New Technology / Ai Development
Track AI development, model progress, product releases, infrastructure shifts and strategic technology signals across the artificial intelligence sector.
Why AI Hasn’t Exploded Yet (And How It Will)
Topic
Enhancing Consumer LLM Adoption
Key insights
- Everyones been focused on agent coding and the SaaS apocalypse
- Basic improvements to chat apps could enhance user experience
- Caching results for frequently asked questions can improve response times
- Instantaneous responses will keep users engaged in apps longer
- ChatGPTs current model 5.2 Instant is not instant at all
- GPT 5.3 Codex Spark low responds in under two seconds
Perspectives
Discussion focuses on improving consumer LLM adoption through various strategies.
Proponents of AI Improvements
- Propose caching frequently asked questions for faster responses
- Highlight the need for instantaneous responses to enhance user experience
- Argue for minimizing model name visibility in user interfaces
- Suggest integrating ads to improve product quality and revenue generation
- Claim that faster response times will increase user engagement
Skeptics of Current Strategies
- Question the sustainability of ad-supported models and user trust
- Critique Perplexitys failure to attract advertisers despite demand
- Doubt the effectiveness of deep research workflows in user engagement
- Challenge the assumption that ads enhance product quality
- Express concern over the potential disconnect between user needs and product capabilities
Neutral / Shared
- Acknowledge the challenges faced by the model routing team
- Recognize the historical context of Googles monetization strategies
- Note the integration of AI models with cloud services
Metrics
response_time
38 seconds
response time for ChatGPT 5.2 Instant
Long response times can lead to user disengagement.
it took 38 seconds to deliver the full response
response_time
under two seconds
response time for GPT 5.3 Codex Spark low
Faster responses can enhance user interaction and retention.
it responded in under two seconds
weekly_active_users
a million weekly active units
user engagement level for Google products
This threshold indicates the viability of maintaining a product based on user activity.
A million weekly active. It's not probably worth keeping around.
Key entities
Timeline highlights
00:00–05:00
The discussion revolves around improving consumer LLM adoption through basic enhancements to chat applications. Key suggestions include caching frequently asked questions for faster responses and minimizing the visibility of model names in user interfaces.
- Everyones been focused on agent coding and the SaaS apocalypse
- Basic improvements to chat apps could enhance user experience
- Caching results for frequently asked questions can improve response times
- Instantaneous responses will keep users engaged in apps longer
- ChatGPTs current model 5.2 Instant is not instant at all
- GPT 5.3 Codex Spark low responds in under two seconds
05:00–10:00
The discussion centers on the challenges faced by the model routing team in enhancing user interaction with AI models and the integration of advertisements into LLMs. It also reflects on the historical context of Google's product monetization strategies and the implications of ad-supported models for AI platforms.
- The model routing team has a hard job to do
- You can already talk to the model in chat
- Deep research is buried under a plus button
- The model router should be intelligent about frequently asked questions
- Ads need to be integrated into the LLM
- Google Reader was killed because it never monetized effectively
10:00–15:00
Perplexity began testing advertising in 2014 but struggled to attract brands, leading to a low adoption rate. The discussion also highlights the integration of AI models with cloud services and the challenges in user interface design that contribute to churn.
- Perplexity started testing advertising in 2014
- Taz Patel, the executive leading the ads effort, left the company
- Perplexity only led in less than half a percent of the brands that wanted to advertise on ChatGPT
- Many reasoning models can already write some Python and execute it
- Most of the data in an average internet users life is mirrored in the cloud
- All three major LLM apps have Gmail integrations already