New Technology / Ai Development
Google COSMO AI App Withdrawal and AI Innovations
Google introduced an experimental AI app called COSMO, designed to assist users by integrating local and cloud AI functionalities. However, the app was quickly pulled from the Play Store, indicating it may not have been ready for public release. The rapid withdrawal raises questions about Google's internal testing protocols and the assumptions made regarding user readiness for such technology.
Source material: Google Just Dropped COSMO Then Mysteriously Pulled It
Summary
Google introduced an experimental AI app called COSMO, designed to assist users by integrating local and cloud AI functionalities. However, the app was quickly pulled from the Play Store, indicating it may not have been ready for public release. The rapid withdrawal raises questions about Google's internal testing protocols and the assumptions made regarding user readiness for such technology.
COSMO aimed to simplify users' lives by organizing their day and answering complex questions. It was intended to run with a local Gemini Nano model or a remote server, allowing for a flexible assistant setup. Despite its potential, the app's features were not fully operational, leading to its premature removal.
In addition to COSMO, Google is advancing its AI Co-clinician, which assists doctors in telemedicine without replacing them. This system has shown effectiveness in addressing complex medication inquiries and guiding patients through physical assessments during simulated calls.
OpenAI's Codex has also evolved, introducing animated pets to enhance user engagement while adding practical features for coding tasks. Meanwhile, Anthropic is preparing for its upcoming developer conference, showcasing its Claude Jupiter version, which is currently undergoing testing.
Perspectives
Support for AI advancements
- Highlights the potential of COSMO to integrate local and cloud AI functionalities
- Confirms the effectiveness of AI Co-clinician in telemedicine
Criticism of COSMO's readiness
- Questions Googles testing protocols due to COSMOs rapid withdrawal
- Notes the high pricing and performance issues of Mistral Medium 3.5
Neutral / Shared
- Acknowledges the competitive landscape of AI development
- Recognizes the advancements made by OpenAI and Anthropic
Metrics
10 million units
projected shortage of health workers by 2030
This shortage highlights the urgent need for AI solutions in healthcare
the world is heading toward a shortage of more than 10 million health workers by 2030.
98 units
primary care questions tested with AI Co-clinician
This indicates the scale of the AI's testing and its relevance to real-world applications
Google tested the system on 98 realistic primary care questions.
97 units
critical errors made by AI in testing
This suggests a high level of accuracy for the AI in medical contexts
In 97 of them, the AI made zero critical errors.
$1.50 USD
cost per million input tokens for Mistral
High pricing may deter potential users compared to competitors
$1.50 per million input tokens
$7.50 USD
cost per million output tokens for Mistral
Expensive output costs could limit Mistral's market appeal
$7.50 per million output tokens
27 billion units
of parameters in Alibaba's Quinn 3.6
A lower parameter count may affect performance compared to Mistral
Quinn 3.6 has 27 billion parameters
72.4%
SWE Bench score for Alibaba's Quinn 3.6
Performance metrics are crucial for competitive positioning
Scores 72.4% on SWE Bench verified
Key entities
Key developments
Phase 1
Google introduced an experimental AI app called COSMO, which was designed to assist users by integrating local and cloud AI functionalities. However, the app was quickly pulled from the Play Store, suggesting it may not have been ready for public release.
- The block primarily promotes Googles AI advancements, including the COSMO app and other AI tools, highlighting their potential applications in various fields
Phase 2
Google's COSMO AI app was introduced but quickly removed from the Play Store, indicating potential issues with its readiness for public use. The app aimed to integrate local and cloud AI functionalities, reflecting the rapid advancements in AI technology.
- Googles AI Co-clinician has shown effectiveness in addressing complex medication inquiries, particularly in telemedicine, where it supports doctors without replacing them
- In simulated video calls, the AI Co-clinician successfully guided patients through physical assessments, although human doctors remained superior in detecting critical warning signs
- To ensure medical safety, Google utilizes a dual-agent system where one AI interacts with the patient while another monitors the conversation
- OpenAIs Codex has added animated pets to enhance user engagement, alongside practical features like auto-detecting configuration files from other coding agents
- Anthropic is advancing its Claude Jupiter version, currently undergoing red teaming in preparation for an upcoming developer conference
- Mistral AIs Medium 3.5 model has received criticism despite its advanced capabilities, reflecting the competitive pressures in AI development
Phase 3
Google's COSMO AI app was introduced but quickly removed from the Play Store, indicating potential issues with its readiness for public use. The app aimed to integrate local and cloud AI functionalities, reflecting the rapid advancements in AI technology.
- Mistral Medium 3.5 faces criticism for its high pricing and is viewed as inferior to competitors like Alibabas Quinn 3.6, which offers a more affordable option with fewer parameters
- Despite these challenges, Mistral remains a key player in the Western AI market, especially among European enterprises that value GDPR compliance and self-hosting capabilities
- The new model incorporates features from earlier versions, enabling configurable reasoning efforts, but it has not yet gained recognition on major independent leaderboards, raising concerns about its competitiveness
- The mixed reception of Mistral Medium 3.5 underscores the increasing preference for open-source models in the AI sector, as developers lean towards options that are freely available and easier to implement