The $225M Chatbot

PLUS: Google's AI leak

Good morning human brains, welcome back to your daily munch of AI news.

Here’s what’s on the menu today:

  • Google’s AI leak 📑 

    Google’s internal AI docs just got leaked, and the statements aren’t flattering.

  • InflectionAI releases Pi ➗ 

    Ex-Deepmind employees raised $225M to build a ChatGPT competitor. The reviews just came in.

  • Slack becomes an AI org 🤖 

    Slack’s slapping on an AI cover and working to integrate new tools across the board.

APPETIZER

Google’s Leaked AI Docs 😵 

Last night, an anonymous researcher at Google leaked its internal AI statements, and it’s not the prettiest look. Here are the biggest finds:

1/ Some Google officials think it has no AI Moat, and neither does OpenAI.

2/ Google researchers believe that open-source AI will outcompete both it and most other AI players, and are advocating for open models themselves.

“While our models still hold a slight edge in terms of quality, the gap is closing astonishingly quickly. Open-source models are faster, more customizable, and pound-for-pound more capable. They are doing things with $100 and 13B params that we struggle with at $10M and 540B.”

The Google Docs

3/ Researchers think Large models aren’t more capable in the long run if they can iterate faster on small models.

4/ Google actively bars many sites from showing up in search (at least as of a few years ago). Some of these are allegedly unjustified.

5/ Google talks about algorithmic unfairness many times internally, which likely means it does emphasize making its AI tools less biased.

This comes right after Google’s Q1 earnings call and Google employees complaining about the CEO’s $226M salary.

What this means: some Google AI researchers aren’t confident in the company’s approach to LLMs, but that’s about all. Most of the 900 files leaked are screenshots of emails sent between Google employees with no context. That said, the narrative painted has interesting insights and may be worth a read.

BUZZWORD OF THE DAY

Data Science

studying data to find meaningful insights for businesses. This combines math, statistics, AI, and computer engineering to analyze big amounts of information.

FROM OUR PARTNERS

Build & Deploy AI Models Quickly

Cerebrium makes deploying ML Models seamless. Deploy & Fine-Tune your model without worrying about infrastructure.

1/ Serverless GPU Deployment - <1s cold start

2/ All Major ML Frameworks Supported (Pytorch, Onnx, XGBoost, Etc).

3/ Don’t have your own model? Access 15+ pre-build models.

4/ Fine-tune FlantT5, GPT-Neo, Stable Diffusion, and more.

MAIN COURSE

The $225M Chatbot 🦣 

Last year, ex-Deepmind members raised $225M from investors like Ried Hoffman and founded InflectionAI to work on a GPT competitor. This week, they unveiled the Pi chatbot. Here’s what’s new:

1/ Pi converses colloquially, getting to know you over time and remembering your past conversations with it.

2/ It’s built on one of Inflection's in-house LLMs, so this isn’t just another GPT wrapper.

3/ It’s intended to be a personal AI that’ll help you organize your schedules, prep for meetings, and learn new skills.

4/ It’ll detect when you appear to be growing frustrated with it and tweak its tone of responses.

5/ It’s about to have access to real-time online content and use your calendar, email, and other docs on your PC to better manage your time.

Inflection's ultimate goal is to create a JARVIS-esque chief of staff that’s more hands-on than GPT and can talk with other AIs to optimize your work.

The reviews: Pi’s release generated a bunch of buzz online, but the consensus after people played around with it is that Inflection just spent $225M on a slower, less useful ChatGPT. Testers at Forbes say it’s relentlessly bland.

Take: its impressive that Inflection built a GPT competitor this quick with a fraction of their budget. With its focus on internal ops, it’s possible Pi’s next versions will resonate with pros - we aren’t writing Pi off just yet.

Curious? Play with Pi here.

MEMES FOR DESSERT

YOUR DAILY MUNCH

Think Pieces

Four investors explain why AI ethics can’t be an afterthought

How Does AI Writing detection work? Robot v Reality.

Startup News

AI-powered supply chain startup Pando lands $30M investment.

Slack updates aim to put AI at the center of the user experience.

Research

FactGen: faithful text generation using factuality-aware pre-training.

”Coopetition” against an Amazon: analyzing cooperative data-sharing between big competitors.

Tools

VaticAI - an experimental text-to-video tool.

PhotoAI - create AI-generated headshots and portraits.

Fiction - the personal AI media generator for pros.

Kalendar - book your calendar on autopilot with AI.

TWEET OF THE DAY

Tag us on Twitter @BotEatBrain for a chance to be featured here tomorrow.

AI ART-SHOW

Until next time 🤖😋🧠

What'd you think of today's newsletter?

Login or Subscribe to participate in polls.