Next Week in The Sequence:

• Edge 339: Our series about fine-tuning continues with the super cool topic of prefix-tuning. We review the original prefix-tuning paper and Hugging Face’s PEFT framework.

• Edge 340: We dice into Microsoft’s AutoGen framework for autonomous agents.

You can subscribe below:

📝 Editorial: Generative AI is the New Wall Street Earnings Kingmaker, and Microsoft is the New Earnings King

It's that week of the quarter when the big three in cloud computing: Microsoft, Amazon, and Google, all report earnings. And what a week it was! Generative AI must have been the most mentioned term in earnings calls this week, signaling the importance that Wall Street places on this as a growth factor for tech incumbents. To put it in context, Microsoft mentioned AI 66 times, Google 79 times, and Amazon 32 times.For this quarter (which is likely to change many times over), Microsoft was crowned the undisputed king of cloud AI. It's fascinating how a single factor can so impactfully drive investor sentiment across three companies, each with trillions of dollars in market cap.

Here are a few highlights: • Microsoft posted a 29% growth in Azure, compared to Amazon’s 12.3% and Google’s 22.5%. • Microsoft struck an optimistic tone about growth driven by interest in generative AI, highlighting their clear strategy based on a partnership with OpenAI. • Amazon also showed demand for generative AI but offered a more diverse range based on different products, including support for Anthropic’s Claude. For Wall Street, quite often, a simpler message is better. • Google adopted a more cautionary tone regarding generative AI demand, noting that many clients were more cautious about spending. Sales fell short of Wall Street estimates.

These results were reflected in post-earnings trading, with Microsoft and Amazon up and Google experiencing its most significant decline since March 2020. Intel, Meta, and Snap were among other companies whose earnings reports were deeply influenced by their AI narratives.

It's rare to see a single factor being so influential across the earnings reports of many large companies, especially when discussing a nascent trend. For almost a decade, the cloud has been the dominant element in the earnings reports of the big three, but it was mostly related to sales. Now, that position is clearly occupied by generative AI, but the narrative isn't about sales; it's about potential. Wall Street seems to recognize that AI doesn't evolve at the same pace as traditional software methodologies and that the winners will unlock value on a scale we've never seen before. Generative AI is the new earnings kingmaker."

📌 Webinar: Fine-Tune and Serve 100s of LLMs for the Cost of One with LoRAX

Date: November 7th, 10:00 am PST / 7:00 pm CET

We are excited to support an upcoming webinar with Predibase where we’ll learn why smaller, faster, fine-tuned LLMs are emerging as more efficient, performant, and cost-effective alternatives to large, general AI models. In this webinar, co-founders Dev and Travis will demonstrate how Predibase’s modular LLM serving architecture – LoRAX – enables teams to dynamically serve 100s of fine-tuned LLMs for the cost of one.

You’ll walk away with an actionable understanding of how to significantly reduce costs and improve performance and efficiency of LLMs in production, all while retaining full ownership of your models and data.

Reserve your spot today!

🔎 ML Research

Eureka

NVIDIA published a paper detailing Eureka, an LLM-based technique used to design reward design algorithms that can be used in complex reinforcement learning algorithms. NVIDIA showcased Eureka applied to complex embodied AI tasks —> Read more.

Table GPT

Microsoft Research published a paper detailing Table GPT, a fine-tuned LLM optimized for different table tasks. The research shows how very simple fine-tuning can lead to outperform ChatGPT in table related tasks such as interactions with Excel spreadsheets —> Read more.

SAT Solving

Amazon Science published a paper detailing a method for generating checkable proofs for incremental SAT problems. SATs are considered the original NP-complete problem and is one of the most important expressions of reasoning in AI —> Read more.

Dataset Provenance

Researchers from MIT, Cohere for AI and other AI labs published a paper detailing a data provenance technique used for attributions in training and fine-tuning datasets. The researchers also released the Data Provenance Explorer that allow researchers to filter thousands of datasets —> Read more.

RAFA

Researchers from Northwestern University published a paper and open source code for “reason for future, act for now" (RAFA), a frameowrk for autonomous LLM agents. The framework explores cutting edge capabilities of autonomous agents such as long term reasoning —> Read more.

CodeChain

Salesforce Research published a paper introducing CodeChain, a code language model that provides a more modular approach than alternatives. CodeChain relies on sub-modulesfor code revisions using a variation of chain of thought prompting —> Read more.

🤖 Cool AI Tech Releases

Habitat 3.0

Meta AI open sourced Habitat 3.0, the newest version of its framework for building embodied AI agents —> Read more.

🛠 Real World ML

Einstein Model Operationalization

Salesforce discusses the processes to operationalize and scale models in the Einstein platform —> Read more.

English Speaking in Google Search

Google details the techniques used to enable users to practice speaking English directly in Google Search —> Read more.

📡AI Radar

• Microsoft beat Wall Street estimates by a wide marging with strong AI demand forecasts.

• Google set to invest $2 billion in Anthropic.

• OpenAI, Anthropic, Google, and Microsoft published an update about their Frontier Model Forum effort for AI safety.

• Google’s earnings were strong but casted doubts over the search giant monetization plans for generative AI.

• CentML raised $27 million for its AI model optimization platform.

• Amazon delivered strong quaterly earnings boosted by generative AI and cloud computing.

• MLOps startup ZenML announced a $3.7 million extension of their seed round.

• Amazon launched new generative AI capabilities to generate images for advertisers.

• AI data center optimization platform MangoBoost raised $55 million in new funding.

• AI-video platform Oxolo raised a €13 million Series A.

• AI-AR Wearable company Brilliant Labs raised $3 million for its AI glasses.

• Autonomous vehicle startup Pony.ai secured $100 million in new funding.

• AI healthcare company Heidi Health raised a $10 million Series A.

• Arteria AI, a platform for applying AI to financial documentation processes, announced a $30 million Series B.

• Credal.ai raised $4.8 million to enable secure pipes for connecting enterprise datasets to LLMs.

• AI-based micro-influencer platform 1337 emerged from stealth mode and announced a $4 million round.