EP64: Microsoft Build, Can We Get This Song to #1? Google AI Fail, GPT-4o, Phi-3-Vision, Mistral-7B-v0.3
May 24, 2024
auto_awesome
The podcast covers Google's AI failures, the reality of AI usage, GPT-4o updates, Microsoft Build innovations, testing Phi-3-Vision, exploring Mistral-7B v0.3, AI startup bubble concerns, and a plea to help a song trend on Udio.
Google's AI search summaries face accuracy challenges due to prompt injection attacks, raising credibility concerns.
Microsoft Build introduces Co-Pilot Plus PCs with recall feature, highlighting privacy risks and data security implications.
AI startups like Adept struggle with stability and saturation, seeking buyers after raising funds without a product.
Deep dives
Introducing Digital Godfather Song and UDO Challenge
A challenge is proposed to make a song, Digital Godfather, trend on UDO after the podcast episode. Listeners are encouraged to listen to the song, share it, and add it to playlists to boost its popularity.
Issues with Google AI Search Summaries
Google's rollout of generative AI search summaries faces challenges with inaccurate information. Examples include prompt injection attacks on websites, leading to misleading search results. The implications of AI search summaries being manipulated raise concerns about credibility and information accuracy.
Concerns over AI Summary Accuracy and Manipulation
The podcast discusses the potential dangers of AI-generated search summaries being manipulated or providing inaccurate information. Examples of fabricated information through prompt injection attacks highlight the risks of trusting AI summaries and the need for accuracy and credibility in search results.
Microsoft Build Unveils Co-Pilot Plus PCs and AI Features
Microsoft Build introduces Co-Pilot Plus PCs with an MPU, enabling new features like 'Recall' to capture screen snapshots for later retrieval. The podcast discusses the implications of local inference on personal data and the potential misuse or security risks associated with constant monitoring and recall functionalities.
Mr. All 7 Billion Vision Model
The podcast episode discusses the Mr. All 7 billion vision model, a small model that can be run locally. The results from testing varied, with examples ranging from correctly identifying book titles to misunderstandings like suggesting eating 'sour patch squirrel.' Despite being unreliable in some scenarios, the model showed potential for local use.
AI Startups and Industry Challenges
The podcast delves into the challenges faced by AI startups like Adept and stability. Adept, a company creating AI tools for software interactions, sought buyers after raising large funds without presenting a product. The episode also contemplates the saturation and financial struggles of AI companies operating in a competitive market where big tech giants like Microsoft and Google are rapidly advancing with innovative solutions.
Thanks for listening and all of your support, we appreciate it.
-------- CHAPTERS: 00:00 - Google's AI Overview Fails 18:50 - The Reality of Using AI 25:02 - Two weeks of GPT-4o 34:36 - Microsoft Build: CoPilot+ PCs, Recall, AI Narks, Phi-Silica, Team CoPilot, CoPilot with GPT-4o Voice 54:41 - Phi-3-Vision Testing 57:26 - Mistral-7B v0.3 Uncensored with Function Calling Testing 1:08:55 - AI Startup Bubble? Lots of AI Startups looking for buyers... 1:15:37 - Help us get this song to #1 on Udio!
Get the Snipd podcast app
Unlock the knowledge in podcasts with the podcast player of the future.
AI-powered podcast player
Listen to all your favourite podcasts with AI-powered features
Discover highlights
Listen to the best highlights from the podcasts you love and dive into the full episode
Save any moment
Hear something you like? Tap your headphones to save it with AI-generated key takeaways
Share & Export
Send highlights to Twitter, WhatsApp or export them to Notion, Readwise & more
AI-powered podcast player
Listen to all your favourite podcasts with AI-powered features
Discover highlights
Listen to the best highlights from the podcasts you love and dive into the full episode