Another Monster Generative AI Week
Sundays, The Sequence Scope brings a summary of the most important research papers, technology releases and VC funding deals in the artificial intelligence space.
Next Week in The Sequence
Edge 275 continues with our series about federated learning(FL) with an overview of vertical FL, the research behind Google’s FL model to predict text selections in Android and the Flower framework.
Edge 277 discusses the research behind Meta AI’s Llama, a large language model that outperforms ChatGPT on several benchmarks.
📝 Editorial
No, I am not repeating last week’s editorial but is just that the momentum around generative AI makes it hard to write about something else. If we thought the previous week was impressive in terms of generative AI releases nothing compares with the last few days. The biggest headline of the week was the release of GPT-4, which now works in multimodal scenarios. The release is available via ChatGPT Plus and requires a waitlist for API access. One of the most relevant OpenAI competitors, Anthropic, unveiled the first release of Claude, its ChatGPT competitor. Is that wasn’t enough, Google announced the release of its highly anticipated PaLM API to interact with its LLM.
Productivity is becoming one of the most important aeras of LLMs. Together with the PaLM API, Google also released MakerSuite, a tool that allow developers to quickly prototype generative AI apps. The next day after this announcement Microsoft hosted and event and unveiled 365 Copilot, an integration of LLM capabilities into the Office 365 suite. LinkedIn also released a generative AI assistant for recruitment ads and writing profiles.
The pace of innovation in generative AI is like nothing we have seen before.
On a non-generative AI news, PyTorch reached its 2.0 release with sizable improvements in its text, computer vision and audio libraries.
🔎 ML Research
Audio-Video Speech Translation Benchmark
Meta AI published a paper and open source code for MuAViC, one of the first audio-video speech translation benchmarks for deep learning models —> Read more.
PALM-E
Google published the research around PALM-E, a model that adapts PALM for robotics scenarios —> Read more.
Alpaca
Stanford University published a paper detailing Alpaca, a fine0tuned version of Meta AI’s Llama optimized in 52,000 instructions —> Read more.
Petals
Researchers from Yandex and Hugging Face published a paper detailing Petals, a framework that can run large language models in a decentralized, distributed network —> Read more.
📌 EVENT: Learn how you can simplify & accelerate feature engineering
Tecton’s latest release, version 0.6, is now available, featuring new capabilities such as notebook-driven development, continuous mode for all stream feature views, new aggregation functions, Stream Ingest API, and ACL CLI commands. These new features make it easier for data teams to shift notebook development into production and approach feature engineering more efficiently. Register today to join us for our product webinar to learn how to shift notebook development into production and learn more about our new feature platform capabilities!
🤖 Cool AI Tech Releases
GPT-4
OpenAI released its highly anticipated version of GPT-4 with broader knowledge and multi-modal capabilities —> Read more.
Claude
Anthropic released the first version of Clause, its highly anticipated ChatGPT competitor —> Read more.
Palm API
Google announced PALM API, a programmable interface to interact with LLMs and MakerSuite, a tool that allow developers to quickly build generative AI powered apps —> Read more.
PyTorch 2.0
PyTorch released its 2.0 version with improvements in domain libraries for text, computer vision and audio as well as APIs for building ecosystem projects —> Read more.
CoPilot 365
Microsoft unveiled CoPilot 365, a platform that integrates LLMs into productivity tools —> Read more.
OpenChatKit
A collaboration between Together, LAION and Ontocord resulted in the open source version of OpenChatKit, a framework to create specialized and general purpose chatbots —> Read more.
LinkedIn Assistant
LinkedIn unveiled an AI assistant for writing profiles and recruitment ads à—> Read more.
Generative AI VMs
Microsoft previews a new generation of Azure VMs optimized for generative AI workloads —> Read more.
Time Series Transformer
Hugging Face unveiled Informer, a probabilistic time series transformer —> Read more.
🛠 Real World ML
Data Governance at Vodafone
Vodafone discusses how they use TensorFlo data validation to improve their data governance processes —> Read more.
📡AI Radar
Adept raised a $300 million series B for its generative AI platform.
AI auto insurance platform Fairmatic raised $46 million.
AI automation platform Cast AI raised $20 million to help companies reduce cloud spend.
MLOps platform Seldon announced a $20 million series B funding.
Presight AI Holding Plc, an AI firm based on Abu Dhabi, drew tremendous traction for its $496 million IPO.
I’ve been following the newsletter for a long time and it’s really an essential part of my learning now.
Just want to bring my attention of my work this week in case it can be a good fit for the newsletter: https://github.com/ethanyanjiali/minChatGPT