Google's True Moonshot, Security Flaw in OpenAI's GPTs, Harvey's Series B, & Apple Optimizes LLMs for Mobile
Ben Thompson proposed a true moonshot for Google, Lakera found a massive security flaw in OpenAI's Custom GPTs, Harvey raised an $80 million Series B, & Apple's LLM white paper could help iPhones
Google's True Moonshot, From "OK Google" to an AI Hardware Game-Changer
Ben Thompson of Stratechery proposed a new "moonshot" for Google, a hardware device leveraging Google’s AI expertise with an exclusive universal AI assistant.
Mobile was the last paradigm where platform market share shifted, as Apple and Google captured meaningful share from Microsoft's Windows. Generative AI could be the next paradigm, and Ben Thompson proposes that Google seizes it.
But given how much this decision will harm their relationships with Android phone makers, as well as how this decision will drive down their gross margins (Google has 56% gross margins vs Apple's 44%), does Google have it in them to go all in on this hardware device?
And even if Google tries, what are the odds that they'll succeed? Winning search so easily is primarily why they have a $1.8 trillion market cap, but they may not have the muscle memory to win a new market like AI hardware.
If founders Larry Page & Sergey Brin weren't involved, we would already know the answer. Seeing that they are, let's see what Google does.
Lakera Finds Security Flaw in Custom GPTs: Private Conversations Potentially Compromised
Benedict Böttger at Lakera discovered a significant vulnerability in OpenAI's Custom GPTs.
Creators of custom GPT models can extract data from conversations by embedding custom images in the model's responses. This method involves appending an invisible image, hosted on the creator's server, to every message. The image URL includes the user's message, thus covertly forwarding it to the creator's server.
As many commented in the LinkedIn post, this discovery raises serious concerns about user privacy and data security in OpenAI’s Custom GPTs. Perhaps this is a reason why OpenAI delayed their GPT Store until next year.
Harvey Raises $80 Million Series B from Kleiner Perkins, Sequoia, & OpenAI
Harvey, an AI startup targeting legal professionals, raised an $80 million series B at a $715 million valuation, led by Kleiner Perkins, Sequoia, & OpenAI.
Harvey aims to overcome the limitations of general-purpose AI models like ChatGPT by providing more accurate, secure, and private legal assistance, addressing concerns like data security and AI-generated errors.
They have already landed clients like Allen & Overy and Macfarlanes, as well as a partnership with PricewaterhouseCoopers. And these logos are paying real money, as Harvey has grown revenue by 10x since April. The new funding will be channeled into expanding their custom model building, team scaling, and product development to further cater to the legal industry's needs.
Apple's New White Paper Optimizes LLMs for Mobile
Apple published a white paper about running large language models (LLMs) on devices with limited memory, such as an iPhone. Their innovative approach involves using flash memory, a more abundant resource in mobile devices, to store AI model data. By employing techniques like 'Windowing' and 'Row-Column Bundling,' the method allows for more efficient data management and faster processing.
This approach enables AI models to run up to twice the devices memory capacity while increasing processing speeds by 4-25x. This advancement will help Apple add more sophisticated AI-driven features in iPhones, such as their own generative AI model, "Ajax".