OpenAI's Big API Update 🤖

And also: AI learnings algorithms, Salesforce and Snowflake Q4 reports, YouTube new priorities..

March 02, 2023

Update 2023-03-02

  • Neil Mohan Outlines YouTube Priorities, Teases Upcoming Features

  • Microsoft Teams Gets Performance Upgrade with New Version

  • Salesforce Reports Q4 Revenue and EPS Growth Above Expectations

  • Snowflake Reports Strong Q4 Earnings and Authorizes $2 Billion Stock Repurchase Program

  • Multicloud Strategies Adopted by 98% of Enterprises Using Public Cloud: Study

  • ChatGPT and Whisper APIs are Now Available, Priced at a 90% Discount and $0.006/Minute, Respectively

  • Scientists Find That AI Models Invent Learning Algorithms

  • Reducing Generative AI System Costs and Sustainability Issues Through Sample Pipelining and Coprocessor Technology

🐘

Big Tech and Startups

Neil Mohan, the new head of YouTube, outlined his key priorities and teased the upcoming features for the media platform in his first address. He highlighted progress with paid channels and the NFL Sunday Ticket rights deal, which will move to YouTube TV and become a stand-alone offering available through YouTube.

He also mentioned that generative AI tools are being developed to help content creators with storytelling and raising production values, as well as a creation feature similar to TikTok’s Duet feature. In addition, he noted that YouTube Shorts had seen a dramatic rise in daily views, and podcasting and AI tools are being introduced to the platform.

Microsoft is launching a new version of Microsoft Teams next month to improve system resource usage on PCs and laptops.

  • It will use 50% less memory, tax the CPU less, and result in better laptop battery life.

  • The app has been built from the ground up, moving away from Electron and to Microsoft's Edge Webview2 technology & React Javascript library.

  • It should open faster, feel more responsive, and address criticisms and complaints about Teams performance.

  • A preview of the new app will be available in late March with a toggle to switch back to the existing app.

☁️

SaaS and Cloud

Salesforce reported Q4 results that topped expectations, sending shares up. Subscription and support revenues for the quarter were $7.79B, a 14% increase year-over-year.

Professional services and other revenues for the quarter were $0.60 billion, an increase of 19% year-over-year. The operating margin was 4.3%, after a 2.4% seen decline seen in the prior-year period.

Guidance projected first-quarter EPS of $1.60 to $1.61 on revenue in the range of $8.16B to $8.18B; full-year 2024 guidance on EPS was estimated at $7.12 to $7.14 on revenue between $34.5B to 34$.7B. Activist investors have flagged concerns about growth and margins as well as a desire for increased share buybacks.

📝 Cloudly quick take: "It was not as bad as we thought" is how we see the reaction from the street. Why does this matter? Thanks to its dominant position and vast customer base, Salesforce is a proxy for the whole SaaS market. So that they see continued growth and give guidance above expectations is positive for the sentiment.

Snowflake reported better-than-expected results in the Q4 FY2023 quarter, with revenue up 53.5% yearly to $589 million. EPS (non-GAAP) was $0.14 vs. analyst estimates of $0.05 ($0.09 beat), Product Revenue Guidance for Q1 2024 is $570.5 million at the midpoint.

Free Cash Flow of $205.3 million, up from $65 million in the previous quarter. Net Revenue Retention Rate: 158%, down from 165% the previous quarter. Customers increased to 7,828, up from 7,292 in the previous quarter; Gross Margin (GAAP): 65.1%, in line with the same quarter last year.

From the earnings transcript

  • Streamlit in Snowflake entered private preview status, and SnowConvert's proprietary conversion tools enable migration from legacy platforms.

  • Michael Scarpelli (CFO) reported an expansion of the partnership with AWS over the next five years to $2.5 billion, with guidance for product revenues between $568 million and $573 million for Q1 FY 2024, representing 44%-45% year-on-year growth.

  • The Board of Directors also authorized a stock repurchase program of up to $2 billion over two years while expecting to add 1,000 employees this fiscal year.

98% of enterprises using public cloud adopt multicloud infrastructure strategy, according to a study commissioned by Oracle Cloud Infrastructure. The study collected information from 1,500 respondents at enterprises, finding that almost every cloud journey is becoming a multicloud journey.

Data sovereignty and cost optimization are the top drivers of multicloud strategies. IT departments are proactively planning a multicloud strategy for the future, including data redundancy, mobility, risk mitigation, and global service delivery.

🤖

AI

ChatGPT and Whisper APIs are now available for developers, giving access to advanced language processing and speech-to-text capabilities.

  • The cost of ChatGPT has been reduced by 90%, while Whisper is priced at $0.006/minute.

  • Dedicated instances are available for deep control over the model version and system performance.

  • Data submitted through the API will not be used for service improvement unless opted in.

  • Terms of Service have been improved to better meet user needs, with data ownership terms that allow users to own both input and output from models.

  • The engineering team is prioritizing stability and uptime improvements in order to benefit all humanity through AI-powered applications.

Scientists have identified a phenomenon called in-context learning, where large language models such as OpenAI's GPT-3 and Google's LaMDA can perform tasks without retraining.

  • MIT, Stanford, and Google researchers found that the model could build upon existing knowledge rather than just memorizing data.

  • Their experiments yielded that AI models may create smaller models inside themselves to learn tasks from prompts they have not seen before.

  • This study suggests that AI models can invent learning algorithms and opens up possibilities for future research into better training methods.

  • While researchers are making progress on uncovering how AI works, ethical concerns remain around how these systems reproduce biases.

Generative AI systems, such as ChatGPT, are becoming increasingly popular; however, this technology is resource intensive and is causing increased power bills, infrastructure costs, and carbon footprints.

Cloud computing resources are the backbone of AI technology development and innovation; however, this popularity will lead to increased demand for these resources and a resulting surge in costs and sustainability issues.

To reduce the cost and sustainability issues of generative AI systems, research into sample pipelining is needed to reduce the data processing and the use of coprocessors or quantum computing for faster processing without reducing accuracy.

There must be a careful balance between cost, risk, and value when implementing AI systems, with an understanding of the potential impact on cost and sustainability.

Reply

or to participate.