🔥 Weekly AI news: Did you miss it?!

👨‍🚀 The most important AI updates at a glance

AI-HOI, AInauts!

Maybe you didn't catch all the news, tools, and hacks about AI last week, or maybe you've only recently joined us. Either way, here's our recap with all the headlines from the newsletter - just one click away!

Click the links to jump right to the article - or read our picks below.

• 🎉 SPECIAL: The next generation AI is here: Discover GPT-4o

• 🎶 Fun song: As an AI language model ...

• 🐙 Microsoft launches the ultimate data-mining giant?

• ✍️ New update of the Claude Prompt Generator

• ✅ Which LLM to use for which purpose

→ Selection of the top posts of the last week ←

🎉 The next generation AI is here: Discover GPT-4o

May 13 will go down in history. Just like November 30, 2022, when OpenAI’s ChatGPT was introduced to the world. Whoever thinks that the new GPT-4o(mni) model is only an incremental improvement is wrong.

This really is a next-generation AI! ChatGPT is still a long way from being a jack-of-all-trades, but it is more human-like and smarter than anything that has come before.

Before we get into the practical ideas, let's first summarize what the highlights of OpenAI’s Spring Update were.

We recommend that you at least take a look at the supercut of the most important updates - even better, the entire presentation and demos to really take it in. And you can find plenty more examples below.

Omni(present) and multimodal

• The new GPT-4o (=omni) model is faster, improved and freshly trained (Knowledge Cutoff: October 2023).

• It is multimodal from the ground up (!) and can therefore process audio, images and text in real time.

• In terms of response quality, it is the most advanced model in many areas, with a super-fast response time.

• It is now gradually being rolled out on the mobile apps, in the web version and also for desktop.

• It is also available via the API interface - and is 50% cheaper than GPT-4 Turbo, but twice as fast. The context window is 128k tokens.

The new voice assistant = your personal helper in all areas

• The new voice features are a game changer (yes, an overused word, but a fact). Siri, Alexa and the others seem like a relic from the past in comparison. He (she?) is the best we’ve seen yet in terms of form and expression - just as we know it from the film "Her". Fun fact: Scarlett is not happy …

• Our prediction: This voice assistant (with or without Scarlett’s voice) will give many of us a completely new user experience and take the frequency and quality of use to a new level.

• The previous voice function is currently experiencing a renaissance and is being used by many users for the first time to get a taste of it.

• Bummer: The new voice features are not yet available for most Plus users, but will be rolled out "in the coming weeks". We are already eagerly waiting ...

Your computer assistant looks over your shoulder (but not always)

• There is also a new desktop version (initially only for Mac), which is always just a shortcut away in day-to-day work.

• We are already using it and are trying to support computer work as much as possible and adapt our routines.

• This shows how important an intuitive user interface is - you can take a screenshot of any app with just two clicks and ask questions about it.

• You'll get a message in the ChatGPT interface when the desktop version is available for you - or you can follow this tutorial here to start using it right away (it's a bit technical).

• And if you are located in Europe, it may make sense to use a VPN with an American IP address in order to be able to use all functions.

A huge upgrade for users of the free version

• The really brilliant thing is that OpenAI will offer the new model with the existing Plus features free of charge for all users.

• This includes access to web results, analysis of uploaded files (texts, sheets, images, ...), the use of GPTs and the GPT store, the memory function and more.

• Gone are the days of GPT-3.5 with mediocre responses. You have a limited number of queries in the free plan before the system reverts to the old 3.5 version.

• The new voice assistant(s) features will initially only be available to paying users.

Where will GPT-4o make an impression in real-life?

• GPT-4o is a multimodal language model and therefore the perfect translator! Even with the previous language version, a lot can already be done (it even understands language dialects). Language barriers, goodbye - and translators or language teachers will soon have a harder time.

• The new voice assistant can also understand human emotions (!) and respond to them, making it a human-like conversation partner. This opens up new possibilities for customer support, therapy and entertainment.

• Virtual boyfriends/girlfriends have long been a popular form of interaction with AI. And of course, this trend will continue, it/he/she will also become the best boyfriend/girlfriend of many, stay in character and remembers thing, thanks to the memory-feature.

• Think of character.ai or the providers of uncensored AI companions. We are curious to see whether these providers will have to face a drop in visitors in the coming weeks and months.

