25th Edition: 2/2/24
Deepfake concern ahead of the 2024 election, Meta's code generator with a massive context window, AI2's new LLM, and enterprises embracing RAG…
Thanks for tuning into my first public-facing edition. Let’s dive right in!
PNW AI News:
AI2 officially released OLMo 7B, an open source LLM with a full pre-training data set Dolma (3T token corpus) and finetuning SDKs included, on Hugging Face and Github. It was developed in partnership with AMD, Databricks, UW, and Harvard's AI Lab
MSFT revenue surges 18% QoQ to $62B, largely due to Azure consumption volume driven by AI + the Activision Blizzard acquisition finalization
Seattle startup Protect AI, focused on monitoring/securing companies' ML systems, acquired Laiyer AI for their open-source module "LLM Guard" on undisclosed terms
AI2 Co-founder Oren Etzioni launched across-the-aisle nonprofit TrueMedia focused on detecting political deepfakes, ahead of upcoming election concerns around AI interference in media
MSFT Employee whistleblower silenced in attempt to illuminate DALL-E 3 model misbehavior issues surrounding explicit content generation by prompt jailbreaking
Direct response to Taylor Swift deepfake controversy
Amazon launches their in-app AI shopping copilot "Rufus" to trained on user shopping history to create tailored product suggestions and reduce abandoned-cart rates
MSFT announced plans to launch Copilot AI app on Apple's Vision Pro headset immediately upon the device's launch (scheduled for today)
Key AI Product Updates:
Meta AI introduces Code Llama 70B, an open-source AI model to advance code generation in Python, C++, Java, and more with a larger context window of 100K tokens (!!!)
Can handle far, far more complex code repo's
Pinecone’s study reveals that retrieval-augmented generation (RAG) to access external docs makes open-source models better than GPT-4. LLMs with RAG achieve a 13% improvement in response quality for “Faithfulness”
RAG showing potential to boost LLM accuracy, even at >1B document scale
Google's chatbot Bard announces slurry of updates including:
Running on the new Gemini Pro model rather than LaMDA (outperforming GPT-4 in several measurements)
Can generate images multimodally using DeepMind's Imagen 2
Can perform RAG on your own G-suite
Google is enhancing Google Maps with LLMs to suggest places based on user queries, trained on the recommendations from its 300M contributors
OpenAI shipped a new ‘GPT mentions’ feature to ChatGPT, allowing users to seamlessly integrate custom bots into a conversation by tagging them
Google has rolled out ImageFX, an AI-fueled image generator powered by Deepmind's Imagen 2, to compete with Midjourney/DALL-E3
Twin Labs is automating repetitive tasks by letting AI take over your mouse cursor using GPT-4V to understand NextJS/React frontend architecture, rather than other agents that automate by mapping API endpoints' connectivity on backend
Tencent publishes research on proliferation of multimodal LLMs, and compares their performance side by side:
Key AI Business/Investment Updates:
Shares of major AI like MSFT, Google, AMD and Nvidia fell sharply this week after the Q4 earnings reports failed to impress investors anticipating an AI-driven boom
Adoption in enterprises still congested by "curiosity budget POCs", pending improved security guardrails before in-prod adoption
The FCC is taking steps to criminalize AI-generated robocalls, aiming to combat the proliferation of unsolicited calls using synthetic voices
School bus platform Zum raised $140M Series E at a $1.3B valuation from SoftBank, Sequoia, GIC and more for its AI-powered route optimization, EV fleet management, and parent communication tools used by 1K+ US school districts
Sam Altman is advancing discussions with Samsung and SK Group in South Korea to forge an AI semiconductor alliance, with major investors like SoftBank and Mubadala in the mix
Apple CEO Tim Cook has confirmed Apple is launching generative AI features on the iPhone “later this year” on their Q4 earnings call
Meta plans to deploy its in-house custom AI chip, “Artemis,” in 2024 to augment its data centers’ AI capabilities and build redundancy against dependence on Nvidia GPUs. This is v2 of the Meta Training & Inference Accelerator (MTIA) launched in May 2023
Sam Altman predicts that a ‘one person, billion dollar company’ will happen thanks to AI in the near future. He is betting on it in his tech CEO groupchat
Volkswagen opened an "AI lab" to create prototypes and form tech partnerships on products leveraging AI, like predictive maintenance & voice recognition
PNW AI Fundings:
Proniras (Seattle, WA) raised $4.7M in post-Series A funding co-led by ARCH Venture Partners and Accelerator Life Science Partners, joined by WRF Capital, Watson Fund, and several angels
AI-enhanced development of a new parenteral medical countermeasure (MCM) to treat seizures and brain injury - TFTD $13.7M
CEO: Christopher Toombs
Quindar (Denver, CO/Seattle, WA) raised $6M in Seed funding led by FUSE, joined by Y-Combinator, Founders Club, K5 Global, Soma Capital, and more
Managing satellites as efficiently as servers by utilizing AI-driven insights and operations to revolutionize the industry’s approach to spacecraft management - TFTD $8.5M
CEO: Nate Hamet
Storedge AI (Seattle, WA) raised $3.1M in Grant funding from the US Dept. of Energy
Offers solar energy using a transportation infrastructure for charging EVs and also provides project development, analytics, power market, and procurement advisory for energy storage to utility-scale and commercial developers - TFTD $3.1M
CEO: Ranjan Gupta
Beagle (Bellevue, WA/San Francisco, CA) raised $3M From Davidovs Venture Capital and ML Capital Investments
AI-enabled platform provides a convenient user-user-friendly interface for discovery assistant that helps find and analyze documents that match one's needs to review large volumes of documents - TFTD $3.4M
CEO: Sergey Demyanov