For such a tiny model (it can run on your phone!) Phi 3 mini 3.8B seems to do a pretty nice job on real world reasoning and planning tasks.

And Phi-3 small model is at Hugging Face right now - huggingface.co/microsoft/Phi-…

Phi3-small, our new open model, at 7B parameters rivals GPT3.5, proving the quality of training data matters more than pure size: arxiv.org/pdf/2404.14219…

The Neuron is a great newsletter for the latest in AI. I’m excited about the podcast:

I'm excited to launch The Neuron's podcast! And give away $500 to a lucky winner :)

After shipping 300+ newsletters to 400K+ readers, we're going deeper with a pod.

TLDR: It’s still about AI, but we’ll dive deeper and cover more than we normally do in the newsletter: extra

phi-3 is here, and it's ... good :-).

I made a quick short demo to give you a feel of what phi-3-mini (3.8B) can do. Stay tuned for the open weights release and more announcements tomorrow morning!

(And ofc this wouldn't be complete without the usual table of benchmarks!)

They should have consulted with me. I could have given them a much more realistic bluescreen: swoosh.nike.

Microsoft launched the best course on Generative AI!

The free 18 lesson course is available on Github and will teach you everything you need to know to start building Generative AI applications.

I'm thrilled to announce that Radius, the cloud agnostic application platform, has been approved by CNCF as a Sanbox project! blog.radapp.io/posts/2024/04/…

Big day for unexpectedly powerful LLM releases.

Microsoft's open source WizardLM 2 (also note that it used synthetic inputs in training, maybe 'running out of data' will not be a big deal): wizardlm.github.io/WizardLM2/

Closed source Reka, which is multimodal: reka.ai

On April 16, 2024 see all the latest and greatest in #AzureCosmosDB ! You won't want to miss the biggest Azure Cosmos DB event of the year.

Register now for #AzureCosmosDB Conf: aka.ms/AzureCosmosDBC…

I never imagined getting served a Kubernetes ad while watching news on TV:

In this post, I share examples of how we approach AI security by researching attacks and the defenses for them:

As generative AI continues to integrate into our daily lives, we continue our commitment to advance safe, secure, and trustworthy AI. Here are the steps we're taking to mitigate risks: msft.it/6016cA9zI

Aspire Institute Also kudos to Microsoft for providing the only free GPT-4 level model available worldwide. Many people could not afford access to a frontier model otherwise. I hope they keep up that commitment.

Today we announced our plans to deepen our investments in Japan, spanning cloud and AI infrastructure, skilling, research, and cybersecurity, as we continue partnering to accelerate the country's AI transformation.

Google's teams coding in Rust are more than twice as productive as teams that code in C/C++: theregister.com/2024/03/31/rus…

As part of our ongoing work on AI safety and security, we've discovered a powerful, yet simple LLM jailbreak that exploits an intrinsic LLM behavior we call 'crescendo' and have demonstrated it on dozens of tasks across major LLM models and services: …ndo-the-multiturn-jailbreak.github.io

Our team in Cambridge is looking to recruit two postdoc researchers to work in the space of Security/Privacy/Systems/PL. It's a great team to work in with a massive potential for impact. You also get Mark Russinovich as your (skip-)skip-level manager as an added bonus. (1/4)

This is such an exciting milestone on our path toward unlocking the scientific and commercial progress that will come from reliable quantum computation.