Our Menu :)

• Beyond the Hype: A Closer Look at Claude 3 🔍

• Think You’re AI-proof? Think Again: Case Klarna 🇸🇪.

• OpenAI Fires Back at Elon Musk. 🍲

• The Biggest Funding Deals in AI This Week 💰

Actionable Tips:

•  3 Steps to Make ChatGPT Write Like You

Read time: 4 Minutes

GEN AI STARTUP
Beyond the Hype: A Closer Look at Anthropic's Claude 3 Models 🔍

Mar 4th, 2024. Anthropic has unveiled its latest innovation, the Claude 3 series, featuring three new models:

• Opus is tailored for deep, comprehensive processing, perfect for tackling complex challenges.

• Sonnet combines speed with efficiency, ideal for scalable operations.

• Haiku offers rapid responses, enhancing live interaction experiences.

Now available on claude.ai and the Claude API, Opus, and Sonnet are ready for global use in 159 countries, with Haiku being released soon.

After the release of Gemini, we are more cautious with the hype:

What is new?

• Each model boasts a knowledge update as recent as August 2023.

• 🥇 Opus leads in text benchmarks, surpassing GPT-4 and Gemini 1.0 Ultra, backed by a 200k context window and 59% vision support on MMMU.

• 200k context window

• Vision support (59% on MMMU)

• Sonnet doubles the efficiency of Claude 2, enhancing task automation and workflows. I

  t's already accessible on Amazon Bedrock and will soon be previewed on Google Cloud’s Vertex AI Model Garden, with Opus and Haiku to follow.

• 🌍 Supports also German, Spanish, French, Italian, Dutch, and Russian

• 🔜 Soon: function calling & interactive coding (REPL)

  𝗖𝗼𝘀𝘁:

What is matter?

Domain Expert Benchmarks: Claude 3 excels in specialized areas such as Finance, Medicine, and Philosophy, offering a clear performance snapshot for professionals in these fields.

This targeted approach is a call to action for all LLM models to provide detailed benchmarks, aiding users in selecting the right tool for their needs.

Refusal Rate Analysis: The tendency of LLMs to default to overly cautious responses is on the radar.

Anthropic acknowledges this challenge and is actively working on refining the balance, ensuring AI interactions are both safe and informative.

𝗖𝗼𝘀𝘁:

• Opus: $15 per 1M input tokens and $75 per 1M output tokens

• Sonnet: $3 per 1M input tokens and $15 per 1M output tokens

• Haiku: $0.25 per 1M input tokens and $1.25 per 1M output tokens

In comparison to the competition:

• GPT-4 Turbo: $10 per 1M input tokens and $30 per 1M output tokens

• GPT-4 (32K context): $60 per 1M input tokens and $120 per 1M output tokens

• GPT-4 (4K context): $30 per 1M input tokens and $60 per 1M output tokens

• GPT-3.5 Turbo: $0.50 per 1M input tokens and $1.50 per 1M output tokens

• Gemini 1.0 Pro: ~$0.50 per 1M input tokens and ~$1.50 per 1M output tokens

GEN AI AT WORK
Think You’re AI-proof? Think Again: Case Klarna 🇸🇪.

Klarna, a leading global retail bank, has introduced an AI assistant powered by OpenAI's ChatGPT.

This tool is designed to handle customer inquiries and resolve errands, and it's been making waves since its inception.

After a successful 4-week pilot, it's now live globally.

What's New: 

• Klarna's AI Assistant, in its first month, managed to handle nearly 70% of customer requests independently.

• It's doing the job of an equivalent of 700 full-time agents.

• It's available in 23 markets, 24/7, and communicates in more than 35 languages.

Why Does It Matter? 

This development is significant for several reasons, according to Klarna:

• Firstly, the AI assistant is on par with human agents regarding customer satisfaction.

• Secondly, it's more accurate in errand resolution, leading to a 25% drop in repeat inquiries.

• Customers now resolve their errands in less than 2 minutes compared to 11 minutes previously.

• Lastly, it’s estimated to drive a $40 million profit improvement to Klarna in 2024.

Think you’re AI-proof? Think again…

The harsh reality is that AI can already do many white-collar tasks better than humans can, and this trend is gaining momentum.

Like, look at the impact ChatGPT has already had on some freelance jobs:

Though correlation ≠ causation, the plunge in writing, translation, and customer support gigs isn’t a coincidence.

So how do you know what’s safe from AI?

AI makes some skills less valuable, such as basic writing, simple data analysis, and mundane admin tasks.

And it elevates the importance of other skills, such as those involving high EQ—communicating well, people skills, conflict resolution, etc.

