Feb 4, 2025 7 min read

DeepSeek Meditation, AI Faster, Kill AI Messaging, SMB SOS

Dialog: Exchange Number 16

Thinking Out Loud

The Meaning of DeepSeek: The DeepSeek fallout and soul-searching continues. DeepSeek is "here to stay," yet there's an effort to ban it and perhaps all Chinese AI. Missouri Senator Josh Hawley introduced a bill that would effectively make all Chinese AI downloads illegal. It won't become law but it probably won't be the last attempt. The media coverage of DeepSeek is a David v. Goliath story: scrappy Chinese company builds highly competitive AI product for $6M using lower-grade chips. Much of this is false: DeepSeek probably used derivative "distillation" of OpenAI's LLM and it's actual AI investments are likely more than $1B, though it is innovative. Putting aside the irony of OpenAI complaining about its IP being used without consent, DeepSeek would have cost more without US foundation models. DeepSeek is now being evaluated by a growing range of companies. Here's the impact: 1) significantly reduced AI costs, 2) firms migrating from OpenAI, Gemini, 3) faster innovation, more open-source AI, 4) efforts to replicate DeepSeek's approach, 5) more competitive global market, 6) perceived commoditization of LLMs, with value creation at the application layer.

OpenAI Goes Deep: Following last week's DeepSeek crisis, OpenAI has introduced "DeepResearch," which happens to have the same name as a nearly identical product from Google. Both are "narrow agents" that focus on a specific domain – in this case research – compared with OpenAI's Operator, which is a general agent and has greater limitations accordingly. We have yet to test DeepResearch (OpenAI's version). We did test the Google version when it was released and while the abstract capabilities were impressive, it didn't seem fully baked. We've looked quickly at OpenAI's version and were moderately impressed (issues: random sources, can't read many sites). But anecdotal early indications suggest OpenAI's version is better and more thorough than Google's. But it's very early, both will get better. And ChatGPT's DeepResearch is a very useful new tool for multiple use cases. But it's also concerning. How many scientific and academic papers will now be substantially AI generated, not to mention high-school and college reports. It raises a number of ethical issues.

Ask for Me: Google is testing a (sort of) new product that uses AI voice to schedule appointments and reservations for consumers. It's called Ask for Me and it's an update of Google Duplex, which was introduced at Google's 2018 developer conference, in the pre-generative AI era. It's available as a Search Labs experiment (waitlist). It's being tested in automotive repair and nail salons. After searching on Google, the Ask for Me module appears below the Map/Local Pack. Users provide certain preferences and information and then "Google’s AI will then make calls to local businesses on your behalf and send you a summary of prices & availability, saving you time and effort." It thus appears to be contacting multiple businesses, but it's not clear using what criteria. Effectively this is an RFP tool that, if it survives, will probably will get the ads treatment – speculation: sponsors would be included in the returned list of prices and availability.

Unimpressed by Apple Intelligence: Apple Intelligence has generally been oversold, at least at this stage. During Apple's Q1 fiscal year 2025 earnings call, CEO Tim Cook reviewed the rollout history of Apple Intelligence, introduced in October 2024. Most notably it provides access to ChatGPT via Siri and has a number of other features: Writing Tools, Image Playground, Genmoji, and marginally improved Siri capabilities. There had been some evidence that Apple Intelligence wasn't boosting iPhone sales. However Tim Cook said, "we did see that the markets where we had rolled out Apple Intelligence that the year over year performance on the iPhone 16 family was stronger than those where Apple Intelligence was not available." He also asserted that people are using all the tools. Siri has a few more capabilities and is slightly more "natural" sounding. However, it lags well behind ChatGPT's advanced voice mode, which is now the standard. Cook has promised a major Siri update in April.

News & Noteworthy

Microsoft CEO Nadella: the "proof of concept" phase of AI is now over.
A UC Berkeley team says it has reproduced DeepSeek's capabilities for $30.
Artificial Specific Intelligence vs. Artificial General Intelligence.
OpenAI now seeking to raise $40B more at a bubblicious $340B valuation.
Google using brute force to get Workspace users to adopt Gemini.
ElevenLabs raised a $180M Series C to voice-enable AI agents/assistants.
Perplexity, Microsoft, others have already deployed DeepSeek on their servers.
Meta XR spending to reach $100B as of this year (but Ray-Bans selling).
AI assisted human creativity can now be copyrighted, says US copyright police.
First EU AI Act provisions now have the force of law with penalties.
Apple's lack of a major proprietary AI model now looks smarter.
The AI/robot doctor will see you now (NYT).
Beatles' AI-enabled "final song" Now And Then wins Grammy award.

AI Faster than Internet

At at a eye-popping $8 million for 30 seconds, the Super Bowl this Sunday is expected to feature a lot of ads for AI. Awareness, if not usage, of AI is already very mainstream. A survey cited by the Federal Reserve Bank of St. Louis, conducted in early Q3 2024, found that 40% of US adults had used AI. Just under 11% used AI daily and 33% used it at least weekly, according to the poll of roughly 4,600 respondents. Dialog's consumer survey (n=1K US adults), conducted in November 2024, extrapolated that 44% had used AI. Among qualified AI users, however, Dialog found that 57% used it daily – a much larger number. The Federal Reserve survey found that while younger males were more likely to be AI adopters, "generative AI use is widespread across gender, age, education, industries and occupations." AI penetration has happened much faster than earlier technologies. The PC took 12 years to get to the same penetration level and it took the internet about four years. Things are going to only go faster from here.

Kill 'AI' in Your Messaging

As hype intensified around AI, SaaS companies and B2B marketers were quick to add "AI-powered" (or similar language) to their sites and marketing copy everywhere they could. Motivated by a fear of falling behind; AI mentions were intended to convey that the technology in question was state of the art. Yet a surprising new study from Irrational Labs found that using "AI-powered" or "generative AI" in your messaging does not give people confidence, increase trust, or motivate them to buy. Indeed, the study found that "using the term 'AI' often lowered expectations about the product’s potential impact, likely due to growing skepticism around AI technologies." The study (n=767 US adults) showed participants landing pages for actual SaaS products with different marketing copy: one leading with "AI" and an alternative page that emphasized features and benefits without mentioning AI. As you might guess, the "value messaging" won. Accordingly, the study recommends using "AI" sparingly and thoughtfully in marketing, rather than spraying it everywhere. Focus on user benefits and tangible value. AI is a black box abstraction that doesn't immediately convey clear value.

SMBs Are Struggling, AI to the Rescue?

From Glen Allsopp on LinkedIn comes anecdotal evidence of SMB interest in AI education. As we used to repeat like a mantra, there's demand for reliable practical education about AI and how it can be applied in different local business contexts. We observed back in August, "There's still a gap between awareness and active [SMB] adoption – and significant need for AI education. This creates a window of opportunity for SaaS providers and technology vendors. Become that trusted information source." A few have done this sporadically but not in any sustained way (without selling). Both our 2024 SMB surveys (Q1, Q3) show the majority of local business owners are optimistic about AI. Many are figuring it out through trial and error; trusted third parties could accelerate that process to everyone's benefit. In the aggregate SMBs are struggling, with increasing market power concentrated in big companies. SMBs are having a harder time getting attention and competing. And since The Great Recession (2007-2009) there's been a shift: more people in the US (53%) are now employed by large enterprises. We won't use the "level the playing field" cliche. But AI presents a viable opportunity to improve SMB marketing, operations and customer service. DIY won't get them there.