(Source)

Breaking! Anthropic launches Claude 3, the latest in state-of-the-art foundation models.

I am adding it at midnight, as it has just been announced. See its key Highlights:

1️⃣ Introducing Three New Models: Opus (large), Sonnet (medium), and Haiku (small).

Claude 3 Opus surpasses OpenAI's GPT-4 across all major benchmarks, setting a new standard for deep reasoning, advanced mathematics, and coding capabilities.

Claude 3 Opus demonstrates unparalleled fluency in task automation, hypothesis generation, and the analysis of charts, graphs, and forecasts.

Claude 3 Sonnet offers an optimal mix of intelligence and efficiency, ideal for enterprise applications.

Claude 3 Haiku excels in speed, enabling the creation of rapid, innovative user experiences.

2️⃣ Enhanced Vision Capabilities: All Claude 3 models are equipped with advanced vision features, capable of processing and analyzing images, photos, charts, graphs, technical diagrams, and text through OCR.

Unfortunately, it is not available in Germany. 😭 

If you have access, I would like to know how you like it.

(Source)

Mistral AI is considered the French rival to OpenAI, with the added commitment to open-source development.

Mistral AI has recently introduced its latest model, Au Large.

Looking at benchmarks such as MMLU, it ranks (now after Claude 3 Opus) third best, after GPT-4. And this with roughly 64 employees?!

They announced a partnership with Microsoft (who invested $16 million), making Au Large available first on the MS Azure Cloud.

For this, Microsoft offers supercomputing infrastructure, scale-to-market, and AI research support.

Unfortunately, as part of this pact, Mistral had to deviate from their open-source commitment, making Au Large a closed-source model.

Interestingly, in 2019, when Microsoft partnered with OpenAI, OpenAI's transition to a for-profit model followed directly.

Today, Mistral is changing its commercial model to closed-source.

Is this even legal?

No, according to Elon Musk. He has filed a lawsuit against OpenAI.

"OpenAI has breached its founding principles by prioritizing profit over the benefit of humanity," the lawsuit states. (See his analogy. 😄 )

What the ruling will be is yet to be seen.

However, I think that the approach by Zuckerberg (+ LeCun) at Meta is much more reasonable and better for humanity: to be fully committed to open-sourcing their models, fueling the community, and then using the community-refined models and iterations to actually improve their AI-driven products.

Don’t believe me? See this report by McKinsey, or see the low-hanging fruits AI can reach in marketing by continuing reading.

Last week, I was at the GenAI for Marketing Summit in London, giving a GenAI tools workshop, meeting high-energy, intelligent teams.

However, marketing teams must prepare for unprecedented turbulence ahead. Making the right decisions is critical.

Using AI just to say you do will not cut it.

I know 3 people who have been let go in Marketing, and cutting costs due to AI capabilities is a cross-industry trend.

A couple of points here (AI tools below):

First, to set up the right measures and implement the right GenAI Tools takes work.

Sometimes, hiring an expert who looks at your situation is best.

I could be your expert if you like. I led 10+ client projects in GenAI end-to-end.

But my calendar is a no-fly zone atm. So, I have to triage what meetings I take in March and April.

Are you interested in booking a call with me? Please apply using the button below—semi-random draw; plus, LMK if you bought my book.

Second, on GenerativeAI.net, you’ll find the online course on Generative AI for Marketing.

Juan Carlos, who was with me at the summit (see image below), is the course’s lecturer and AI Lead at Bolt.

The course provides tangible insights. There is no time for BS.

Tip of the iceberg:

• Content costs will approach $0. The democratization of content.

• Hyperpersonalization: real-time personalization tailored to individual customers.

• New marketing channels and formats, like AI Optimization. (0.01% know how to utilize this; get a competitive edge!)

• How to improve marketing efficiency + AI automated decisions.

• Creative is the new targeting: Stating that creative often outperforms technical choices.

Juan Carlos and I studied together (and even shared a flat at some point; supreme time!), and I was always impressed with his analytical mind and way of sharing insights.

So I had to get him on GenerativeAI.net, producing a course filled with valuable insights.

See yourself.

Third, I asked if you would like to hear about the top GenAI Tools for Marketing. Over 80% said yes.

Here are the first 4 tools.

Ideogram.ai

Version 1.0 dropped recently. It is an easy-to-use image generation. The text rendering is better than Midjourney’s.

Test it out by simply prompting and using “for the text.”

If you have the plus version, you can upload images to be manipulated.

I love the following output. It uses this prompt: "Super Divorce Family" Pixar-style title. A Disney movie poster, 3D characters, a father, mother, and a little son, sad faces, and superhero clothes, the father is holding documents and the mother holding a pen.

Jasper.ai

This app has probably thought the most about what marketers need.

It dynamically builds the workflow depending on the task you want to take on next.

For instance, writing a brand email asks for unique selling points, the call-to-action, and other knowledge required, building the v0 text; with high-quality.

Just confirmed: Jasper announced its acquisition of Stability AI’s Clipdrop. Now, they also have top-notch image generation capabilities.

ChatGPT, Mistral, Gemini

If prompted correctly, LM systems can produce an even better outcome.

In my workshop, I went into the top 5 prompting rules to immediately upgrade your output quality:

1. Expert Persona: Define the skill profile that the model should emulate. Example: "You are a seasoned marketing strategist…”

2. Context: Provide external or user-specific details. Example: “The following marketing campaigns were launched last quarter... ”

3. Task Specification: Instruct the model on required actions. Either a single instruction, or a sequence of instructions. Example: "Develop a hypothesis to understand the reasons behind the customer's behavior...”

4. Execution Rules: Directives that shape the model's decision-making during task execution. Example: “Adhere to our ethical standards...”

5. Output Constraints: Specify conditions for the output in terms of structure, content, length, etc. Example: "Keep it under 500 tokens."

A pro tip for working with AI is using existing GPTs.

One example of brainstorming (incl. different PoVs) is the Ideator GPT by Jonathan Kahan.

When I asked how I could sell my book best, it came up with many ideas from Psychology, Network theory, Feedback loops, flow theory, and more.

Extremely valuable. It is called Autonomous Problem Solving - Ideator.

There is much more to leverage, but where to start?

Again, if you’d like to book a call with me in March/ April, please apply using the button below—semi-random draw; plus, LMK if you bought my book.

More videos are dropping as more people get access to the Sora. See below more mind-bogling examples.

Three points:

1. This is v1. I expect v2 in 6 months (looking at the track record). Each new version has much better quality and longer videos.

2. Expect Sora to be part of GPT-5.

3. Marketing departments will once again benefit greatly from this new tech.