Next Week in The Sequence:

• Edge 327: We start a new series about….drum roll….fine tuning in foundation models! We review Meta AI’s LIMA fine-tuning paper and H2O’s no-code fine-tuning stack.

• Edge 328: WE dee dive into AudioCraft, Meta AI’s new family of generative audio models.

Subscribe Below!

📝 Editorial: NVIDIA, The Most Influential VC In Generative AI

These days, NVIDIA is synonymous with AI computing. The growing dependence of AI workloads on NVIDIA GPUs has transformed the chip manufacturer into one of the most influential tech companies in the world. However, NVIDIA's influence is rapidly expanding beyond computing and into an unexpected area: venture capital.

Last week, AI news was dominated by Databricks' massive $500 million round. Who was at the center of that fundraising? NVIDIA. This news is only surprising if you haven't been paying attention to NVIDIA's recent investment spree. The recent additions to the portfolio feature some of the hottest companies in the AI space. Take a look and judge for yourself:

• Hugging Face: NVIDIA was part of the recent $200 million round that pushed Hugging Face's valuation to $4 billion.

• AI21 Labs: NVIDIA invested in the recent $155 million Series C that valued the OpenAI competitor at $1.4 billion.

• Imbue: NVIDIA joined the $200 million Series B that valued the AI reasoning platform at $1 billion.

• Cohere: NVIDIA was part of the $270 million round that valued Cohere at $2.1 billion.

• Adept AI: NVIDIA participated in the $350 million round that valued Adept at $1 billion.

There are plenty of other AI companies in NVIDIA's portfolio, but the investments in the last few months have been remarkable. NVIDIA's influence in the AI space expands far beyond its computing capabilities, as it strategically invests in some of the biggest innovators in the generative AI space. Quietly and rapidly, NVIDIA has become one of the most influential venture investors in AI. Neither Microsoft nor Google have this level of presence in AI innovation at the moment. Quite amazing.

🔥Building Plaid’s ML Fraud Detection Application—an apply() Fireside Chat

How did Plaid, a leading fintech company, build Signal, a ground-breaking ML fraud detection application that analyzes 1000+ risk factors per ACH transaction, resulting in accelerated transfers and a better user experience?  

Find out in this fireside chat with guest speaker Renault Young, Software Engineer at Plaid, who will share his hands-on experience building the ML infrastructure for Signal. 

SAVE MY SPOT!

🔎 ML Research

LLMs as Optimizers

Researchers from DeepMind published a paper unveiling Optimization by PROmpting (OPRO), a technique to use LLMs as optimizers. Given a specific problem, OPRO generates solutions from the prompt based on previous solutions regularly optimizing the final outcome —> Read more.

Phi-1.5 Technical Report

Microsoft Research published a paper explaining the groundbreaking techniques behind their phi-1.5 model. This is the model that was first published in the “Textbooks is All You Need “ paper —> Read more.

RL in Google Maps

Google Research published a paper proposing a massively scalable inverse reinforcement learning(RL) algorithm called Receding Horizon Inverse Planning (RHIP) used in Google Maps. The paper shows how the combination of RHIP and techniques such as graph compression and parallelization address some of the scalability limitations of RL methods —> Read more.

InstructDiffusion

Microsoft Research published a paper detailing InstructDiffusion, a technique to align computer vision tasks with human instructions. The method learns to predict the pixel space based on specific human instructions —> Read more.

Language Agents Architectures

Researchers from Princeton University published a paper proposing Cognitive Architectures for Language Agents(CoALA), a systhematic framework for creating LLM-based agents. CoALA abstracts capabilities such as reasoning, grounding, learning and decision making under a single framework —> Read more.

🤖 Cool AI Tech Releases

TensorRT-LLM

NVIDIA announced the new TensorRT-LLM compiler that drastically improves inference on NVIDIA H100 GPUs —> Read more.

Stable Audio

Stability AI announced the release of Stable Audio, its model for music and sound generation —> Read more.

Persimmon-8B

Adept AI open sourced Permimmon-8B a relatively small but incredibly high performant LLM —> Read more.

🛠 Real World ML

Ray at Pinterest

Pinterest discusses their adoption of Ray to increase velocity in ML pipelines —> Read more.

📡AI Radar

• Databricks announced a monster $500 million funding round.

• Despite a conservative opening price, Arm delivered a very sucessful IPO.

• Salesforce announced the next version of its Einstein AI platform.

• Essential AI, a secretive company started by two of the authors of the famous transformers paper, is raising $40 million.

• Pixis raised $85 million to expand its AI marketing platform.

• AI defense company Helsing announced a $223 million funding round.

• GitHub for data startup Gable.ai announced a $7 million funding round.

• HiddenLayer raised $50 million for its AI cyber security platform.

• Enterprise conversational AI platform Druid announced a new $30 million funding round.

• EY announced its EY.ai platform backed by a $1.4 billion investment.

• Adobe announced the general availability of its Firefly generative AI suite.

• Treefera raised $2.2 million to improve the carbon credit ecosystems usign AI.

• LastMile AI, a generative AI productivity platform, raised $10 million.