Hello everyone,
Welcome back to our newsletter! This week we're diving into the latest AI developments, with a special focus on agent technology and some exciting new models that are changing the landscape.
Gemini 2 is really good for building agents!
When building AI agents, computation is everything - and cost matters. Google's Gemini 2 is a game-changer, offering powerful capabilities at just 20 cents per million tokens on average. That's 500 times cheaper than GPT-4.5 and 30 times cheaper than Claude 3.7! Learn how this economic advantage enables more complex reasoning, multiple solution paths, and continuous self-improvement loops for your agents.
Deploying an open source version of Manus
Curious about the hyped Manus agent but want to run your own version? In this tutorial, I break down how to deploy an open source alternative that gives you similar capabilities without the limitations of the hosted service. Perfect for developers who want more control over their AI agents.
The truth about Manus: Claude Sonnet with tools?
There's been a lot of hype around Manus, but what's actually under the hood? This deep dive reveals how Manus leverages Claude 3.7 Sonnet along with a suite of specialized tools to create its seemingly magical capabilities. Understanding this architecture can help you build similar systems with your preferred models.
• Manus Agent Takes the AI World by Storm: The new general AI agent from China's Butterfly Effect has spread globally, with tech influencers like Jack Dorsey praising its performance. Despite some system crashes and server overload issues, it's showing real promise with its intuitive interface and multi-model approach. Read the review
• GPT-4.5: Breakthrough or Disappointment? OpenAI's latest model has received mixed reactions. While it offers enhanced emotional intelligence and marginal improvements in response fluidity, its extreme cost ($75 per million input tokens and $150 per million output tokens) is pricing out many potential users. This raises questions about the sustainability of AI development at this scale. Read more
• Diffusion-Based LLM Shows Promise: Inception Labs has released a new language model based on diffusion techniques rather than traditional transformer architectures. Early results show impressive performance, potentially opening new avenues for LLM development. Learn more
Best regards,
Gui Bibeau