Kompas AI

So, it is clear that the direction is to move forward from a simple AI model to an AI system. Plus, GPT-4.5 will be delivered in weeks, and GPT-5 in months.

Introducing Expert Mode and Celebrating a Product Hunt Milestone 1. Expert Mode Release

I'm working on this 'Start from Blank' feature—create a report instantly by simply outlining and hitting a button. What do you think about the interaction flow? Is it intuitive?

A few days ago, I demonstrated how to conduct comprehensive research and create a detailed report using AI. Today, I'll show you a demo of how we can achieve equally good results on a Mobile phone.

Developing LLM products with chat-based UX inevitably leads to competing with ChatGPT. Since most developers first experienced LLMs through ChatGPT, it became the default interface, causing most LLM products to follow suit. Competing with ChatGPT is pointless. To create better products with unique positioning, LLM UX design must move away from chat interfaces.

A brief look at an AI-based approach to research and report writing. See how the process shifts from traditional methods to a more dynamic workflow.

Introducing a Research + Report gen AI product Write professional research reports in minutes, not days.

While IQ isn't the whole story of intelligence, the launch of o1 marks a major leap in AI prowess. Achieving an IQ of 120 by combining and repeating various LLMs opens the door to cheaper models with expanded contexts and agentic flows. Soon, we'll see swarms of AI agents tackling tasks across countless platforms.

Why Great Bosses Excel at Prompt Engineering? AI commands, known as 'prompts', can drastically change the results we get.

AI will soon surpass human abilities and become superhuman? This notion seems contradictory. The definition of human abilities and how LLMs are trained and developed are crucial in this context. LLMs are inherently limited by the human language and texts they are trained on; they cannot exceed the boundaries set by these texts. No matter how perfect and complete an LLM becomes, it cannot surpass the level of human texts. In terms of capabilities, LLMs might match human abilities by working much faster, handling more tasks, and being replicated to manage even more work. If this is what we mean by surpassing human abilities to a superhuman level, then it is possible. Essentially, it would mean having many highly capable entities. However, can something beyond this emerge? Is there a potential for a new creative ability that transcends text? While blending and combining current human capabilities in new ways might create new value, it would still be within the realm of human understanding. It wouldn't signify a level of ability beyond human comprehension. What do you think?

Let me introduce the latest version of Kompas AI: it features multi-session capabilities, multiple AI agents, and autonomous AI agents. Anyone can easily get great results without complex prompts.

Kompas AI employs similar approaches. A meta agent analyzes and redefines tasks, while another selects or instantly creates suitable agents. Additionally, Kompas AI also deploys multiple agents simultaneously for complex projects.

As Flux.1 can create images without copyright concerns, I created some city images in different renowned painters' styles. Check this out. 🧵 The first one is San Francisco in Yayoi Kusama's style

If there is a clear ceiling, it implies that everyone will eventually reach it. Language models (LLMs) are based on human language, which acts as a constraint. This means that LLMs cannot surpass human cognitive abilities. Even if some 'capability' emerges that exceeds human abilities, we may not be able to perceive or evaluate it, dismissing it as mere hallucination. Therefore, the so-called Artificial General Intelligence (AGI) is likely to be an inductive sum of all capabilities that humans can comprehend. Regardless of whether the models are closed or open, they are expected to converge to the same level within a few months. The competition will then shift from capability enhancement to speed and cost improvements. And It seems like OpenAI has already shifted its focus towards multimodal UX. The revenue structure for models will continue to deteriorate, and inference costs are likely to be reduced to compute costs. We must prepare for this scenario and pay closer attention to AI agents that can effectively leverage these models. The focus should be on developing applications and agents that can harness the full potential of LLMs, rather than solely relying on model improvements.

Contemplating the concept of a new internet, future applications, and websites: 1. Generative AI is producing vast amounts of content, optimizing for SEO, and dominating search engine rankings. This trend is expected to intensify. 2. The proliferation of similar content will exponentially increase the volume of material one must consume to find original ideas and develop comprehensive understanding. 3. AI agents will likely become necessary intermediaries for information retrieval and integration, reviewing and summarizing extensive content on our behalf. 4. This shift may lead to the development of AI agent-friendly websites, RSS feeds, or APIs, potentially reducing direct human web browsing. This could diminish the significance of websites designed for human interaction. 5. AI agents will recompile gathered information, presenting it in user-preferred formats such as summaries, reports, slides, or even short-form videos. Personalized outputs will be facilitated by generative AI capabilities. 6. Similar transformations may occur in interpersonal communication methods and vertical applications. Traditional forms of communication such as texting and email may evolve into different formats. 7. The internet ecosystem may shift from provider-controlled information dissemination to user-centric information acquisition, mediated by AI agents. This premise underlies my approach to AI agent development and product creation. What's your idea? I'm open to discussing alternative perspectives or ideas related to this concept.

AI search often resembles this meme. Many AI search products prioritize speed, potentially leading to hallucinations and superficial information. As professionals, researchers, and diverse users, you need a solution that offers depth, accuracy, and reliability.

Many people still do not use AI in their work, despite the fact that the current level of AI is sufficiently advanced to be applied in the workplace, and has the potential to bring revolutionary improvements to the majority of people's work areas. Nevertheless, the reason why many still do not utilize it largely stems from issues with the user experience (UX). At first glance, Conversational UX can seem deceptively simple, this may lead users to misuse it. When we engage in 'conversation', we do so against a backdrop of various contexts, and the acceptance of chat-based AI often comes with the mistaken belief that the AI will understand context as well as a 'human' does. This leads to incorrect 'prompting', which can result in subpar outcomes and, consequently, users, disappointed with these results, may quickly underestimate the AI's capabilities. Therefore, it could be said that conversational UX is a double-edged sword. Overcoming these challenges, I believe, will lead to the true mass adoption of AI.