- AI Emergence
- Posts
- Microsoft is 'Build'ing AI and it is enabling you too!
Microsoft is 'Build'ing AI and it is enabling you too!
Why Copilot+ PCs are the talk of the town?
Hey there,
When most of the industry was talking ChatGPT 4o vs. Gemini, Microsoft did a Microsoft! They launched multiple exciting products (silently, relatively speaking) at their campus.
The launches include new Hardware (Copilot+ PCs), new Software (new Phi 3), and multiple new products in their Tech Stack!
Google also launched Gemini for Gmail, allowing you to ask questions directly from your inbox, like “Where is the design pending on X project,” and it will summarize relevant conversations.
It’s an exciting week!
Have you ever wondered how powerful AI tools like ChatGPT-4, Claude, and Gemini function? If you're interested in understanding the workings of LLM, join us at DataHack Summit 2024 for the workshop "Mastering LLMs: Training Finetuning and Best Practices" led by Sandeep Singh.
Let’s dive into this week’s news!
What would be the format? Every week, we will break the newsletter into the following sections:
The Input - All about recent developments in AI
The Tools - Interesting finds and launches
The Algorithm - Resources for learning
The Output - Our reflection
Question to ponder before we meet next!
Table of Contents
If I told you 10 days back that I am going to buy a laptop with 14-core CPUs, a few cores can clock 4 GHz and others 3.4 GHz. It comes with NPU to enhance AI tasks and has an all-day battery life. The only thing fitting that description was Apple Silicon-based MacBooks! Not anymore…
Recently, Microsoft announced the launch of Copilot+ PCs, a new category of AI-powered Windows PCs, at a special event on their campus.
Some key points about the device -
NPUs to enhance AI performance
40+ trillion operations per second,
Providing all-day battery life and
The ability to leverage advanced AI models directly on the device.
Additionally, some of the notable features are -
Recall: Helps users find and remember information on their PCs.
Cocreator: AI-powered image creation and editing in real-time.
Live Captions: Translates audio from 40+ languages into English.
While the real-life use cases and reviews are still to come out, this looks like a Mac with an App like Rewind and ChatGPT plus a desktop app. I also think that PCs would be less taxing on the pockets.
This is going to heat up the space - what do you think? Can Microsoft finally get a meaningful market share for Surface products? Only time will tell…(source)
Microsoft has expanded its Phi-3 family of models by introducing new offerings, including the Phi-3-vision, a multimodal model that combines language and vision capabilities, available for use on Microsoft Azure.
Alongside the previously released Phi-3-small and Phi-3-medium, and the Phi-3-mini, these models are designed for generative AI applications requiring efficient reasoning with limited compute resources.
Phi-3 models stand out for their robust performance across various benchmarks, including language, reasoning, and math tasks, making them both cost-effective and powerful (source).
Scarlett Johansson has accused OpenAI of using her voice without consent for its new AI voice system, "Sky." She claims that she had declined the offer from OpenAI CEO Sam Altman last year to voice the ChatGPT 4.0 system.
Johansson’s statement expressed her “shock, anger, and disbelief” at OpenAI’s demo, which used a voice so similar to hers that even her closest friends and news outlets couldn’t distinguish it.
OpenAI has paused the use of “Sky” and clarified in a statement that the voice used was not an imitation of Johansson but that of another professional actress. (source)
This is one of the rare occasions when OpenAI is struggling with defining the public narrative. They first said that the voice is of “another professional actress”, alongside “We believe that AI voices should not deliberately mimic a celebrity's distinctive voice”. Sam Altman also tried to defend that ChatGPT 4o is safe for use post this incident.
Reddit shares surged 11% after announcing a partnership with OpenAI, allowing the AI company to train its models on Reddit content via its Data API.
In return, Reddit will offer AI features powered by OpenAI and partner with them for advertising.
CEO Steve Huffman highlighted Reddit's vast archive of authentic human conversations as a valuable resource for AI training. OpenAI CEO Sam Altman, a major shareholder in Reddit, supported the deal, which was led by COO Brad Lightcap and approved by OpenAI's board. (source).
Google recently introduced an AI feature that provides AI-generated answers to user queries, known as AI overviews. Alongside this google will include ads just like it does with regular search results.
