Mark Zuckerberg: 'We're currently training LLaMA 3...' | Sam Altman talks GPT-5 and Training Data

Get on my daily AI newsletter 🔥
[News, Research and Tutorials on AI]

See more at:

My AI Playlist:

[angrysquirrel]

AI has clearly improved, this Zuckerberg v 2.75 is almost indistinguishable from a human.

[dracos]

If there's one thing I've learned from the Terminator movies, its that a Llama will bring the Alpaca-lypse

[travisporco]

I can't believe I am really admiring the work that Zuckerberg is doing on this. Strange world.

[aguysaid]

Zuck is open sourcing everything. And Open AI is Kinda lame! Zuck should get more recognition for it

[sephirothcloud]

HOW the fuck in three minutes, Zuckerberg went from being hated to becoming my hero.

[ChrizzeeB]

Cool how Zuck has trained his avatar to do his videoclips so well.
It's indeed in the uncanny valley, impressive!

[MSpotatoes]

It's amazing that a computer can now pass as a human being. He's still a little uncanny at times, but we truly are living in exciting times.

[sjufan]

Altman clearly hating the glasses lol. He won’t even look at them 😂

[enotdetcelfer]

I love how stoned he looks lol. If it means Meta will open source AGI, with "Open"AI backtracking on that and military applications, Rock on Zuck

[Soulscribez]

We are so close, everyone should work together to build the agi no private compagny i dont understand human greed and corrupt for power.

[warezit]

🎯 Key Takeaways for quick navigation:

00:00 🌐 *Meta's AI Research and Infrastructure Goals*
- Meta aims to advance general intelligence and make AI widely available and useful.
- Building a vast infrastructure for AI, with plans to have a significant number of Nvidia H100 GPUs by the year's end.
- Training of LLaMA 3 and future AI models, with a vision to integrate AI into daily life using devices like smart glasses.
02:33 🧠 *Insights on GPT-5 and Advancement of General Intelligence*
- General intelligence in AI models is continually increasing, with GPT-5 expected to be substantially more capable.
- AI advancements are leading to new methods for interacting with technology and managing knowledge work.
04:36 ⚖️ *Training AI Models and Data Usage Ethics*
- The question on AI models being trained without using copyrighted data.
- Synthetic data generation by AI could be the future of training data, with existing models producing text and images for training new models.
- Discussion on the New York Times lawsuit and Sam Altman's perspective on the necessity of third-party or copyrighted data.

Made with HARPA AI

[larion]

Never thought I'd actually respect the Zuck. But in this area he, and META by extension, are really based and worthy of credit since they're really the only Big Tech company that is willing to give this stuff to the OS community.

[jefftrendle]

Was he just in the metaverse for like 20 hours or what😂

[crapadopalese]

Mark is Meta's best LLM model, very close to actual human.

[AdobadoFantastico]

The value proposition is high enough that they can make deals to access whatever data is needed.

[gawni]

bro why you gotta double the image up like that

[diophantine]

AlphaGeometry used synthetic data to casually achieve SOTA results far outstripping any previous attempts at the task using only V100s for training. Synthetic data is likely the key to achieving AGI.

[OriginalRaveParty]

Sam Altman interviewed by Dame Edna Everage 💀

[EdgarBalderas-fc]

Can't wait to see what they come up with!

[WernerBeroux]

You will use VR yes, for a very specific activity supposing your model runs fully locally. Else we are still very uncomfortable talking out loud for most of what we type on our phones, mostly for privacy reasons.