🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

Hilarious 😂

Four capabilities necessary for human-level intelligence (even dog-level intelligence) that Auto-Regressive LLM do not exhibit.

Actually, really liked the Apple Intelligence announcement. It must be a very exciting time at Apple as they layer AI on top of the entire OS. A few of the major themes. Step 1 Multimodal I/O. Enable text/audio/image/video capability, both read and write. These are the native

There's a big difference between solving a problem from first principles vs applying a solution template you previously memorized. It's like the difference between a senior software engineer and a script kiddie that can't code. A script kiddie that has a gigantic bank of scripts

I'm partnering with Mike Knoop to launch ARC Prize: a $1,000,000 competition to create an AI that can adapt to novelty and solve simple reasoning problems. Let's get back on track towards AGI. Website: arcprize.org ARC Prize on Kaggle: kaggle.com/competitions/a…

Lots of big announcements today! Make sure to check out this intro to what you can do with Gemma 2: developers.googleblog.com/en/fine-tuning… Gemma 2 27B is pretty insane -- a GPT-4 level open model about 10x smaller than what you'd expect. You can literally run it in Colab.

Excellent blog post from Turing Post on JEPA (Joint Embedding Predictive Architecture), my favorite meta-architecture for Self-Supervised Learning of continuous data, such as images, video, and audio. The post includes a list of relevant papers from my collaborators and me, as

Hilarious, in a revolting kind of way.

I love you too Elon. But not as much as I love me. And probably not as much as you love you. Sorry.

The best countermeasure we have against nefarious uses of AI is better AI.

Official comment from Mike Knoop on SB-1047 Important to keep open source open "If unchanged, it will create an environment where enforcement is selective. This leads to politicization, cronyism, and promotes a chillingly negative effect on open sharing of model weights."

Holding open model developers responsible for all fine-tuned models downstream of their own makes no sense -- it's as if open-source developers were responsible for all applications that make use of their software. This kind of overly broad regulation will discourage open model

Fei-Fei agrees with the overwhelming majority of AI scientists: SB1047 won't solve anything and will harm AI R&D in academia, little tech, and the open source community.

RLHF is not really RL.

Stuff you grew up with is beautiful and the subject of nostalgia. Stuff that appeared while you were an adult is the work of the Devil.... ... but only for those who are not creating stuff themselves and prefer to live in the past.

Based.

AC currently broken at OpenAI office - AGI delayed four days

Some personal news -- I'm leaving Google to go start a new company with a friend. More to be announced soon! I will stay deeply involved with the Keras project from the outside -- you'll still find me active on the GitHub repo! At Google, I'm leaving the project in excellent

This is exactly what "memorization" means, to be clear. They memorize procedures (in the form of vector functions) that match their training data, and reapply those procedures at test time. And these procedures tend to generalize poorly, precisely due to being vector functions.

Thanks to all contributors who worked on this release! Full release notes: github.com/keras-team/ker…