32nd Edition: 3/22/24
Saudi's $40B AI fund, Satya's OpenAI remarks to the MSFT board, the Biden admin's first $ allocated under the CHIPS Act, Nvidia's insane announcements at their GTC 2024 conference, and more....
I’m back from Asia, so my more consistent cadence will resume. Here's this week's AI update, and it’s a meaty one. I hope you enjoy!
PNW AI News:
MSFT CEO Satya Nadella goes viral from his board statement on OpenAI: "If OpenAl disappeared tomorrow, we have all the IP rights and all the capability. We have the people, we have the compute, we have the data, we have everything. We are below them, above them, and around them."
Boom!!!! MSFT has clearly been building redundancy against OpenAI following their 2023 board saga, proven by their Mistral AI investment
As general-purpose LLM arms race continues, app builders' switching cost/model dependency decreases asymptotically
Seattle-based Open Interpreter unveiled the 01 Light, a portable open-source voice interface that connects to a user’s computer, allowing the AI to control apps, learn skills, and observe screens
Microsoft's $650M "licensing" deal with AI startup Inflection raises Q's about its true purpose. Officially focused on integrating Inflection's AI models into Microsoft Azure, deal terms revealed that MSFT also hired most of Inflection's team
Including DeepMind co-founder Mustafa Suleyman, who will lead Microsoft's consumer AI division
Amazon benefits heavily from Astera Labs' blockbuster IPO, whose stock skyrocketed 72% on the Nasdaq, valuing the data center chipmaker at nearly $9.5B
Amazon owns 2.3M Astera shares (worth $144M) and is a major client of Astera, which supplies connectivity chips for AI and cloud infrastructure
Key AI Product Updates:
NVIDIA announces a massive series of products/partnerships at their GTC 2024:
Next-gen GB200 chip, boasting 20 petaflops (petaflop = 10^15 floating-point operations/sec!!!), while the Grace Blackwell Superchip promises 30x faster LLM inference
NIM microservices for rapid AI deployment across environments, supporting support Retrieval Augmented Generation (RAG), integrating with vector databases, and can deploy on cloud, Linux servers, or serverless
Its shared VR technology comes to Apple Vision Pro, enabling enterprise developers to stream high-fidelity 3D digital twins
A strategic partnership with Oracle for global sovereign AI solutions
Earth-2, a cloud-based digital twin for climate modeling
Project GR00T, a foundational model for humanoid robots
The NVIDIA Quantum Cloud enabling researchers to simulate quantum computers using AI chips, including the H100 Tensor Core GPU
Github Copilot launches a new AI tool "CodeQL" that can automatically fix code vulnerabilities, querying the codebase as if it were structured data to parse the open-source SBOM
Apple researchers just published a new paper unveiling MM1, a family of multimodal AI models that combine visual and language understanding to enable advanced capabilities
Neuralink debuted its brain-computer interface (BCI) with its first human test subject, a 29-year-old paralyzed man, who demonstrated his ability to play online chess and Civilization using the BCI
Elon also claims Neuralink's brain chips have restored sight in blind monkeys with "Blindsight"
Character AI introduced a new feature allowing users to add and create voices for their characters with just 10 seconds of audio input
Google DeepMind introduced TacticAI, an AI system that can advise coaches on soccer tactics by using predictive and generative models to analyze previous play data
Created in a collaboration with Liverpool FC!!
Tokyo-based startup Sakana AI just released three new Japanese-language AI models, created by “merging“ existing open-source models to produce optimized “offspring” including: 1) A language model with strong Japanese fluency and math capabilities, 2) A diffusion model for high-speed image generation, and 3) A vision language model for Japanese text and images
Seattle-based Khosla partner Jon Chu led the $30M Seed of this company
Stanford researchers just developed Quiet-STaR, a new training method that enables AI models to generate an inner monologue, or ‘think’ before responding
When applied to Mistral 7B, Quiet-STaR doubled its scores on math + common sense reasoning benchmarks
Key AI Business/Investment Updates:
Saudi Arabia reportedly plans to create a $40B fund to invest in AI (allocated by Andreesen Horowitz), potentially becoming the world's largest investor in the sector
Biden Administration Invests $8.5B (with additional loans up to $11B) to fund Intel's domestic chip manufacturing, aligning with the CHIPS Act to bolster U.S. semiconductor production
Intel's $100B investment over five years aims to create over 30K US jobs
LLM company Cohere is looking to raise $500M at $5B valuation, despite only being at $22M annualized revenue (up from $13M end of 2023)
Google was fined $272M by French regulators for allegedly using content from news outlets to train the company’s Gemini AI model without consent
PNW AI/ML Fundings:
On the slower side this week…
Nomad Go (Seattle, WA) raised $1.9M in Seed II funding from undisclosed investors according to an SEC filing
An AI-enabled inventory automation platform designed for automated inventory counts and power decision-making - TFTD $3.4M
CEO: David Greschler
Mebot AI (Portland, OR) raised $250K in Pre-Seed funding from undisclosed investors
AI chatbots trained with your voice and personality - TFTD $250K
CEO: Peter Kirwan