to instruct a brewmeister and then to market the result. The beer sold out – true story.
Beyond beer, AE Studio’s data scientists, designers and developers have done even more impressive things working 1:1 with founders and executives. They’re a great match for leaders wanting to incorporate AI and just generally deliver outstanding products built with the latest tools and tech.
If you’re done guessing how to solve work problems or have a crazy idea in your back pocket to test out, ask AI Ideas by AE Studio for free solutions, right now.
SIDE SALAD
AI tells AI its wrong 🧑🔧 📟
On Wednesday, Google DeepMind unveiled LongFact and SAFE. It’s a benchmark dataset and a long-form content fact-checking method that surpasses human abilities.
What’s LongFact?
It’s a benchmark dataset that uses GPT-4 to generate thousands of questions across 38 topics.
What’s SAFE?
SAFE is an LLM evaluation method. It rates LLM’s factuality by breaking down responses into facts and verifying them via Google Search.
Why do I care?
SAFE shows that LLMs are over 20 times more cost-effective than human annotators. In 100 disagreements between SAFE and humans, SAFE was correct 76% of the time.
So it’s better than humans?
It depends. SAFE's effectiveness depends on the LLM's capabilities and the comprehensiveness of Google Search results.
What else has Google DeepMind been up to?
In February, we reported on Google DeepMind’s Genie. It creates playable 2D platform games from text prompts, images, videos, and more.
In March, we covered Google DeepMind’s SIMA. It’s a versatile AI agent trained on nine complex video games.
A week later, we reported on Google DeepMind and Liverpool FC’s TacticAI. It’s an AI system that analyzes past plays and gives soccer strategy insights.
RECOMMENDED READING
Join over 110,000 people learning actionable tips and growing personally and professionally with Amanda.
Each week, subscribers receive a deep dive that covers topics ranging from personal growth, productivity, career growth, mental health, money, and relationships.
A LITTLE SOMETHING EXTRA
Clone your voice 🗣️ 🫢
On Friday, OpenAI introduced Voice Engine. It generates natural-sounding speech from 15 seconds of audio and text input.
What is it?
Voice Engine has been in development since late 2022. It fuels preset voices in OpenAI's text-to-speech API, ChatGPT Voice, Read Aloud, and more.
What’s the point?
OpenAI claims it collaborates with private partners to test Voice Engine with reading assistance, global content translation, support for speech impairments, and more.
Let me guess… I can’t use it?
Correct. OpenAI says the potential for misuse is too great, especially in an election year.
But, I need voice-cloning tools.
We got you. In January, we reported on Dubbing Studio from ElevenLabs. It’s an AI tool for video translation in 29 languages, transcript editing, and more.
In March, we covered Microsoft’s NaturalSpeech 3. It’s a text-to-speech model that can clone voices, emulate realistic emotions, and more.
On Friday, we reported on Rabbit and ElevenLabs’ partnership. Rabbit integrated ElevenLabs voice tech into its upcoming r1 device.
YOUR DAILY MUNCH
Alice
Instantly boost productivity using Alice's AI suite. Streamline tasks with shortcuts and automation in a single app. Craft social media posts, manage expenses, and automate your schedule. Transform your productivity now.
Think Pieces
Why was Google DeepMind’s CEO knighted? Demis Hassabis was awarded a knighthood for “services to artificial intelligence.”
Beginner’s guide to Grok. What it is, how to use it, and more.
Salt-sized brain sensors? Brown University researchers developed brain-machine interface sensors that transfer data similarly to neurons.
Startup News
OpenAI and Microsoft are reportedly developing Stargate. It’s a $100 billion AI datacenter for a supercomputer with millions of GPUs.
Google announced Google.org Accelerator: Generative AI. It’s a $20 million AI accelerator program.
Midjourney announced new Personalized Models. They boast better image and video generations and faster prompt understanding.
Research
ReALM — Apple’s LLM that achieves near GPT-4 performance.
UPD — Unsolvable Problem Detection is a challenge to evaluate the trustworthiness of Vision-Language Models (VLMs).
Gecko — a versatile, small text embedding model that competes with much larger models.
MEMES FOR DESSERT
TWEET OF THE DAY
AI agents can create entire websites from the ground up.
Tag us on Twitter @BotEatBrain for a chance to be featured here tomorrow.
AI ART-SHOW
Until next time 🤖😋🧠