🤝 Are you using the right chatbot?
Dear curious minds,
This week, as top models become increasingly close according to benchmarks, we elaborate on how to choose the right chatbot.
Furthermore, the text-to-image models have seen major updates from Ideogram and Midjourney, showcasing just three weeks after the FLUX.1 release how competitive this area has become.
In this issue:
💡 Shared Insight
Choosing the Right AI Assistant: Features and Personal Preference
📰 AI Update
Ideogram 2.0 Improves Text-to-Image Generation
Midjourney Launches Web Version with Free Trial
🌟 Media Recommendation
Structured Journaling: Cal Newport's Key to Self-Discovery and Career Alignment
💡 Shared Insight
Choosing the Right AI Assistant: Features and Personal Preference
In the rapidly evolving world of large language models (LLMs), the best models are performing quite similarly across most comparisons. This convergence in capabilities raises interesting questions about how we evaluate and choose between these AI assistants.
The chatbot arena
One popular method for comparing different AI chatbots is through platforms like the LMSYS Chatbot Arena. This arena allows users to interact with multiple AI chatbots side-by-side and rate their performances. The overall rankings are calculated based on these user ratings, providing a seemingly objective measure of chatbot capabilities.
However, recent discussions, e.g. in the Benchmarks 201: Why Leaderboards > Arenas >> LLM-as-Judge issue of the
newsletter, have highlighted that these arenas may not be showing us the full picture of model capabilities. Instead, they often reflect user preferences and interaction styles. Each chatbot has its own unique way of answering queries, which can significantly influence a user's preference and, consequently, their rating.Explore chatbot styles
Given this subjectivity in chatbot evaluation, my recommendation is to explore various chatbots and find the one that resonates most with you. One excellent tool for this purpose is the Chathub Chrome extension, which allows you to interact with up to six chatbots simultaneously, making comparisons much easier.
Explore chatbot features
While the conversational style is important, it's not the only factor to consider. The capabilities of the web application and, for mobile users, the functionality of the dedicated app can be significant differentiators.
I've created a table comparing the different capabilities of various websites of popular chatbots. This table provides an overview of features such as file uploads, image generation, voice support and more. Web and Python interpreter describe the capability to run code and render webpages generated in the responses.
Mobile apps often offer a more tailored experience compared to using a website on a mobile browser. For instance, they might allow you to easily share text passages with the app or directly create and use a photo with the device camera.
Free options
The tables above show also that there are multiple free options available, which are perfectly suitable if you're not a power user. Companies like OpenAI (ChatGPT) and Anthropic (Claude) now offer their best models in free modes, but with stricter usage caps. These free tiers provide an excellent opportunity to explore advanced AI capabilities without any cost, making them ideal for casual users or those just starting to experiment with AI assistants.
Open-source options
While this article primarily focuses on cloud-based, commercial LLM chatbots, it's important to mention the rapidly growing field of open-source models. These models aren't covered in depth here as they typically require more advanced hardware and technical knowledge to set up and use effectively.
However, there are now many great open-source models available. If you prioritize privacy and prefer not to share your queries with cloud-based tools, it's well worth the time to familiarize yourself with these alternatives. They offer a level of control and customization that cloud-based solutions can't match, making them particularly attractive for users with specific needs or privacy concerns.
To get started, I recommend you to pick either the command line based Ollama or the GUI-based LM Studio application. Both are great options to run open models.
Conclusion
While the core capabilities of leading LLM chatbots are converging, the user experience, platform features, and individual preferences still play crucial roles in choosing the right tool. I encourage you to explore different options, consider your specific needs, and find the AI assistant that works best for you.
The feature comparison shows that ChatGPT in its paid option tick all except of one box. Even without major updates in recent months, it remains my top recommendation due to its versatility. I continue to rely on ChatGPT as my primary tool. Its wide range of capabilities and consistent performance across various tasks make it an invaluable asset for my daily work.
📰 AI Update
Ideogram 2.0 Improves Text-to-Image Generation
Ideogram has released version 2.0 of its text-to-image AI model, marking a significant leap in image generation technology.
After sign-up, you get 10 free credits per day and can start to explore the capabilities in a nice and intuitive user interface.
