Replit Agent Automates Software Development

Replit now allows us to spin up real apps with natural language, Reflection-70B tops the charts, and I’m loving Anthropic’s new accelerator…

Replit’s New AI Agent Will Help You Ship Faster Than Ever

Remember that startup idea you had 10 years ago? Looks like you can finally bring it to life with Replit Agent.

• The new application development tool allows both developers and non-developers to describe their app ideas in natural language, enabling the creation of fully functional applications with ease.

• Replit believes its platform holds a distinct advantage by not only generating code but also managing infrastructure and deployment.

• Currently, Replit’s AI agent is available in beta for Replit Core and Teams subscribers.

• Check out the reveal demo from CEO Amjad Masad here!

So What?

This saves developers time on repetitive tasks, allowing them to focus on creativity, while offering businesses a quicker and more cost-effective way to launch software. However, as AI tools like Replit’s advance, they will inevitably reshape the industry, pushing developers to focus more on creative problem-solving as AI handles routine tasks.

You may have recalled our coverage of Cursor, which is a tool to speed up the actual development process in a IDE. Replit Agent is a little different in that it caters to a wider audience but looks to automate the entire development lifecycle.

We might be on the verge of a future where coding becomes as easy as telling a story… Check out our take!

Reflection-70B Reigns Supreme Among Open Source Models, Or Does It?

Late last week, HyperWrite CEO Matt Shumer (@mattshumer_ on X) dropped the world’s top open source model, Reflection-70B, and when I say it’s blowing minds…

• Reflection-70B is an open-source large language model (LLM) that uses a technique called reflection tuning.

• Reflection tuning enables the LLM to detect mistakes in its reasoning and self-correct; the model performs better in tasks requiring complex reasoning and accuracy due to this new technique.

• Reflection 70B outperforms other models like GPT-4 and Sonnet 3.5 on benchmarks such as MMLU, MATH, and IFEval, achieving scores of 89.9% on MMLU, 79.7% on MATH, and 90.1% on IFEval.

• The model has been trained entirely on almost on synthetic data, powered by startup Glaive AI.

• A more capable, 405B parameter version is coming soon, but for now, you can give it a spin here and check it out on Hugging Face.

An interesting outcome of this development is that much of the AI community accused Matt of faking the SOTA model build, as issues were discovered with how its weights were uploaded to Hugging Face, is it too good to be true? 🤔

Anthropic Releases The “Quickstarts” Repo, Aimed to Accelerate Development with its API

Anthropic is cooking with this one… The AI startup just released a Quickstarts repo, which is a collection of projects designed to help developers quickly get started with building deployable applications using the Anthropic API.

• The first example is a Customer Support Agent powered by Claude, where you can have a simulated chat.

• You will need an Anthropic API key to use this accelerator. If you don't have one, you can register for free at console.anthropic.com.

So What?

This is a great step for Anthropic, and I’m sure the other major players will be releasing their own versions of this soon. It gives practitioners a starting point when developing applications powered by Claude, instead of scratching and clawing to find documentation!

The AI Craze is Seriously Distorting the VC Market

Unlike previous tech booms, venture capitalists are not playing a central role. Instead, industry giants such as Microsoft, Amazon, Alphabet, and Nvidia are investing billions to drive the growth of capital-intensive companies like OpenAI, Anthropic, Scale AI, and CoreWeave.

• AI startups are also receiving additional incentives from tech mammoths, such as cloud credits and business partnerships, which VC’s simply can’t match.

• The U.S. VC exit value is projected to reach just $98 billion in 2024, down 86% from 2021, and IPOs are expected to hit their lowest point since 2016.

• Generative AI has raised $26.8 billion across 498 deals in 2024, continuing a trend from 2023 when AI companies saw a 200% increase in funding.

This speaks to the massive monopoly big tech may pose upon AI in general. If they own the latest and greatest models, as well as being big investors in early unicorns, true democratization of AI may be in their hands…

Stay updated with our favorite highlights, dive in for a full flavor of the coverage!

Google DeepMind announced AlphaProteo, the first AI system for designing novel, high-strength protein binders to serve as building blocks for biological and health research.

Canva is dramatically increasing prices for some customers. Canva Teams subscribers on older pricing plans will see a 300% increase for a five-person plan, jumping from $119.99 per year to $500 per year for AI features.

Anthropic's prompt engineering experts reflect on how prompt engineering has evolved, highlight practical tips, and showcase their insights on how prompting might change as AI capabilities grow.

Apple iPhone 16 event shows off AI muscle. IPhone 16 is the first Apple smartphone designed around AI features.

A North Carolina man has been arrested after allegedly making more than $10 million by stuffing AI-generated songs onto streamers.

We’ve been refining our newsletter delivery schedule, and going forward, you can expect to receive two issues per week—on Tuesdays and Thursdays. Be on the lookout 😉 - RM

If you have any suggestions or specific feedback, simply reply to this email or fill out this form. Additionally, if you found this insightful, don't hesitate to engage with us on our socials and forward this over to your friends!

You can find our past newsletter editions here!

This newsletter was brought to you by Digestible AI. Was this forwarded to you? Subscribe here to get your AI fix delivered right to your inbox. 👋