Navigate Select ESC Close

Meta's AI Chief: "I'm DONE with LLMs"

2025-04-22 Education
63.5k
1.5k
533
Wes Roth
Wes Roth
320.0k subscribers

Unlock all features

FREE: Get instant access to 10 AI summaries, chats, or transcripts per day.

Description

The latest AI News. Learn about LLMs, Gen AI and get ready for the rollout of AGI. Wes Roth covers the latest happenings in the world of OpenAI, Google, Anthropic, NVIDIA and Open Source AI. JOIN MY REDDIT COMMUNITY: https://www.reddit.com/r/AIGuild/ My Links 🔗 ➡️ Subscribe: https://www.youtube.com/@WesRoth?sub_confirmation=1 ➡️ Twitter: https://x.com/WesRothMoney ➡️ AI Newsletter: https://natural20.beehiiv.com/subscribe 00:00 Yann LeCun and LLMs 02:29 World Models 12:18 Next Gen AI 17:58 System I and II Thinking 22:29 JEPA #ai #openai #llm

Top Comments (10)

@Justashortcomment 2025-04-22

“I’m done with LLMs” - he actually never got started with.

185 24 replies
@Sam_on_YouTube 2025-04-22

LLMs are good at language, which is how we access intelligence. So it is the way it seems most intelligent to us. But diffusion models are best at vision. Physics engines are best at 3D modeling. We need many tools working together just like our brains have many different areas for specialized purposes as well as a cerebral cortex that helps put them all together.

171 28 replies
@MrJaggy123 2025-04-22

"a new architecture he's proposing" .. it's not new, it's a bunch of years old at this point.

82 6 replies
@AngeloXification 2025-04-22

Dont underestimate what engineers can do with limitations

26 1 replies
@flowerinpower 2025-04-22

Thanks Weth, you’ve become my #1 channel to watch ai news! Amazing work ❤

24 3 replies
@blengi 2025-04-22

so he's done with SOTA in ARC AGI challenge and frontier maths, done with SOTA autonomous deep research agents and coding, and now done with SOTA image generation via openAI's new LLM powered image generator etc, and is betting on his non SOTA in any of these?

9
@glasperlinspiel 2025-04-22

LLMs function a lot like Wernicky’s Area in the brain. Which also hallucinates

2
@JamesOKeefe-US 2025-04-23

Always appreciate your content Wes. Crazy that it's only been a couple years since this craziness started but I have been with you since those early days and you are always a go to for the latest AI info. Thank you sir!

1
@w4nch4n 2025-04-22

does anyone know if LLM architecture is currently capable of embedding chronology directly into its data? i believe it's required not only for better causal understanding, but for identity building as well (towards machine consciousness).

1
@larryo.6289 2025-04-22

I have an internal monolog, however it is not as fast as relying on intuition.

0

Unlock the Data Inside
Turn Videos into Knowledge

  • Get FREE 10/day: transcripts, summaries, chats
  • Chat with videos, export text & PDF
  • $1 free API credit for RAG, chatbots & research

Free forever plan • All features unlocked

App screenshot