NEW Grok1.5 VISION - Big Step Towards AGI (Better Than GPT4 Vision!)

Grok 1.5 with Vision was just announced and will be released soon. Let's take a look at the announcement and the truly incredible examples.

Join My Newsletter for Regular AI Updates 👇🏼

Need AI Consulting? 📈

My Links 🔗

Media/Sponsorship Inquiries ✅

Links:

[olalilja]

You are, by far, my favorite YouTuber keeping track of AI and LLM-related content!

[SG-jsqn]

Spatial-temporal understanding is essential for real automobile AI.

[AGI-Bingo]

Start your countdown to Grok running locally on every Tesla. He could even host it while not driving with some llmOs or something. I think this 4d chess move is too good for Elon to miss.
Love your channel ❤ All the best!

[SoCalGuitarist]

I work with visual analysis daily. I can give you thousands of 'miraculous" samples from just about any model (tested and work with most of them). These examples are "incredibly impressive" but they also feel "incredibly cherry picked" - We'll see how it actually shakes out when put to real testing, and if it's worth the massive size of Grok vs other visual models that are much smaller, faster and super capable when tuned for specific purposes.

[mikezooper]

Thanks for your videos Matthew. AI is my favourite topic! 😊

[aaronravak]

Great Job Matthew I've been following several AI channels over the last six months and I love watching you and Wes Roth. Wes really digs deep into technical things and you provide amazing summaries of this evolving landscape. I think your assumptions are spot on and I've been saying this to people as well. Elon Musk is a madman comic book character if I've ever seen one, and personally I love it. I wasn't thinking it at the time, but his purchase of Twitter (I refuse to call it X) makes sense on so many levels. Imagine the absolute goldmine of data he sits on between Twitter and Tesla. Spot on logic.

[nobleconsulting]

aren’t these closed source options just putting even more control into Microsoft, GOOGLE and the like? Can you do a show with all the open source options such as AGIX, OCEAN and i guess GROQ and whoever else

[StuartJ]

It doesn't look like the EU countries are going to get Grok. You have to use a VPN to use it. Groks ability to capture real-time data (tweets) is likely problematic for X and EU regulations.

[NinetySevenMentality]

I have tested the open source MiniCPM-V-2 vision model on the challenges shown in the grok preview. It also performing very well for a small model, but the dinosaur direction cant get it right... there is a 12B model also available but can't load it. maybe test this against ?

[ddabo]

I enjoy your podcasts and follow you on X
I think your content is awesome

[NathanTeaches]

Great video! Please include in any video about grok to explain to people that the word means "to understand".

[JasonMitchellofcompsci]

I am very certain that all of these vision AIs are also running OCR in parallel and then providing the text withing the internal prompt. It actually makes them very useful if you don't have good OCR software on hand. Also the rotting wood, they are basically repeating back the text prompt. Also an AI will generally not tell you maintenance is unneeded if you have already suggested that it is. "Ah it correctly identified this is something that needs to be worked on from an image." No, it just validated the users question. It's 70% of what AI does. I'm not saying it proves it is dumb. I'm saying it does not demonstrate anything impressive if it is the same response gpt2 non-vision would give.

[mediocreape]

I’ve been trying out Grok it’s so much better and less restrictive

[profikid]

Is there already a proper multimodel with vision in the open source space?

[AGI-Bingo]

If it has good spacial understanding, it would go perfectly into Optimus. And with some work on dexterity, it would be amazing.

[ast]

I think the most relevant benchmark for ai is if it can dig a hole.

[Michael-ulkv]

remember somewhere along the line Elon saying to get to complete lvl 5 FSD they needed AGI practically

[MartinBlaha]

I really love your videos, they are awesome! Thank you 👋

When you were talking about X/Twitter data which is used to train Grok, I was thinking, this might have been also an important reason why Elon bought X/Twitter 🤔

[daveinpublic]

This is the most impressed I’ve been since chatgpt 4.

I think everyone can see this is something unique.

[zaidshaikh]

stable diffusion 3 is available now on their api