• Lore
  • Posts
  • Lore #72: Google Gemini 1.5 Launch Results In $70B Market Cap Loss

Lore #72: Google Gemini 1.5 Launch Results In $70B Market Cap Loss

Author

Nathan Lands
February 28, 2024

Good morning.

This week, Google finally released Gemini 1.5, their competitor to GPT-4. And immediately, their stock dropped almost $70B as there was a severe backlash against the product.

As soon as it came out, Twitter was flooded with people sharing how the LLM has an apparent racial bias — but in the opposite of how you might imagine. Gemini 1.5 basically refused to generate photos of white people. This occasionally got comical as people shared that Gemini produced photos of Vikings, and most of them were not what you’d normally imagine a Viking to look like.

Obviously, they look cool, but they are not very historically accurate…

Many Silicon Valley leaders have taken this very seriously, including one of the most influential people in tech, Marc Andreessen. Marc thinks this isn’t a mistake and that it will only get far worse from here, with the ultimate goal of rewriting history and forcing their personal politics into all future products. And that the only answer is to support truth-seeking AI and open-source AI.

I personally have mixed feelings about this. I feel that solving bias in AI is a challenging issue. Most attempts at solving it will likely result in just creating a new form of bias, and there aren’t any easy answers. In general, the positive future I imagine with AI really requires AI that is absolutely truthful with us — to empower all kinds of people to live better lives by showing them reality and guiding them to a better life. That is what will result in the most significant positive impact on people’s lives and the greatest leveling of the playing field we could ever hope for.

With all of that said, Gemini 1.5 itself is actually very impressive technology. With a context window of 100k, you can put entire books into it and then ask questions of the LLM, which is extremely useful. I hope Google gets its act together and returns to its mission of organizing the world's information and making it universally accessible and useful.

Until then, I’ll probably continue to use Perplexity more and more instead of Google.

Lore - Venture Capital

  • Glean raised $200M for its AI-powered work assistant (Link)

  • Abridge raised $150M for AI model for clinicians (Link)

  • Exodigo raised $105M to make underground maps (Link)

  • Intenseye raised $64M to develop AI that can detect and help fix potential hazards in the workplace (Link)

  • Fabric raised $60M to grow its conversational healthcare platform (Link)

  • Napier raised £45M for its financial crime and regulation technology compliance software (Link)

  • Photoroom raised $43M for its AI image editor (Link)

  • Inkitt raised $37M for its self-publishing platform using AI to develop bestsellers (Link)

  • Qloo raised $25M to expand cultural reach with AI (Link)

  • Bria raised $24M for its image generator trained on licensed content (Link)

  • Loora raised $12M for its English learning platform (Link)

  • NLX raised $12M to grow conversational AI platform (Link)

  • Datology raised $11.65M to automate data curation for more efficient AI training (Link)

  • Parspec raised $11.5M to revolutionize construction (Link)

  • Monumo raised £10.5M for advanced EV motor design (Link)

  • Collov AI raised $10M to revolutionize home furnishing (Link)

  • FlowGPT raised $10M for its consumer-focused platform for sharing artificial intelligence apps (Link)

  • Novity raised $7.8M to streamline industrial predictive maintenance (Link)

  • Ema App raised almost $2M for its conversational AI for women’s health (Link)

Coming up:

  • Figure AI to Raise $675 Million for Human-Like Robots (Link)

  • AI Chip Startup DeepX Nears $90 Million in Fresh Funding (Link)

News - Lore

Google has announced Gemma, a new series of open AI models, including Gemma 2B and Gemma 7B, built on the technology behind the Gemini models. Aimed at fostering responsible AI development, Gemma comes with model weights, tools for innovation, and a Responsible Generative AI Toolkit to ensure safe application creation. Designed for compatibility across major frameworks and hardware, including NVIDIA GPUs and Google Cloud TPUs, Gemma models offer state-of-the-art performance with optimization for efficiency.

Scale AI has been selected by the Pentagon's Chief Digital and Artificial Intelligence Office (CDAO) to develop a comprehensive framework for testing and evaluating large language models (LLMs) in military contexts. This one-year contract aims to establish reliable methods for assessing the performance of AI models that can support military planning and decision-making. The initiative will focus on creating specialized evaluation sets and offering real-time feedback mechanisms for military applications, addressing the unique challenges of generative AI technologies like text, code, and image generation.

Microsoft has made a significant move in the AI space by investing €15 million ($16.3 million) in Mistral AI, a Paris-based AI startup known for developing foundational models that could potentially compete with OpenAI's GPT-4. The exact stake Microsoft will acquire depends on the valuation at the next funding round.

Mistral AI recently introduced Mistral Large, a proprietary LLM intended to rival other leading models. This strategic partnership also includes a distribution agreement for Azure. However, the investment has drawn regulatory attention, with the European Commission planning to scrutinize the deal amid ongoing investigations into Microsoft's relationships with AI companies.

Other news

  • Google announces an expanded partnership with Reddit (Link)

  • Tyler Perry puts $800 million studio expansion on hold because of OpenAI’s Sora (Link)

  • The Walt Disney Company Announces the 10th Year of the Disney Accelerator Program and its 2024 Participant Companies - including ElevenLabs (Link)

  • Perplexity Pro is coming to all SK Telecom users (Link)

Coolest Things - Lore

Gemini can write Selenium code to automate complex tasks

The release of Mistral Large

GPT store updates

Xiaomi’s AI portraits

AI Tools - Lore

  • Reveal: Create, launch, and analyze one-on-one research interviews (Link)

  • Opinly: Automate your competitor analysis (Link)

  • Tome: Create polished presentations and docs faster (Link)

  • Mixio: Launch a website in seconds (Link)

Find more tools at FutureTools.io.

AI Further Study - Lore

  • LongRoPE: Extending LLM Context Window Beyond 2 Million Tokens (Link)

  • Neural Network Diffusion (Link)

  • OpenCodeInterpreter: Integrating Code Generation with Execution and Refinement (Link)

  • Genie: Generative Interactive Environments (Link)

  • MobileLLM: Optimizing Sub-billion Parameter Language Models for On-Device Use Cases (Link)

That's it!

See you next week!

If you want to support Lore, please:

  1. Share this newsletter 

  2. Follow us on Twitter @NathanLands & Lore.

  3. Join our Generative AI community on LinkedIn.