New Features:
Style Options: Users can now choose from models tuned for Realistic, Design, 3D, and Anime styles.
Color Palette Control: Allows for precise control over image colors.
Flexible Aspect Ratios: Supports various image dimensions, including 3:1 and 1:3.
Advanced Prompting: Enhanced "Describe" and "Magic Prompt" features for better creative iterations.
Expanded Accessibility:
iOS App: Ideogram is now available on mobile devices through the App Store.
API Access: Developers can integrate Ideogram 2.0 into their applications.
Ideogram Search: Over 1 billion community-generated images are now searchable for inspiration.
Ideogram 2.0 claims to surpass DALL·E 3 and FLUX.1 Pro in human evaluations. But is there a reason why they don’t show a comparison to Midjourney?
I prompted Ideogram to generate a handshake between a robot and a human. Hands are often a problem in AI-generated images, and this is still true for Ideogram 2.0 as three hands seem to miss one finger.
My take: The txt2img space is getting crowded by high-performing tools. This update of Ideogram represents a significant advancement in AI-powered image generation, offering improved quality and nice options to control the generation process for both casual users and professionals in creative fields.
Midjourney Launches Web Version with Free Trial
Midjourney, the popular AI image generation platform, has expanded its accessibility by launching a web-based version open to all users. Previously limited to Discord for most users, Midjourney is now offering a more user-friendly interface directly through web browsers.
To celebrate this milestone, Midjourney is providing a generous free trial to both new and existing users. Everyone can now create 25 AI-generated images at no cost, allowing them to explore the platform's capabilities.
To get started, simply sign up using your Google or Discord account on Midjourney's website. The intuitive web interface lets you input text prompts to generate images, with options to adjust various settings like stylization levels and aspect ratios. This move not only simplifies the user experience but also opens up Midjourney's powerful AI art tools to a broader audience.
I prompted Midjourney to generate a handshake between a robot and a human. Hands are often still a problem in AI-generated images, and this is also still true for Midjourney, as one of four result images shows a hand with too many fingers. Furthermore, there are multiple generations where both hands look rather robotic to me.
My take: This strategic move by Midjourney comes at a crucial time in the AI image generation landscape. With competitors like Ideogram and FLUX.1, Midjourney's simultaneous release of a web UI and free trial appears to be a calculated effort to maintain its pole position in the market. By lowering barriers to entry and enhancing user experience, Midjourney is clearly aiming to solidify its user base and attract new creators.
🌟 Media Recommendation
Structured Journaling: Cal Newport's Key to Self-Discovery and Career Alignment
I enjoyed listening to episode 313 of Cal Newport's "Deep Questions" podcast, titled "Structured Journaling".
Cal introduces the concept of structured journaling as a tool to help you identify what truly resonates with you. He suggests keeping a notebook to document experiences and environments that appeal to you, then reviewing these notes monthly to spot patterns. This practice can be instrumental in gaining clarity on your personal values and preferences.
He emphasizes on a lifestyle-centric approach to career choices. Rather than fixating on job titles, he advocates for aligning your work with your ideal lifestyle and personal values. This perspective shift could be transformative for those feeling stuck or unfulfilled in their current paths.
If you're seeking more depth in your life but struggling to define what that means for you, this episode provides practical strategies to help you clarify your goals and make intentional changes. It's a thought-provoking listen that could spark significant personal insights as you plan for the future.
My Take: While Cal's structured journaling technique is powerful on its own, I believe AI can significantly enhance this process. AI algorithms can help uncover patterns and extract insights from your journal entries that might not be immediately apparent to the human eye. This could provide an extra layer of self-discovery and personal growth.
You don’t need to use cloud-based AI solutions like ChatGPT or Claude, which might raise privacy concerns when dealing with personal journal entries. As discussed in several past issues, open AI models are becoming increasingly sophisticated and accessible. These models can be run locally on your own device, ensuring that your most intimate thoughts and reflections remain private while still benefiting from AI-powered analysis.
Disclaimer: This newsletter is written with the aid of AI. I use AI as an assistant to generate and optimize the text. However, the amount of AI used varies depending on the topic and the content. I always curate and edit the text myself to ensure quality and accuracy. The opinions and views expressed in this newsletter are my own and do not necessarily reflect those of the sources or the AI models.