Welcome to the 222 new subscribers who joined last week. I’m thrilled that you’re here to learn with me.

What’s brewing in AI #31

Cognition AI launched Devin, a new tool that can code better than any other chatbot so far, referring to it as “the world’s first fully autonomous AI software engineer”. It can plan and execute complicated engineering tasks.

On SWE-bench, a benchmark for language models to solve Github issues, Devin outperforms all of the biggest LLMs with good margin.

If you’re curious how it works, I recommend checking out some of Cognition’s demo videos of Devin in action:

Devin is currently in early access only, you can apply for access here.

Why it matters: Do I think Devin is going to replace software engineers? No. But I’d say it’s probably going to change software engineering; ChatGPT has already made most developers far more efficient, and tools like Devin stand to increase productivity further.

It’s not going to be able to produce production-grade for most businesses just yet, although it might at some point. However, it could likely already automate a lot of the simpler and often tedious work assigned to junior developers.

I think Cognition’s own co-founder put it well:

A lot of AI stuff happening with Apple this week.

Why it matters: It’s hard to say exactly what Apple is cooking up with all these advancements, but Tim Cook did tease a huge AI announcement earlier this year. Is Siri in for some major AI upgrades soon? I think so! Then again, that’s just a guess – so I’ll keep you posted.

Elon Musk, having recently sued OpenAI over claims that they are too closed, recently said he would open-source xAI’s model, Grok-1.

Well, Elon kept his word and has now released the weights and architecture of the Grok-1 model the Apache 2.0 license (meaning it’s free for commercial use). It’s a 314 billion parameter model trained from scratch by xAI and not fine-tuned to any particular task. The license does not give access to X for real-time data.

In terms of performance Grok-1 lags behind the biggest open-source models, like Mistral and Falcon, but the size of the model means it can improve a lot with continued training.

Why it matters: Developers can now use Grok-1 for free in their own projects – it doesn’t come with training code but still, it has less usage restrictions than most other open models. Perplexity has already said they will fine-tune it for conversations and make it available on their Pro plan.

a16z released another of their brilliant analyses, this time their semi-annual top 100 Gen AI consumer apps.

Key takeaways from the report:

Highly rated, new GPTs featured in OpenAI’s official GPT store (from the last week)

I’m collaborating with writers to create easy tutorials and reviews of GPTs to achieve different tasks. Here’s what we have this week:

We’re still working on improving the content and format of these articles. Any specific topic you’d like to see covered by us? Leave your feedback in the poll at the end of this email.

Converts text to speech, max 4096 chars, 6 voices

❞ Far out! A simple and fast GPT that does what it says, and does it well. You give it a piece of text and it gives you six voices to choose from; you select a voice and it quickly generates a lifelike voiceover from your text.

Want your GPT (honestly) reviewed here? Submit it to whatplugin and get featured in the newsletter.

what i’m reading right now

simple.ai 🕵🏻‍♂️ - deep-dives on AI by Hubspot’s co-founder (highly recommended)

The Neuron 😸 - easy weekday read on AI’s latest developments

Bagel Bots 🥯 - best hands-on tips & tricks

AI Minds 🧠 - semi-technical breakdown of trending AI topics

*if you sign up for these newsletters I may earn a commission at no cost to you. If you do, THANK YOU, it will make it possible for me to continue to do this.

Reach 10k professionals leveraging AI by sponsoring this newsletter.