AI: Weekly Summary
...week ending August 25, 2023
Welcome to the Saturday weekly AI summary here on ‘AI: Reset to Zero’. Hope everyone is having a great Summer with the end of August approaching. Here’s this week’s AI highlights:
Nvidia quarter continues AI Gold Rush: Nvidia’s results this week made AI believers of any stragglers in the public markets, with the last negative analyst calls muted. The report had upside that surprised even the bulls , with revenues doubling to $13.5 billion in a year. Nvidia remains in the GPU catbird’s seat at least into 2025, regardless of ramping competitors. In a related story of note is this ‘origin’ piece on AI datacenter CoreWeave. They have investments by Nvidia and Microsoft, as well as early access to NVIDIA GPU chips, the most precious AI commodity.
Open Source Funding News: In other public market news, the long-anticipated ARM Holdings IPO was filed with the SEC, driven by Softbank, and likely the year’s biggest IPO. As the leading open-source chip fabrication company, with key customers ranging from Apple to Nvidia to Qualcomm and many others, the deal likely rings another bell for IPOs in 2023. Others expected this year include Instacart and others. In private open source funding news on the software side, saw the $4.5 billion valuation for LLM AI hub Hugging Face Funding led by Salesforce, and a host of other strategic investors. Highlights the on-going rush to build the underlying AI infrastructure in this AI wave.
Meta continues AI open source initiatives: As expected, Meta rolled out its open source Code Llama, based on the recent Llama 2, free for research and commercial use, complete with model weights. Of course it’s targeted against CoPilot for Github by Microsoft, and other coding focused LLM AI products. It’s all a race for the hearts and habits of Developers worldwide, as they build AI into software products and services everywhere. Separately, Meta also released ‘SeamlessM4T’, an open source multi-language and multi-modal LLM AI translation product that also got positive feedback. Both Code Llama and Seamless were highlighted by Meta AI Chief Scientist Yann LeCun online. I have a deeper dive on this AI universal translation trend with related context for those with further interest.
Google YouTube Universal deal: While the writers/actors continue to wrestle with the studios on a settlement, Google YouTube crafted a partnership with Universal Music Group, one one of the largest music publishers. The deal envisions an AI Music Incubator, and includes artists like Frank Sinatra estate, Yo Gotti, Anitta and others. Senior folks from both sides had positive things to say on the opportunities to innovate while getting artists paid for their content, and maintaining copyright and safety. Broader questions around AI and copyrights of course remain. Separately, YouTube also provided more details on their Shorts Algorithm, focused on the $250 billion “Creators Economy” (forecasted to grow to almost $500 billion by 2027 on Goldman Sachs estimates). YouTube Shorts of course competes with Meta’s Reels product (running at over a $10 billion run rate this year), and TikTok, which is on a parallel revenue path. All this of course while YouTube continues to evaluate other AI innovations as well around video.
Microsoft OpenAI build other AI partnerships: While Microsoft and OpenAI continue to move forward on their partnership, Microsoft on the Azure cloud business front continues to build relationships with other LLM AI providers. We saw the open source initiative with Meta’s Llama 2 a few weeks ago, and now another relationship with Databricks around AI software is in the news. Conversely, OpenAI is also working with other AI software infrastructure partners, with this week seeing a partnership with Scale AI, allowing companies to fine-tune GPT3.5. And presumably a similar deal with GPT 4 down the road. Remember Scale AI was the company picked by the White House for the ‘Red Team’ exercise for AI vendors Google, Anthropic, Hugging Face, Microsoft and others not too long ago. So both Microsoft and OpenAI are building AI infrastructure ecosystem partnerships while moving ahead on their own joint initiatives.
AI Papers and Developments:
New AI paper from Google DeepMind on improving Reinforced self-training (ReST) for language modelling, using existing output.
McKinsey introduces an internal generative AI tool called Lilli, leveraging a vast dataset. Example of vertical LLM AI efforts to expand data sources for LLM AIs in specific industries, that I’ve discussed in earlier pieces.
Posts of note this week:
I did a piece on the consumer and business AI applications of “Digital Twins”, the other major set of AI services beyond “Smart Agents”. All in Box 6 of the AI Tech Chart we’ve discussed.
Also did a piece on AI “Data Exhaust”, the next set of data sources both large and small that could power the upcoming set of more powerful Foundation LLM AI models.
Thanks for joining this Saturday with your beverage of choice. Lots of AI items to sample this weekend.
For your Sunday tomorrow, this week’s ‘AI In my View’ is coming up. Stay tuned.
(NOTE: The discussions here are for information purposes only, and not meant as investment advice at any time. Thanks for joining us here).