MC.34: My Fears About AI Adoption

Discover the unexpected dip in ChatGPT, China's new entry into the AI video race with Vidu, and the real challenges behind using AI in production with "Air Head".

Author

Samuel ROY
May 02, 2024

Hi everyone,

Stumbled on the deadly Datos chart showing that a lot of people are trying out ChatGPT but quickly drop it. The assumption that search could be replaced by AI assistants in the near term looks unlikely from these findings.

It shows we are still at the evangelization phase. People are not invested in learning AI tools. The value foreseen and state of UX are not enough to motivate them.

It scares me because we're building a product for macOS in the space. I thought we could rely on the evangelization made by the bigger players. The plan was simple: provide a 10x better experience of AI for macOS power users who understood the value of ChatGPT. Now I’m unsure, who’s there to find if they've already dropped out?

This could also explain the frenzy towards agents. It’s easier to market and use agents (if they work) on specific tasks. But in my opinion, we’re far from production-ready agents. It’s already hard to get one-shot AI actions not to behave like a slot machine, so multiple agents with multi-turns and lengthy contexts? It’s too slow and costly. At least until Groq opens the gate to everyone with high limit rates.

The path taken by GitHub with their Copilot native workspace is the smartest one. Agents might help but you stay in control. Agents aren’t even named, it’s just a smarter dev environment avoiding criticism if it’s dumber than expected.

My last thought is that AI adoption might happen in industries where the front customer has no choice but to use it: customer support, call centers, and the like. Even if the system is not perfect, just like it actually is with humans, AI by its scale and rapid enhancements can easily take these markets.

Are you still using ChatGPT or a similar service?

Login or Subscribe to participate in polls.

Cheers,

Sam

PS: updated Alter website with lots of demo videos, check them out.

In this issue:

📚 Joseph Thacker: Incremental Learning LLM Pattern
📖 Jim Nielsen: You Are What You Read, Even If You Don’t Always Remember It
💡 HubSpot: Spotlight with HubSpot AI
👨‍💻 GitHub: Copilot Workspace
🔗 Anon: User-permissioned integrations for sites without APIs
🇨🇳 Vidu: China's Answer to OpenAI Sora
🎬 FXGuide: Actually Using SORA
📈 Bobby Pinero: SaaS metrics for fundraising
🔍 Datos: Predicted 25% Drop In Search Volume Remains Unclear
💵 Microsoft: US$1.7 billion investment in Indonesia

Updates & tools

Joseph Thacker: Incremental Learning LLM Pattern
Feeding a book to AI in chunks to better understand its nuances sounds fascinating. It's akin to teaching a machine to read and remember a story as we experience it, bit by bit, potentially changing our interaction with stories and information retrieval. MORE

Jim Nielsen: You Are What You Read, Even If You Don’t Always Remember It
Jim Nielsen says that books change us, even if we don't remember all the details. He compares it to how food affects our bodies. This hits home for me because I often forget the names of books and movies I've watched. MORE

HubSpot: Spotlight with HubSpot AI 
HubSpot's new feature automatically generates replies for reps using AI. They leverage extensive data to tailor responses, a capability smaller players lack due to data ownership constraints. This is a strategic and effective move. MORE

GitHub: Copilot Workspace 
GitHub is launching a Copilot-native development environment. It's a clever approach. Instead of overpromising with an AI developer, it adopts a smart workspace narrative to sidestep criticism while clearly paving the way for AI development agents. MORE

Anon: User-permissioned integrations for sites without APIs 
Anon is tackling the tough issue of enabling developers to build integrations for websites without APIs. This approach allows automation or system integration even without direct API support, targeting AI agents developers that need this authentication layer. MORE

Vidu: China's Answer to OpenAI Sora
Vidu isn't as good as Sora. It only makes 16-second videos, while OpenAI can make up to 60-second videos that look better. Congrats to Vidu on achieving significant results with a budget and compute power far from OpenAI's capabilities. The race has begun. MORE

FXGuide: Actually Using SORA 
The post-production tweaks necessary to maintain consistency in the AI-generated character in "Air Head" were eye-opening. They demonstrate that while AI can create impressive outputs, significant work is required to effectively use them. MORE

Bobby Pinero: SaaS metrics for fundraising 
Bobby Pinero explains how to write an effective data deck. It focuses on demonstrating your business's raw performance through data, a powerful method to build trust and clarity. Five keywords: revenue, acquisition, engagement, retention, financials. MORE

Datos: Predicted 25% Drop In Search Volume Remains Unclear 
The article discusses the drop in user interest in AI chatbots like ChatGPT after initial enthusiasm, in contrast with Google's consistent use. Check the chart; it's scary. MORE

Microsoft: US$1.7 billion investment in Indonesia 
Microsoft is investing $1.7 billion in Indonesia to enhance its tech infrastructure, focusing on cloud and AI. Is there a country where Microsoft hasn't invested? MORE

Reply

or to participate.