250: The Cloud Pod Goes Nuclear Powered

Welcome to episode 250 of the Cloud Pod  podcast – where the forecast is always cloudy! Well, we’re not launching rockets this week, but we ARE discussing the AI arms race, AWS going nuclear, and all the latest drama between Elon and OpenAI. You won’t want to miss a minute of it!

Titles we almost went with this week:

• The Paradox of AI choice
• Amazon just comes across super desperate on RACING to AI foundation model    
•      support
• Your new JR developer Test-LLM
• If you can’t beat OpenAI, sue them

A big thanks to this week’s sponsor:

We’re sponsorless this week! Interested in sponsoring us and having access to a specialized and targeted market? We’d love to talk to you. Send us an email or hit us up on our Slack Channel. 

General News 

01:12  IT Infrastructure, Operations Management & Cloud Strategies: Chicago (Rosemont/O’Hare), Illinois 

• Want to meet cloud superstar Matthew Kohn in person? He’s going to be giving a talk in Chicago, if you’re going to be in the neighborhood. *Maybe* he’ll have some stickers.
• 11:30am – 12:30pm: Using Data and AI to Shine a Light on Your Dark IT Estate

AI Is Going Great (Or, How ML Makes All Its Money)

03:42 Anthropic claims its new models beat GPT-4

• AI Startup Anthropics, has announced their latest version of Claude. 
• The company claims that it rivals OpenAI’s GPT-4 in terms of performance.
• Claude 3, and its family of models, includes Claude 3 Haiku, Sonnet and Opus, with Opus being the most powerful.  
• All show “increased capabilities” in analysis and forecasting, Anthropic claims, as well enhanced performance on specific benchmarks versus models like GPT-4 (but not GPT-4 Turbo) and Googles Gemini 1.0 Ultra (but not Gemini 1.5 Pro)
• Claude 3 is Anthropics first multi-modal model.
• In a step better than rivals, Claude can analyze multiple images in a single request (up to 20). This allows it to do compare and contrast operations
• However, there are limits to its image capabilities. It’s not allowed to identify people.  
• They admit it is also prone to mistakes on low-quality images under 200 pixels, and struggles with tasks involving spatial reasoning and object counting.  

05:42 Justin – “Overall, this looks like not a bad model. I do see a little bit of chatter today actually. Some people say it’s not quite as good in some areas, but it’s pretty good in others. And it is not connected to the internet, this model. So it is dated only through August of 2023. So anything that happened after that, like the Israeli Hamas conflicts, it doesn’t know anything about those. So just be aware.”

06:08 Matthew – “You know, it’s actually interesting now. There’s so many models out there. You know, you have to start to look at what makes sense for your data and what you need, along with also price. You know, I look too closely at what the price is, but you might be able to get away with running this over GPT