This move aims to extend Google's dominance in search advertising into the AI era. The ads will appear in a clearly labeled "sponsored" section and will be sourced from advertisers' existing campaigns. (source)
OpenAI has disbanded its Superalignment team, which was established just a year ago to address the long-term risks associated with artificial intelligence.
This team, tasked with making scientific and technical breakthroughs to control highly advanced AI systems, is being reassigned to other teams within the company.
The decision follows the departures of team leaders Ilya Sutskever and Jan Leike, who cited concerns that the pursuit of new products was overshadowing OpenAI's safety culture and processes. (source)
Sony has recently shared a video where its remote-controlled microsurgery assistance robot capable of performing extremely delicate operations, demonstrated by stitching a corn kernel with tiny suture needles.
This robot, designed for precise procedures like microvascular anastomosis, highlights the potential future of advanced surgery.
The robot features an automatic instrument exchange function, mimicking real-life surgical workflows. (source)
Researchers decode the AI language model's inner mechanisms
Researchers at the AI company Anthropic have made significant progress in understanding the inner workings of LLMs.
By using a technique called "dictionary learning," they uncovered patterns in how different neurons in their model, Claude 3 Sonnet, activate in response to various topics.
These patterns, called "features," help explain why the AI responds in certain ways and can even be manually adjusted to alter the AI's behavior. (source)
Snowflake aims to acquire Reka AI for $1 Billion to enhance AI ecosystem
Snowflake is in talks to acquire AI startup Reka AI for over $1 billion as part of its strategy to build the world's leading AI ecosystem.
This acquisition aims to enhance Snowflake's generative AI capabilities, following its recent investments in other AI companies like Mistral AI and Landing AI.
Reka AI, founded in 2022 by former Google and Meta researchers, specializes in large language models for various applications. Snowflake had already invested in Reka and established a partnership to integrate third-party models into its platform. (source)
Chinese chipmakers CXMT and Wuhan Xinxin advance in HBM semiconductor Development
Two Chinese companies, CXMT and Wuhan Xinxin, are making progress in developing a memory called high bandwidth memory (HBM) essential for advanced AI chips.
Their aim is to reduce their reliance on foreign technology, especially since there are restrictions from the U.S. on exporting advanced tech to China. (source)
Sales is resource-intensive and relies heavily on individual performance. Are LLMs capable of convincing customers to buy a product?
Problem Statement: Develop a sales agent for an AI training program targeting the USA market.
Solution:
Sign up on the platform.
Choose from multiple voices (we chose Scarlet!).
Adjust parameters: “Patience Level,” “Language,” “Custom Vocabulary.”
Set up the prompt: Add the script for the LLM, and specify actions like “transfer to human.”
Test your assistant!
Our Thoughts: This product is valuable for sales-intensive operations, offering a cost-effective alternative to building a native sales team, capable of speaking with a native accent.
For more information about the tool, check out our guide on using GPT-Omni on the Analytics Vidhya blog.
In this video, the CEO of Salesforce, Arundhati Bhattacharya, talks about the power of AI and how businesses can harness it to enhance efficiency, reduce costs, and seize new growth opportunities.
In the recent episode of Leading with Data, I had an amazing conversation with Aamod, Director and Analytics Lead at Meta, about his extensive experience and insights into organizational trends, the impact of generative AI on the field, driving cultural changes through data, and the exciting opportunities in startups and new ventures.
The article "Ways to Think About AGI" by Benedict Evans explores the complexities and uncertainties surrounding artificial general intelligence (AGI). It discusses the unpredictable nature of AGI and the lack of consensus among experts about its risks and behaviors.
Deeplearning.ai launched a course on On-device AI - This could be interesting for developers given that CoPilot+ PCs could become big.
You can probably tell by now - I am kicked about the CoPilot+ PCs and the fact that this could mean meaningful options for people looking to Build! While I love my MacBook - I do feel the serious lack of similar quality alternatives. This should hopefully not be the case shortly.
It should also enable more on-device AI applications and their broader availability. What do you feel about CoPilot+ PCs?
How do you rate this issue of AI Emergence?Would love to hear your thoughts |
Reply