Announcing Magic's $23 million Series A to build an AI software engineer: magic.dev/blog/series-a CapitalG (Alphabet), Nat Friedman (fmr CEO GitHub), Elad Gil, Amplify, many others. We're hiring! magic.dev/join

Speaker: Sebastian De Ro Sebastian De Ro ist Mitbegründer und CTO von Magic (magic.dev). Sicher dir jetzt deinen Platz für unser Digital Event am 25.Mai 2023: tuw.media/event/future-t…

Meet LTM-1: LLM with *5,000,000 prompt tokens* That's ~500k lines of code or ~5k files, enough to fully cover most repositories. LTM-1 is a prototype of a neural network architecture we designed for giant context windows.

Watch LTM-1 generate complex suggestions:

Watch LTM-1 reuse and synthesize information across files:

$100/GiB dataset bounty! We Magic train LLMs on trillions of text tokens. We want even more tokens. Introducing the data bounty: up to $1M in payouts, $100/GiB: magic.dev/blog/data-boun… If you submit good data, we'd also love to hire you for our data team.

We've raised $117M from Nat Friedman and others to build an AI software engineer. Code generation is both a product and a path to AGI, requiring new algorithms, lots of CUDA, frontier-scale training, RL, and a new UI. We are hiring!

Big congrats to Eric Steinberger and Magic on $117M fundraise to build code-gen models 🔥 Magic.dev

Magic.dev has trained a groundbreaking model with many millions of tokens of context that performed far better in our evals than anything we've tried before. They're using it to build an advanced AI programmer that can reason over your entire codebase and the

Very excited to welcome Andrej Karpathy as Magic's latest investor!

Looking for an intern to help with a microplastics research project this summer in Silicon Valley. Apply here: forms.gle/i17uYcixVwuzGi…

LTM-2-Mini is our first model with a 100 million token context window. That’s 10 million lines of code, or 750 novels. Full blog: magic.dev/blog/100m-toke… Evals, efficiency, and more ↓

We want to build an AI model that can design, code and secure the next version of itself. This is obviously really hard. I thought it’d take <=2y lol. It’s been 2y now. It will take >2y. But I still believe it’ll happen!

We are 23 people, 8000 H100s, $465M raised. Thanks to our partnership with GCP, we can scale up to tens of thousands of GB200s over time. If you like really hard problems, please come join us! magic.dev/blog/100m-toke…

I think the HashHop long context eval is really cool. Curious what people make of it!

Update!

Excited to partner with Google Cloud as we scale up on H100s and GB200s.

This blog post by Magic does a great job highlighting the weaknesses of popular long-context evals and introduces HashHop as an alternative. Very impressive work from the Magic team and congrats on the new funding!

Sharing a brief write-up of our safety priorities: magic.dev/safety I’m also endorsing SB1047 – regulation is necessary to shape the race track we are all on. SB1047 is only part of what's needed, but it is a good start. I hope Gavin Newsom signs it into law.

Our AGI Readiness Policy (magic.dev/agi-readiness-…) was developed in collaboration with METR and we've started work on v2, focusing on safety testing models capable of adaptive inference-time compute utilization.