World's best personal AI?

This newsletter just crossed 10,000 readers ✨ THANK YOU for trusting me to bring the latest in AI to your inbox every week.

I’m excited to keep improving the format to give you the most meticulously curated digest of what's brewing in AI possible. No hype included.

‍

What’s brewing in AI #30

• OpenAI’s response to Elon Musk (and Elon’s latest surprise)
• Inflection launches “the world's best personal AI”
• Google penalizes AI generated content
• GPTs: Great article with real-life examples of using GPTs at work
• The other top stories in AI this last week

‍

Dario’s Picks:

I. OpenAI fires back at Elon Musk

‍

‍

Elon recently sued OpenAI claiming the company has been transformed into a closed-source subsidiary of Microsoft, betraying its agreement to develop AI for the benefit of humanity rather than for profit.

OpenAI responded by sharing their own perspective, including publishing several email exchanges Elon has had with the company. The gist is that Elon initially understood that the Open in OpenAI doesn’t mean open-source, but started complaining “when we started making meaningful progress towards OpenAI’s mission without him”.

And yesterday came a small plot twist: Elon Musk will open-source Grok

‍

‍

As the tweet reads, Elon Musk has decided to not just talk the talk, but also walk the walk.

Why it matters: Any personal vendettas between Elon and OpenAI aside.. I think there’s a real and legitimate question here: Is OpenAI really open and should it be? I think the answers here are no and yes. But openness is both risky for a company and it’s a spectrum, so there’s definitely nuances here.

‍

II. Inflection’s new model: “the world's best personal AI”

‍

‍

Inflection are the makers of Pi, a more personal and empathetic AI with 6 million active users. They just dropped their new model, Inflection-2.5, and it’s on the level of world-leading LLMs like GPT-4 and Gemini (according to the company, anyway).

If you’ve used Pi in the past, you can expect it to now be much better at things like coding and mathematics.

Why it matters: Pi is one of the most popular AI companions out there – and make no mistake AI for companionship will be big. With its recently improved reasoning skills, Pi and Inflection’s latest model might be ideal for people wanting a ChatGPT alternative with better EQ.

‍

In Focus

Google penalizes spammy sites – including AI generated content

‍TLDR: Google’s new update is affecting sites with “scaled” content negatively, making businesses more cautious about using AI, and pushing them to diversify to other traffic channels.

An important one for marketers out there.

Google’s March update includes new ways to fight spammy content. Senior marketers agree it’s one of the biggest algorithm updates ever and, with Google accounting for over 90% of global searches, that’s affecting a lot of businesses.

Two changes are affecting a lot of content generated with AI negatively:

1) the policy change towards scaled content (content produced at scale to boost search rankings). This type of content has increased a lot after the launch of ChatGPT.

2) manual penalties to specific sites with Google flagging, or even deindexing, sites for “automatically generated gibberish” among other things.

‍

My take on this:

With several major websites taking a huge hit to their search traffic, having been flagged for scaled content or received a manual penalty, this is likely to set precedence for how people and businesses use AI in their content going forward.

My short-term bet is that legitimate content creators will get more cautious about using AI to scale their content. Having your site deindexed by Google is a risk not worth taking for most. Smart creators will also focus on diversifying their traffic channels going forward, especially to email, social media and other online communities, and YouTube.

‍

GPTs

I'm having some technical issues with whatplugin today, so I’ll be back with the top new GPT arrivals next week.

Meanwhile, check out this useful article with real-life examples on how to use GPTs at work: You Should be Playing with GPTs at Work

‍

Featured GPTs (sponsored)

Coinz.com

💬400+

A Crypto specialist by coinz.com.

‍

‍

❞ Pretty neat. This is a GPT that provides answers to questions related to crypto, and is connected to a cryptocurrencies portal with information about the different coins (Coinz.com). I tried asking some questions and it gave me fast and succinct insights on how different coins are performing.

Submit your GPT to whatplugin and get featured in the newsletter here.

‍

Bytes

• A great overview of the economic case for AI. Check out this 10-minute presentation by a16z: The Future of AI is Amazing.
• For some forward-thinking AI use cases, check out the winners from the South Park Commons x OpenAI hackathon. My personal favourite has got to be an animated laptop cat that introduces itself as "I’m Toby, the AI cat, created by Michael. I’m here to keep him company because he’s a lonely loser”. 😸
• OpenAI introduced a new tactic to get free users of ChatGPT over to a paid plan: there’s now an option next to chat responses to upgrade the response with GPT-4, asking users to “Upgrade to plus”.

Legal

• OpenAI completed a board review which confirms Sam Altman and Greg Brockman will continue in leadership. They also introduced some new and improved Governance structures.
• Ex-Google engineer charged with stealing and transferring AI secrets to a Chinese company.
• Microsoft engineer warns that the DALL·E-powered Copilot Designer has weaknesses that can make it generate sexual and violent content.

Partnerships

• Cohere + Accenture: New collaboration to bring Cohere’s models and RAG capabilities to Accenture’s clients.
• Mistral AI + Snowflake: Goal is to bring Mistral’s models securely to enterprises through Snowflake’s data cloud platform.
• Zapier acquired Vowel: Zapier is a popular automation platform, and Vowel is an AI-powered video meeting tool

Product

• Zapier launched Zapier Central - an experimental AI workspace where you can teach bots to work across 6,000+ apps.
• ChatGPT now has multi-factor authentication.
• Pika Labs can now do lip syncing.

Research

• Crowdsourcing LLMs evals (testing LLMs against each other) is trending. You may already be aware of the popular Chatbot Arena leaderboard on Hugging Face. There’s now a new leaderboard aimed at comparing LLMs on complex tasks specifically, the Wildbench leaderboard. If you’re wondering why grading AIs like this is necessary in the first place, check this great article by Michael Taylor.
• Scale AI released a safety evaluation benchmark for LLMs: the Weapons of Mass Destruction Proxy (WMDP). It aims to give a measure of how much an LLM knows about risky stuff like biosecurity, chemical security, and cybersecurity. Scale AI has previously made a widely-used benchmark for LLMs, the Massive Multitask Language Understanding (MMLU).

‍