FOD#14: Does AI alphabet also start with A: Amazon and Apple reveal their AI plans

Today, our FOD will be shorter than before. I just came back from a family trip to Sardinia and Norway, and now, I'm on my way to Las Vegas to participate in the Ai4 conference in two roles.

First, I'll be a moderator for an exciting discussion titled "Creating Your All-Star AI Team." Stay tuned for insights!

Second, I will have a fireside chat with Che Chang, General Counsel at OpenAI. We'll be discussing how to Navigate the Legal Complexities of Cutting-Edge AI.

Do you have questions for Che Chang? Send me a note at [email protected]. If you participate: hope to meet you there.

Now to the last week's news.

_{Some of the articles might be behind the paywall. If you are our paid subscriber, let us know, we will send you a pdf.}

Does the AI alphabet also start with A?

Apple's Tim Cook made headlines recently with a declaration that Apple is "building AI into every product." It does not just talk; their whopping $22.6 billion in research is largely funneled toward AI. And over at Amazon, CEO Andy Jassy revealed that AI initiatives are popping up in every corner of their business realm (Stratechery did a good analysis of Jassy’s Earning Call remarks).

Microsoft's strategic alliance with OpenAI, Google's self-developed AI models, and Meta's open-source AI strategy (though Open Source Initiative says that the license of LLaMa models such as LLaMa 2 does not meet the terms of the open source definition) have already paved a clear AI narrative for them. This leaves Apple and Amazon in the limelight, with both uniquely positioned to lead AI's next wave without a conventional strategy.

Amazon, for instance, envisions generative AI as a three-tiered ecosystem, with a solid foundation of computing capabilities. Jassy showcased Amazon's custom AI chips like Trainium and Inferentia as game-changers in this space. Then there's the intermediate layer, Bedrock, offering a buffet of large language models. And finally, atop this structure are the applications, like the innovative Amazon CodeWhisperer, aiming to redefine the developer experience.

While Amazon's aspirations are vast, they notably lack access to OpenAI's models and are yet to get their hands on Meta's Llama 2. This AI tug-of-war among tech behemoths is fascinating, to say the least.

In all of this, Apple's AI card remains somewhat hidden, especially in the developer AI tools domain. But knowing Apple, they might just be waiting for the right moment to unveil their masterstroke.

Additional info: You can read this overview by Devansh on How Amazon tackles a multi-billion dollar bot problem.

News from The Usual Suspects

OpenAI

• Six improvements have been made to ChatGPT.

Now GPT4 actually talks to me and asks me questions after each task I assign to it. To be honest, it’s a bit creepy and I’m not sure if actually helpful.

• OpenAI now allows users to prevent their web crawler from accessing and utilizing website data, enhancing user privacy. You can read the documentation on GPTBot.

• On the other hand, Zoom's (OpenAI’s partner) terms of service that allow the use of user content to train AI without opt-out raised a wave of concern and rage on Hacker News.

Google

Google researchers are exploring the integration of various medical data types, like images and lab tests, into multimodal AI systems. Recent papers showcase their strides in grafting medical imaging models onto LLMs, aiming to create versatile medical AI tools. The next challenge? Ensuring these innovations mesh well with the medical community.

Meta

Meta's new open-source framework, Audiocraft, converts text prompts into audio or music. Comprising three models - MusicGen, AudioGen, and EnCodec - it offers intriguing possibilities. No public demo is available yet, but you can check the code on Github.

MuskX

Since February 2023, the domain AI.com has been redirecting visitors to ChatGPT. However, last week, it was revealed that Elon Musk acquired the rights to AI.com. It would be quite peculiar if OpenAI indeed sold the rights to him.

Additional info: A recap on big tech's Generative AI moves from their recent Q2 earnings calls by Tanay’s Newsletter.

Also, Analytics India Magazine raises some interesting points about us possibly living in a simulation orchestrated by the tech giants.

Is AI Good or Bad for the Economy?

While we are working on the article about the AI bubble, a16z presents an argument as to why generative AI is different and what it might mean for the VC industry: “AI companies have demonstrated unprecedented consumer interest and speed of adoption. Since entering the zeitgeist in mid to late-2022, generative AI has already produced some of the fastest-growing companies, products, and projects we've seen in the history of the technology industry.”

At the same moment, Gary Gensler, the chairman of the SEC, raises alarming concerns about AI integration into the financial system. “This technology will be the center of future crises, future financial crises,” Gensler said. “It has to do with this powerful set of economics around scale and networks.”

a16z is famous for setting the narrative that aligns with the main thrust of their investments, and SEC is known for being overly cautious. Stay tuned for our analysis of the AI bubble this Friday.

Might also be an interesting read: The state of AI in 2023: Generative AI’s breakout year by McKinsey.

Other news, categorized for your convenience:

Climate Analysis and Forecasting

IBM, in collaboration with Hugging Face and using NASA's satellite data, has open-sourced the watsonx.ai geospatial foundation model. This AI model, designed for advanced climate science and Earth studies, marks NASA's first publicly available AI foundation model and is also the largest geospatial model hosted on Hugging Face. Tapping into NASA's extensive 250,000TB climate dataset, this project, initiated just 6 months ago, has birthed the most potent AI tool for climate understanding. This AI promises unparalleled accuracy in climate predictions. A commercial version is slated for release by IBM later this year, offering industries foresight on weather and climate risks.

Btw, both IBM and Hugging Face recently became members of the PyTorch Foundation.

Tools and Software

• llama2.c: Released by Andrej Karpathy, it translates a trained version of Llama 2 into 500 lines of C code →go to GitHub

Model Launches & Collaborations

• Alibaba’s Open-Source Models: Alibaba's cloud computing unit unveiled two open-source AI models: Qwen-7B and Qwen-7B-Chat. Aimed at assisting SMEs with AI, the entirety of these models (code, weights, documentation) will be openly available. However, large corporations will need a license →read more

• Alfred-40B-0723 Model: Introduced by LightOn AI, this open-source language model is energized by Falcon RLHF, the Generative AI Copilot. It underscores the power of Reinforcement Learning from Human Feedback and has been rigorously trained for high-quality output delivery →read more

Multimodal AI Models

• Instructblip-vicuna-13b: A model by Salesforce Research, it recently led in a multimodal LLM benchmark →read more

• UnIVAL: Introduced by Sorbonne University Researchers, this is a unified AI model that integrates image, video, audio, and language tasks without the need for large data sets →read more

Efficiency in LLMs

• Skeleton-of-Thought: A novel method aiming to reduce generation latency in LLMs by using parallel decoding, mimicking the human thought process →read more

We are watching

Keynote address by Yann LeCun: “Towards Machines that Can Learn, Reason, and Plan”

Thank you for reading, please feel free to share with your friends and colleagues 🤍

Another week with fascinating innovations! We call this overview “Froth on the Daydream" - or simply, FOD. It’s a reference to the surrealistic and experimental novel by Boris Vian – after all, AI is experimental and feels quite surrealistic, and a lot of writing on this topic is just a froth on the daydream.