Next Week in The Sequence

• Edge 287: We start our longest and most ambitious series yet. We are going to be exploring emerging techniques in foundation models.

• Edge 288: We deep dive into Microsoft’s DeepSpeed-Chat, the new open source framework for creating ChatGPT like models using RLHF.

📝 Editorial: The Generative AI Cyber Security Week

The RSA conference is widely regarded as one of the most relevant cybersecurity events of the year. Last week, cybersecurity technologists from all over the world descended on San Francisco to discuss the latest developments in cybersecurity technology. This year, generative AI was front and center of the announcements, with major cybersecurity products incorporating generative AI capabilities and new platforms being announced.

 The impact of generative AI in the cybersecurity landscape should not come as a surprise if we consider that many traditional security workflows, such as threat detection and response, involve text analysis capabilities that can be drastically improved by LLMs. Among the biggest announcements at the RSA conference, we have Sec-PaLM, a new cybersecurity specialized LLM announced by Google. IBM also unveiled a new suite of generative AI-based cybersecurity products. SecurityScoreCard launched a ChatGPT power search engine for its security ratings platform, and Microsoft continues to incorporate generative AI features in areas such as API security, app governance, and Azure AD.

Cybersecurity has quickly become one of the most active markets for applying generative AI.

🔎 ML Research

LayerNAS

Google Research published a paper detailing LayerNAS, a new type of neural architecture search(NAS) algorithm. LayerNAS reduces the complexity NAS models with combinatorial optimization —> Read more.

Low-Code LLM

 Microsoft Research published a paper outlining Low-code LLM, a technique for optimizing human-LLM interaction for workflow tasks. Low-code LLM improves upon traditional human-LLM approaches by using two LLMs for planning and execution respectively with a human in the loop —> Read more.

Chameleon

Researchers from the University of California, Los Angeles and Microsoft Research published a paper discussing Chameleon, a framework for compositional reasoning with LLMs. Chameleon augments LLMs with information retrieval mechanisms based on search engines, Python functions, rule-based systems and several others —> Read more.

 DiffFit

Huaweii Research published a paper discussing DiffFit, a fine-tuning approach for diffusion models. DiffFit combines different techniques for fine-tuning parameter optimization in large diffusion models —> Read more.

🤖 Cool AI Tech Releases

HuggingChat

Hugging Face unveiled HuggingChat, an open source chatbot that exhibits ChatGPT like capabilities —> Read more.

Ray 2.4.0

AnyScale announced the release of a new version of the Ray platform optimized for LLM lifecycle management —> Read more.

NeMo Guardrails

NVIDIA open sourced NeMo Guardrails, a framework for building LLMs aligned with enterprise security practices —> Read more.

Phoenix

AI startup Arize open sourced Phoenix, a new platform for LLM evaluation —> Read more.

🛠 Real World ML 

Yelp Content Embeddings

The Yelp engineering team discusses the best practices for generating multi-dimensional embedding representations of content for different ML tasks —> Read more.  

Anomaly Detection at Lyft Part II

Lyft’s engineering team publishes the second of an article describing the process of building anomaly detection models in the LyftLearn platform —> Read more.

📡AI Radar

• Google announced Sec-PaLM, a security first foundation model.

• Vector database platform Pinecone announced a $100 million series B.

• Yelp announced new search features powered by LLMs.

• Microsoft unveiled new AI-powered features in its Designer app.

• Replit, the CoPilot competitor, raised $97.4 million to accelerate the growth of its AI-based IDE platform.

• AI app development platform AirOps announced a $7 million seed round.

• Gitlab announced a new batch of AI-based security features.  

• SecurityScorecard unveiled new search features for its security ratings platform powered by GPT-4.

• IBM announced QRadar, an AI-powered security platform.

• Microsoft made a series of announcements of AI-powered security capabilities at the RSA conference.

• AI powerhouses Databricks and Hugging Face announced a collaboration to streamline dataset processing for AI models.