Microsoft Released Phi-2, a SLM That Beats LIama 2
Plus: Anthropic launches Claude for Google Sheets, Channel 1's AI news anchors.
Hello Engineering Leaders and AI Enthusiasts!
Welcome to the 167th edition of The AI Edge newsletter. This edition brings you “Microsoft’s Phi-2, a small language model that beats the Llama 2.”
And a huge shoutout to our incredible readers. We appreciate you😊
In today’s edition:
🎉 Microsoft released Phi-2, a SLM that beats the Llama 2
👏👏🔢Anthropic has Integrated Claude with Google Sheets
📰 Channel 1 launches AI news anchors with superhuman abilities
📚 Knowledge Nugget: Guide to fine-tune your own general purpose Stable Diffusion models [Part 1] by
Let’s go!
We need your help!
We are working on a Gen AI survey and would love your input.
It takes just 2 minutes.
The survey insights will help us both.
And hey, you might also win a $100 Amazon gift card!
Every response counts. Thanks in advance!
Microsoft released Phi-2, a SLM that beats the Llama 2
Microsoft released Phi-2, a small language model AI with 2.7 billion parameters that outperforms Google's Gemini Nano 2 & LIama 2. Phi-2 is small enough to run on a laptop or mobile device and delivers less toxicity and bias in its responses compared to other models.
It was also able to correctly answer complex physics problems and correct students' mistakes, similar to Google's Gemini Ultra model.
Here is the comparison between Phi-2 and Gemini Nano 2 Models on Gemini’s reported benchmarks. However, Phi-2 is currently only licensed for research purposes and cannot be used for commercial purposes.
Why does this matter?
Microsoft's Phi-2 proved that victory doesn't always belong to the biggest models. Even though it is compact in size, Phi-2 can outperform much larger models on important tasks like interpretability and fine-tuning. Its combination of efficiency and capabilities makes it ideal for researchers to experiment with easily. Phi-2 showcases good reasoning and language understanding, particularly in math and calculations.
Anthropic has Integrated Claude with Google Sheets
Anthropic launches a new prompt engineering tool that makes Claude accessible via spreadsheets. This allows API users to test and refine prompts within their regular workflows and spreadsheets, facilitating easy collaboration with colleagues.
(This allows you to execute interactions with Claude directly in cells.)
Everything you need to know and how to get started with it.
Why does this matter?
Refining Claude's capabilities through specialization empowers domain experts rather than replacing them. The tool's collaborative nature also unlocks Claude's potential at scale. Partners can curate prompts within actual projects and then implement them across entire workflows via API.
Channel 1 launches AI news anchors with superhuman abilities
Channel 1 will use AI-generated news anchors that have superhuman abilities. These photorealistic anchors can speak any language and even attempt humor.
They will curate personalized news stories based on individual interests, using AI to translate and analyze data. The AI can also create footage of events that were not captured by cameras.
While human reporters will still be needed for on-the-ground coverage, this AI-powered news network will provide personalized, up-to-the-minute updates and information.
Why does this matter?
It's a quantum leap in broadcast technology. However, the true impact depends on the ethics behind these automated systems. As pioneers like Channel 1 shape the landscape, they must also establish its guiding principles. AI-powered news must put integrity first to earn public trust and benefit.
Enjoying the daily updates?
Refer your pals to subscribe to our daily newsletter and get exclusive access to 400+ game-changing AI tools.
When you use the referral link above or the “Share” button on any post, you'll get the credit for any new subscribers. All you need to do is send the link via text or email or share it on social media with friends.
Knowledge Nugget: Guide to fine-tuning your own general purpose Stable Diffusion models [Part 1]
This article by
provides a detailed step-by-step guide on how the authors fine-tuned a Stable Diffusion model called Vodka using a dataset of Midjourney AI-generated images. Their goal was to quickly assemble an initial general-purpose model to test whether this approach of "distilling" synthetic images could produce a high-quality model.They explain their methodology for compiling the Midjourney dataset, cleaning and preparing the data, setting up the training parameters, running the fine-tuning, and testing the results after 25 and 50 epochs.
Why does this matter?
While the model has some flaws, like struggling with photorealism, it can generate creative and fun images on par with some top models. This is meant as a starting point for iteration and improvement, with many ideas proposed for enhancing the model further. The authors release the model and intend to continue refining it based on feedback and usage.
What Else Is Happening❗
🤖 Tesla reveals its next-gen humanoid robot, Optimus Gen 2
It is designed to take over repetitive tasks from humans. The robot allows it to walk 30% faster and improve its balance. It also has brand-new hands that are strong enough to support significant weights and precise enough to handle delicate objects. Tesla plans to use the robot in its manufacturing operations and sell it. (Link)
🦊 Mozilla launches MemoryCache, An on-device, personal model with local files
MemoryCache includes a Firefox extension for saving pages and notes, a shell script for monitoring changes in the saved files, and code for updating the Firefox SaveAsPDF API. The project is currently being tested on a gaming PC with an Intel i7-8700 processor using the privateGPT model. (Link)
🕶️ Meta rolling out multimodal AI features in the Ray-Ban smart glasses
The glasses' virtual assistant can identify objects and translate languages, and users can summon it by saying, "Hey, Meta." The AI assistant can also translate text, show image captions, and describe objects accurately. The test period will be limited to a small number of people in the US. (Link)
👻 Snapchat+ subscribers can now create & send AI images based on text prompts
The new feature allows users to choose from a selection of prompts or type in their own, and the app will generate an image accordingly. Subscribers can also use the Dream Selfie feature with friends, creating fantastical images of themselves in different scenarios. Additionally, subscribers can access a new AI-powered extend tool that fills in the background of zoomed-in images. (Link)
🧠 A New System reads minds using a sensor-filled helmet and AI
Scientists have developed a system that can translate a person's thoughts into written words using a sensor-filled helmet and AI. It records the brain's electrical activity through the scalp and converts it into text using an AI model called DeWave. Its accuracy is 40%, and recent data shows an improved accuracy of over 60%. (Link)
That's all for now!
If you are new to The AI Edge newsletter, subscribe to get daily AI updates and news directly sent to your inbox for free!
Thanks for reading, and see you tomorrow. 😊