28th Edition: 2/23/24
AI humanoids, open-source SLMs, Reddit's groundbreaking $60M AI content-licensing deal to beef up IPO plans, NVIDIA's record-breaking stock surge from 265% YoY revenue growth, and more...
General-purpose LLMs are continuing to drive rapid improvements in speed/accuracy, but 95% of enterprise AI model spend is still from run-time inference costs rather than pre-training. To adopt, enterprises began depending on RAG architecture + SLMs to drive efficiency and specificity of output in production. Accordingly, I thought I'd start this week with a graphic that I found useful from Menlo Ventures (thanks Derek!) on the AI maturity curve.
Now, let's get started!
PNW AI News:
Microsoft has confirmed its plans to integrate OpenAI's Sora video-generating chatbot into its Copilot coding assistance tool
Microsoft to invest $2.1B in Spain to expand AI and cloud infrastructure, furthering MSFT's overseas public-private collaboration efforts
Key AI Product Updates:
Google introduced Gemma, a suite of lightweight, open-source AI models optimized for NVIDIA GPUs + Google Cloud, derived from the technology underpinning Google's larger Gemini models. Both 2B and 7B parameter model sizes available.
In line with shift towards small language models (SLMs) for introductory or lower-throughput use cases
ElevenLabs can now generate sound effects, using OpenAI's text-to-video tool to create audio like "waves crashing" based on text descriptions, which could greatly enhance the realism of AI-generated video
Princeton researchers developed an AI system capable of predicting disruptive "tearing" events in fusion plasma up to 300 milliseconds in advance, trained on real-world experimental data rather than theoretical physics
Could make fusion reactors far more reliable and stable by micro-adjusting reactor power
OpenAI has updated its GPT Store, expanding developer profiles and enabling users to rate the third-party AI tools on the platform, providing feedback for both OpenAI and creators
The Groq AI model (not Elon's Grok) has surged into the spotlight due to its impressive response rates powered by its proprietary ASIC chip, generating ~500 tokens/second, outpacing ChatGPT-3.5's 40 tokens/sec
Novel language processing unit (LPU) entirely circumvents needs for GPUs
Stability AI released Stable Diffusion 3.0, an updated text-to-image generation model promising improved image quality and performance
Google is pausing Gemini’s ability to generate images of people, following heavy criticism after the model continuously overcorrected for diversity in its outputs
Google is developing ScreenAI, a new Vision-Language Model for understanding UIs and infographics, surpassing larger models in efficiency while introducing a novel textual representation of UIs enabled by LLM-generated training data
Key AI Business/Investment Updates:
Reddit signs an AI content licensing deal for $60M/yr to unnamed AI company looking to train its LLM, ahead of it's S-1 filing in preparation for its IPO. Reddit is seeking at least a $5B valuation
Emerging biz model for content owners to monetize their user data, following New York Times/MSFT lawsuit settlement originated from IP disputes over LLM training
Simultaneously, the S-1 reveals OpenAI CEO Sam Altman as the third-largest shareholder, with an 8.7% stake. Safe to guess who this "unnamed AI company" is?
SoftBank's Masayoshi Son seeks to raise $100B for AI chip venture Izanagi, in partnership with SoftBank-backed chip-designer Arm
Signals pressurization of AI chip competitive landscape, as demand swells
NVIDIA earning reports significantly outperform Wall Street estimates (265% YoY revenue growth)
Caused $277B surge in stock value (!!!), surpassing Meta's all-time single day record
Sabrina @ Madrona's Substack "Aspiring for Intelligence" explores Nvidia's growth in greater depth here
Figure AI raised $675M at $2B valuation from Jeff Bezos, MSFT, OpenAI to develop humanoid autonomous robots to tackle jobs dependent on unsafe human labor
Scale AI has been contracted Pentagon to develop a framework for testing and evaluating LLMs, aiming to safely deploy AI in military applications
Magic AI raised $117M Series B from Nat Friedman and CapitalGÂ to further its mission of automating software development end-to-end using AI
Disney announced its investment into five startups in its Accelerator program, three of which are AI native — ElevenLabs, PrometheanAI, and AudioShake
Chinese startup Moonshot AI raised over $1B Series B from Alibaba and others for its development of LLMs handling long-form text
Largest recorded Chinese LLM investment to date
Recogni raised $102M Series C to further develop its AI inference chip for generative AI use cases in autonomous vehicles
PNW AI/ML Fundings:
Finpilot (Seattle, WA) raised $4M in Seed funding led by Madrona, joined by Ascend and several angels
Help financial analysts speed up their investment workflows using AI without hallucination - TFTD $4.5M
CEO: Lakshay Chauhan
Levr.ai (Vancouver, BC) raised $1M in Seed II funding from Weave VC and Sprout Fund
A fintech software company transforming the way businesses access and manage loans - TFTD $2.1M
CEO: Kaylan Pepin
Nate @ Ascend's market map of the Seattle AI ecosystem highlights the depth of 50 Seattle-built AI products:
