It’s the Data, Stupid
2023 is the year of AI. It’s really the year of generative AI, but generative AI stands on the shoulders of great AI, and great AI stands on the shoulders of great data.
[This is the Insider Lite, the free newsletter from TalkingPointz]
So, 2023 is the year of great data: finding it and harvesting it. We’ve tolerated companies like Google and Facebook harvesting our information for decades, but this round is different. Now the goal is to make large language models (LLMs) better. My data helping LLMs may or may not help me, but it doesn’t exploit me like targeted ads or behavior manipulation. I’d be kind of disappointed if they excluded TalkingPointz.com or other forms of my content from their training data. If an LLM gets a comma wrong, I’d like to say, “I did that.”
Speaking of getting things wrong, it was silly for the world web to attack Zoom this month for its updated terms of use. I don’t think Zoom ever intended to use data without consent. They were pretty clear about consensual data sharing during their analyst event two weeks ago. (BTW, kudos to Zoom and Smita Hashim for the quick response. The company has proven itself quite “agile” at responding to controversies.)
The more significant issue is how ridiculous these end-user license agreements (EULA) are. No one reads them, and if they did, no one understands them. Even if you did read and understand them, vendors change them on a whim — usually right before you need to access the service again. Let’s require EULA changes to have a one-year notice period.
Without a doubt, LLMs are changing the way we look at data. Web crawling is OK; web scraping is bad. A lot of good data was free before it was recently deemed valuable (Reddit and Twitter/X).
Can Zoom Win at LLMs?
There’s no proven path to success here. Most likely, success will require money, vision, data, and speed. Zoom has all of that.
It is clear to me, and everyone else who attended #ZoomPerspectives23, that Zoom considers generative AI a big opportunity. They shared their plan to federate multiple LLMs (Anthropic, Open AI, Llama AI, and probably more). It’s a complex strategy, but it’s clear that every LLM behaves differently, meaning that a blended approach will be 1) unique and 2) possibly quite versatile. I get it — it’s why I prefer blended whiskies over single malts. It will also take a ton of cash, but Zoom happens to have that.
This approach still requires training data. Still, I don’t anticipate significant opt-out rates. That’s partly because Zoom has a freemium model, and generosity begets generosity. Zoom’s data is also valuable due to its diversified customer base: enterprise, government, education, SMB, and consumer. “That’s gold, Jerry.”
Additionally, Zoom has XD, or more formally, Xuedong Huang. He made his first appearance in the TalkingPointz paid Insider Report last July, when he moved from Microsoft to become CTO at Zoom. I pointed out that XD is the third senior exec from Microsoft to move to Zoom recently.
Microsoft recruited XD directly from Carnegie Mellon 30 years ago based on his work with AI speech recognition. He was put in charge of Microsoft’s spoken language initiatives in 1993 and was instrumental in creating Microsoft’s Speech Application Programming Interface (SAPI), the release of Microsoft Speech Server, and modernizing spoken language and integrative AI services via Azure AI.
Let me make this point very clearly: XD has had an amazing career in speech AI. His entire 30-year run has been at one employer, and his LinkedIn profile shows six titles with words like VP, Chief, GM, and Fellow. His colleagues in Redmond referred to him as Mr. Speech. This guy sees a huge opportunity with generative AI and LLMs, and he quit Microsoft — to go to Zoom! That’s where we are now. There’s more to this story, but it hasn’t been told yet. (More videos from Perspectives).
From the July Insider Report
MS Antitrust: The European Commission has opened a formal investigation to assess whether Microsoft may have breached EU competition rules by tying or bundling Teams to Office 365 and Microsoft 365 subscriptions.
The Commission said it’s “concerned” that Microsoft may effectively be forcing customers to include Teams with their Microsoft 365 subscriptions. No shit. That’s why Slack filed the initial complaint (3+ years ago!). Perhaps if the EU had acted sooner, Slack would still be a viable standalone vendor — though at least they acted. Where was the FTC?
This will amount to nothing for Microsoft for two reasons. First, Teams already won. The giveaway stage was great while it lasted, but that has predictably ended. Heck, even Microsoft has moved past giving it away and has engineered premium licenses and Copilot as creative fees on its free product.
The second reason is that the competitors have packed up and disappeared like the Lorax. The providers need to convince the customer to (1) pay for the apps and (2) not use the apps they already paid for. Cisco, Google, and Zoom all have robust chat apps that can compete against Teams. Most of the UCaaS chats are bundled with UCaaS (the exact opposite GTM strategy as Microsoft). There are quite a few chat apps if we expand our search to include Discord, Chime, and others, but most gave up on the mainstream enterprise adoption pre-pandemic.
It will fizzle, but it’s open season on big tech firms. In an Op-Ed in The New York Times, Sens. Elizabeth Warren and Lindsay Graham (since when do they agree on anything?) proposed creating a Digital Consumer Protection Commission charged with “licensing and policing” the biggest tech companies.
Meta Opened a Can of AI: Meta released its Llama 2 LLM as open source. Even more importantly, it released the training and fine-tuning weights that work it. Zuck even threw in a commercial license so everyone (except Google and Microsoft) can use it. (The free license is for companies with fewer than 700M MAU.)
Llama 2 is the first openly released model on par with ChatGPT. Llama 2 likely cost millions to create, so what has occurred here is quite extraordinary. It will both accelerate AI (as if we needed that) and cause some serious headaches for OpenAI/Azure/GCP and others.
Zuck’s move is disruptive but not completely original. AI has blossomed as a field because company researchers, academics, and other experts have been willing to share notes. Google built and publicly shared the transformer model that has become foundational in the newest AI research. Now, some of the most advanced AI in the world is available to more companies, students, and countries.
Preview of the August Insider (subscribe)
- Faded Blue Jeans
- 3 Angles to Aceyus
- Zoom Perspectives 23
- MQ CCaaS (upcoming research) (Does CCaaS Matter?, recent research).
- Dear Tarek
Upcoming Events: CPaaS Acceleration Summit in Amsterdam, Cloud Communications Alliance in Copenhagen, UCExpo in London
2023 Insider Reports
- Insider Jan 2023
- Jan Insider Lite 2023
- Insider Feb 2023
- Insider Feb Insider Lite
- Insider March 2023
- Insider April 2023
- Insider Lite April
- Insider May 2023
- Insider June 2023