• On TikTok, some resourceful influencers got the viral "Dan" trend rolling and built themselves a flirty ChatGPT chatbot - still based on the old voice model, the new o model even more human and emotional.

An expert in all disciplines

• o’s ability to analyze and visualize data makes it easy to present complex information in an understandable way and answer specific questions.

• What was not shown in the demo are the new possibilities to create images, fonts, logos and other design elements, or to generate 3D models from text descriptions. These examples look insane!

• You can use it to create prototypes or visualizations super-fast, from marketing to product development to architecture. Check out the most important use cases here on the page.

• You can also very easily explore different perspectives on a topic, refine arguments and prepare for real discussions - or expand your own world view.

• One of the biggest benefits is the ability to use the chatbot as an interactive teacher! This democratizes access to education and everyone has their own personal tutor/teacher.

• The desktop version also includes a meeting assistant that can listen in, take meeting minutes and provide summaries or answer questions - or even moderate the meeting.

• The new Omni model can be used as an assistant for visually impaired people. Through object recognition and navigation assistance, it enables greater independence in everyday life.

• We have just spoken to someone in our private circle who suffers from macular degeneration and has more options as a result this new release.

The AInauts' conclusion and what it means now

The race has once again intensified by several measures. GPT-4o may be the new benchmark, but Google has already presented similar ideas.

Despite all the enthusiasm, we are still wondering how this race will develop and what else will happen in the coming months.

In any case, we will continue to put GPT-4o to the test and keep you up to date with the latest developments and possibilities.

🎶 Fun-Song: As an AI Language Model …

Wow, that was a lot of GPT-4o talk... To round things off, here's a catchy AI song, composed with Udio.com. Listen in and be amazed!

✍️ New update of the Claude Prompt Generator

We all know that prompts are essential. Because particularly good prompts also deliver particularly good results.

We no longer just write our prompts ourselves, but have them written and improved by AI.

We have presented various tools for this, and you may also be familiar with one of our favorites on Google Colab.

However, this is not as easy to use as we would like it to be.

Recently, Anthropic, the creators of the Claude chatbot, updated their in-house prompt generator.

We tested it, and it really delivers great results - and is incredibly easy to use.

This is how it works:

Step 1: Create a free account with Anthropic

You get 5 USD credit to start with, so you don't even have to pay anything. If you only use the prompt generator, the 5 USD will probably last for months.

Step 2: Describe your task

Once you have opened the prompt generator directly from the dashboard (see above), simply enter your task.

You will then receive a very well-structured, optimized prompt in just a few seconds.

This opens in a dashboard where you can make adjustments, fill in variables and test them immediately.

Creating a Facebook ad from a one-liner thus became:

You will be writing a high-converting Facebook ad designed to drive direct response and sales for a specific product, targeting a defined audience, with a compelling offer. 

First, carefully review the following product details:
<product>
{{PRODUKT}}
</product>

Next, analyze this information about the ad's intended target audience:
<audience> 
{{ZIELGRUPPE}}
</audience>

Consider this specific offer or deal that the ad will promote:
<offer>
{{ANGEBOT}}
</offer>

Write the text for the Facebook ad inside <ad_copy> tags. Focus the ad copy on the key product benefits and features that matter most to the target audience. Use persuasive language and proven direct response copywriting techniques to encourage clicks and conversions. Make sure to include a clear call-to-action that creates a sense of urgency and compels the reader to take immediate action.

After you've written the ad copy, provide a brief justification inside <justification> tags explaining your approach and why you believe this ad will be effective at driving sales from the target audience.

Something like this will give you much better results than the original notes on the task. If you are satisfied, you can then save the prompt and only have to fill in the variables in the future.

Have fun, and happy prompting!

✅ Which LLM to use for which purpose?

Finally, a quick tip from one of our favorite podcasters, Lex Fridman.

We are often asked which model/tool should be used for what.

His tweet sums it up quite well:

We would add the following here:

When we analyze figures, we still find that Claude Opus sticks to the real data much better than GPT 4o.

It's not irrelevant, so that the analyses are also correct 😎.

That's enough for today. See you soon with a fresh round of news, hacks and insights!

Your AInauts, Fabian & Reto

Follow us on Twitter & LinkedIn!

Your feedback is essential for us. We read EVERY comment and feedback, just respond to this email. Tell us what was (not) good and what is interesting for YOU.