On Open Source and Praveen's Latest
We continue to believe that the rise of open source models and AI agents will be the defining trends of AI's impact in 2024.
Open-source VS Proprietary Models
We have seen a decisive increase in the competition between these two types of models. Having tested broadly across both, we think the status quo still currently favors proprietary models for most text-based and creative tasks. Specifically, the launch of Opus, the cutting edge model by Anthropic (their chat is called Claude) has shown a substantial leap in reasoning, creativity and capabilities.
OpenAI has responded with an upgraded Turbo version of GPT4 recently, but we still vote for Opus as the best performing model at the moment (for text queries). Sadly, Opus is only available through the playground/console and the paid APIs, while the Claude chat interface now offers the somewhat less capable new model (called Sonnet).
A Turning Tide
But we believe capabilities will shift more in favor of open source, and soon. This is especially important for those of us who value unconstrained conversations, less censorship and less filtering. Looking ahead, we see a future where customized open-source AI models are the norm.
These models, if made with care and responsibility, could mark the start of a new chapter – one where AI is both ethical and built on the up-and-up. We’re thinking this could be the real deal later in 2024 or next year. For those of you who want to test the waters, the launch of Facebook’s open source Llama 3 model less than a week ago marks another leap in that direction. It is currently seen as the best open source model out there, and can quite easily be accessed by non-coders through software like Ollama and LM Studio. Feel free to reach out if you have any questions on those!
But, as Praveen points out, we’re not there yet. Right now, we need AI that’s not just smart but also ethical – as well as systems that can learn on their own and gather data without relying on unlicensed data.
Heading in this direction means tackling a few big issues:
Responsible Open-Sourcing: We’ve got to be real about the risks of just throwing open-source models out there without thinking it through. Platforms like HuggingFace, which hosts thousands of these models, need tighter rules to make sure open-source AI is helping, not hurting, content creators.
AI’s Learning and Adaptation Challenges: AI today is great at repeating tasks based on what it’s seen before, but not so hot at picking up new tricks on its own. This “model drift” means AI has to get better at adapting and growing, kind of like how we humans do. Agent models, which we keep going back to, promise to tackle some of these issues, but we are struggling to find solid, reliable framework, and they are definitely not ready for the average tech consumer! We also see some fringe companies like Verses who are hinting in their social media activity at such capabilities, but proof of their claims has been a little scarce for our taste. We are skeptical but watching.
Ethical Data Sourcing: The data we feed AI makes a tremendous difference. We’ve got to make sure it’s coming from the right places, so the AI doesn’t end up biased or breaking copyright laws. There has been a fantastic announcement by Fairly Trained AI, the gist being that ethically sourced AI is possible and is coming. Some good steps have been taken with text-to-video models, too, with Adobe paying video owners to train its competitor to OpenAI’s Sora model (which has already been outed by its CTO, embarrassingly, as having scraped YouTube content without licensing it).
The open-source AI scene has a ton of potential, but getting there means facing these big issues head-on. We will definitely coming back to these issues with some proposals of our own.
Praveen’s Update on Productivity and Innovation
Hello everyone,
I wanted to share some updates on what I’ve been working on lately, particularly my newest project, called MevinX.
MevinX is a tool I’m building to help increase online presence and engagement, something that I believe many of us struggle with. I’ve been using MevinX myself and have seen promising results, being able to engage more meaningfully with posts that align with my interests in just 30 minutes a day.
This tool isn’t here to replace us, but to enhance our capabilities. Think of MevinX as a copilot who navigates online discourse with you.
MevinX helps grow social media engagement with the help of customizable AI
It will allow users to feed their marketing documents to MevinX and the tool will find relevant tweets, LinkedIn posts, and automate engagement to keep their social presence active
Further on, I envision MevinX as helping build a brand with a tailored voice. It will help generate necessary text, images, GIFs, and video posts from existing marketing content.
Alongside building MevinX, I’ve been exploring various AI tools and models to increase my productivity and efficiency. I’ve discovered some useful hacks, like using custom instructions with ChatGPT to get more concise and relevant responses (my latest favorite is, weirdly and somewhat funnily, to “quite yapping!”). It’s important to note that while AI can be a powerful tool, it’s crucial to use it responsibly and not rely on it entirely, as it still lacks the full autonomy and decision-making capabilities of the human mind.
On a more personal note, I’ve been reflecting on my career journey and the balance between being a generalist and a specialist. I realize that being a generalist has allowed me to gain a vast amount of knowledge, but diving deep into a specific problem as a specialist can lead to groundbreaking solutions. I believe the key is to be a generalist to gather knowledge and connect the dots, but to specialize when a problem demands it.
As I continue to build products and startups, I’m always eager to share my experiences and lessons learned with the community. I believe that by sharing we can all grow and support each other in our entrepreneurial and creative journeys.
Stay tuned for more updates on MevinX and my other projects. Reach out if you have questions or just want to connect!