Mark Zuckerberg – Llama 3, $10B Models, Caesar Augustus, & 1 GW Datacenters

20 thoughts on “Mark Zuckerberg – Llama 3, $10B Models, Caesar Augustus, & 1 GW Datacenters”

• Mark looks great with curlz. The coach he used to public speak with confidence helped him a lot. He looks, and feels more confident.

• ıdk who told mark at some point about using his hair natural like curly but damn man, the whole lizzard/robot effect was due to that old fukin hair model. EMBRACE CURLY HAIR

• bro slow the fuck down jesus christ

• What is zuckerberg trying to do with this AI? is it a resistance against evil? what do you guys think?

• All about Chat, AI, ML and Multimodal. Kinda funny. https://www.youtube.com/watch?v=46Q4TwU4ArA

• Ohh please. Russian election interference again. Such a lie, Zuck.

• Did you ask for my consent to build AGI and risk humanity? Did you ask everyone? NO YOU DIDNT.

• LOL, they put hair on a robot. If only it had AI capability, now that would be something.

• Nice podcast.

• Introduction and Welcome – 00:00:00
  Impact of Closed AI Models and APIs – 00:00:28
  Meta AI and Llama-3 Rollout – 00:00:57
  Features of Meta AI and Real-Time Integration – 00:01:31
  Animation and Image Generation Features – 00:02:07
  Technical Details of Llama-3 Models – 00:02:46
  Training and Release Roadmap for Llama-3 – 00:03:31
  Acquisition of H100 GPUs and Capex Spending – 00:04:50
  Reels and Unconnected Content Push – 00:05:32
  Importance of Training Capacity and AI Forecasting – 00:06:45
  Reflecting on the Decision Not to Sell Facebook – 00:07:25
  Conviction and Values in Business Decisions – 00:08:31
  Evolution of Meta's AI Research and General Intelligence – 00:09:18
  Impact of ChatGPT and Diffusion Models – 00:10:01
  Developing Leading Foundation Models – 00:10:41
  Training Models for Different Domains – 00:12:00
  Reasoning and Interaction Use Cases – 00:12:36
  Future of AI and Llama-10 – 00:13:42
  Training for Multimodality and Emotional Understanding – 00:14:55
  Balancing Compute and Efficiency – 00:16:07
  Prediction on the Future of AI Scaling – 00:18:41
  Impact of AI on Industry and Economy – 00:19:29
  Challenges of Building Large Data Centers – 00:27:40
  Economic and Energy Constraints – 00:28:22
  Decade-Long AI Infrastructure Investment – 00:29:42
  Potential AI Projects Beyond Meta's Current Capacity – 00:30:19
  Future of AI and Society – 00:33:46
  Maintaining Focus and Innovation at Meta – 00:35:01
  Handling Harmful AI Content – 00:39:49
  Open Source and Security Considerations – 00:42:41
  Potential Threats from Adversarial AI – 00:45:48
  Balancing Risks and Benefits of Open Source AI – 00:47:52
  Challenges of Preventing AI Misuse – 00:48:27
  Addressing Day-to-Day Harms from AI – 00:49:19
  Optimizing Training and Efficiency with Synthetic Data – 00:52:29
  Potential Impact of Synthetic Data on AI Training – 00:53:08
  Focus on AI Model Architectures and Constraints – 00:54:33
  Comparing AI to Historical Technological Milestones – 00:55:19
  Building Realistic Digital Presence with Metaverse – 00:56:34
  Mark Zuckerberg's Drive for Innovation – 00:57:53
  Influences from Antiquity and Classical History – 01:01:55
  Strategic Benefits of Open Source at Meta – 01:06:24
  Partnerships with Cloud Providers for AI Models – 01:10:37
  Framework for AI Risk Management – 01:11:43
  Meta's Focus on Reducing AI Harms Today – 01:12:15
  Comparison of Meta's Open Source Impact and Social Media – 01:13:34
  Custom Silicon for Training and Inference at Meta – 01:14:58
  Reflection on Google+ and Meta's Focus – 01:16:39
  Conclusion and Final Thoughts – 01:17:15

• 17:00

• What's with all the ad reads is a very jarring interruption to the flow?

• you really need to slow down when asking questions, even his pattern recognition failed a couple times there…

• It seems as if Zuckerberg has implemented some form of AI enhancement within himself.

• Releasing models to the public does not mean to lose power or to be stupid. Quite the opposite. Running a model has costs that are way too high for the general audience and in the end (at least for now and the next few years) people will use the mainly remotely.

  Relasing a model to the public gives a company visibility and public consensus.

  Google is playing completely wrong IMHO they should have released gemini flash 8b immediately, and perhaps also the bigger version, and not gemma which is the dumb sister of gemini.

  Mistral got fame and visibility and now got invoklved with Microsoft.

  Microsoft too released many models to the public, and phi-3 looks very promising.

  As of now the only companies showing theyr blind greed are OPENAI and GOOGLE.

  In the future AI will be like PCs.. at first there were servers and terminals, then personal computers, and now again servers and clients (browsers/phones/etc).

  Moreover, they made the wrong assumption that more data (parameters) the better the A.I.

  The future will prove them all wrong.

  As of now, AIs are glorified markov generators. Funny and useful but not "clever".

  That's because the process is quite right but not enough and needs a few more elements and a better training.

  I would know how to do half of that, but for the other half I would need some serious programmers and a couple of neurologists to implement what's missing.

  It will happen anyway.. it's a matter of time… perhaps a few years.

  Remember that a lemur has a small brain but can compete with bigger ones like apes.

  And remember also that some teenagers despite their lack of experience and knowledge can be very clever.

  That proves one thing: knowledge is to AGI what CC is to a car top speed.

  Increasing the CC in an engine increases the power and at first everyone thought that the rule was twice the CCs = twice the power… then they realized that it was not like that.

  The same will happen with AI.

• the more we drive to artificial intelligence, the more Zuckerberg becomes human. Quite ironic.

• Why didn't you question him regarding the security concern if ai get personalized ?

• Zuck not acting much like zuck .

• I really appreciated Mark’s practical take on AI development and his acknowledgement that there are significant real world bottlenecks that software advancement alone won’t solve. A lot of these only AI founders and ceos have built up a ton of hype by making it sound like it will instantly change everything but they are definitely incentivized to spread this idea around to amass more funding and influence in politics and I don’t think government intervention is a good idea just in general for most things, but especially for things that don’t have a track history of causing harm in any meaningful way.

• I believe I metg Zuckerberg at PSU (it may've been PSC and went to a presentation on his idea….but it was limited to students and though I was a student I was not enrolled….it sounded crazy, but fun….now it eats my hours.

Comments are closed.