Our guidance: Learn what parts of your job an AI can do and what’s uniquely human about your job. But be honest.

A friend of ours recently told us, “there’s no way an AI could craft PowerPoints like me…we follow a very specific template”. Uhhh, how do we break it to him that this is the exact thing AI will be able to do soon?!

For the immediate future…

• Know how to describe your work to an AI.

• Be willing to tinker with AI tools and update your skills.

GEN AI STARTUP
OpenAI Drama pt2: OpenAI Fires Back at Elon Musk. 🍲

(Previously in OpenAI Drama)

What is new?

Elon Musk sued OpenAI in a lawsuit last week that alleged that the startup had abandoned its original mission and started chasing profits instead.

What was said ?

Elon Musk’s lawsuit against OpenAI may fail, or even get dismissed. But it’s already brought some important facts to light.

OpenAI was so upset about the optics that they rushed out a statement, proclaiming ‘We are dedicated to the OpenAI mission and have pursued it every step of the way.”

Once united by a dream of open-source AI for the betterment of humanity, the narrative twists as OpenAI's path veers into a shadowed path of exclusivity and profit. According to Elon, right ?

TL;DR:

• Elon Musk initially agreed to fund OpenAI to compete with Google, but backed out when OpenAI refused to give him majority ownership, board control, and the CEO position.

• Elon also agreed that it was acceptable to keep AI private as it became more powerful.

• In a recent development, Elon jokingly proposed that OpenAI rebrand to "ClosedAI" if they want to see the lawsuit dropped.

  
  😅 

GEN AI STARTUP
The Biggest Funding Deals in AI This Week 💰

• Tech giant Alibaba invests $600M in Chinese AI startup MiniMax, in a deal that values the startup at $2.5B. Minimax builds ChatGPT-like services.

• Blackrock invests $75M in RapidSOS, an AI startup that helps emergency response teams.

• AI productivity startup Ema announces $25M raise to build a ‘Universal AI employee.’

• Overjet, an AI platform for dentists, has raised $53.2 million in Series C funding, marking the largest ever AI investment in the dental field.

• Baseten raises $40M to make it easier for companies to deploy AI applications.

GEN AI AT WORK
3 Steps to Make ChatGPT Write Like You

A new chat with ChatGPT is a blank slate.

To get what you want, you need to be descriptive.

When it comes to describing tone of voice, we ALL use the same words:

• 😊 Friendly

• 🔥 Warm

• 💬 Conversational

• 💼 Professional

• 🤪 Quirky

• 🎣 Engaging

It's no wonder that we get the same boring results.

It's better to SHOW ChatGPT what you want to sound like, instead of telling it.

Here's how:

1️⃣ Give it examples

Gather examples of pieces of your writing that you like: emails, LinkedIn posts, articles, etc.

If you don't have anything, you can sit and write a paragraph or two and use that instead.

2️⃣ Prompt a ToV guide

Use this prompt:

Act as an extensively experienced copywriter, trained in writing tone of voice guides specifically for ChatGPT. I'm a [insert job title], I [insert job details] and I'm looking for you to write an in-depth tone of voice guide based on examples I'm going to give you.

I want you to focus on the four dimensions of tone of voice:

Humour

Respectfulness

Formality

Enthusiasm

Also analyse the content patterns around sentence structure, word use, perspective, language types, grammar, punctuation and emoji usage.

Remember the aim here is to be able to replicate this tone of voice precisely by using the guide you produce, so be detailed.

I'm going to feed you one piece of content at a time: an email, a LinkedIn post, an article, etc. I'll tell you each time what it is. Please read and analyse each one and ask me for the next one. When I've finished, I'll say "I'm finished" and you can create the guide.

Produce the report in markdown. When you're ready for the first piece, let me know.

Then add a few pieces of content that really worked for you: LinkedIn posts, emails, articles, whatever you've got. Or sit and write a paragraph or two.

3️⃣ Use it in your new prompts

Go and save the resulting guide in a separate doc so you can find it again.

Then, whenever you are next prompting to create content, add this guide into the bottom.

***

Ta-da - maybe not perfect, but content that should sound a lot more like you.

*Replace ChatGPT with any other LLM: Claude, Gemini, Copilot, etc.

Important notes:

💡 Remember the 'context window' - that's how much ChatGPT remembers of your conversation. It's about 2000 words for the free version and about 6500 words for the paid version.

💡 You can get up to about 30,000 words in the ChatGPT Playground, which is the developer version of ChatGPT, with more features to fiddle with.

💡 Not a fan of your own tone of voice? Use extracts from a creator or author you do like as your examples instead.

All Credit to Heather Murray

Thank you, see you next week!