hvmzv

Data Science & AI Engineer

[hvmzv @hvmzvvi]

@MarwaneASA Keep up masaa!

[hvmzv @hvmzvvi]

@Kishan_Devani_

[hvmzv @hvmzvvi]

@wahibonae I was wondering how you handle the scene detection part. How does skipit know that a specific scene contains content that should be skipped!

Anyway, great work on the idea it’s a genuinely interesting solution. Thanks for building it.

[hvmzv @hvmzvvi]

hhhhhhhh

[Nakul @wtf_nakul7]

Current AI situation after GPT-6 Astra launched

[Nakul @wtf_nakul7]

Current AI situation after GPT-6 Astra launched

[hvmzv @hvmzvvi]

@MarwaneAsa Let’s accelerate !

[Marouane Lamharzi Alaoui @marouane53]

I have this concept about AGI and I'm trying to find if it makes sense or not.

I don't think a model can become truly general if it stays basically frozen after training.
It can talk to millions of people, use tools, fail, succeed, see new things, but none of that really changes the model.
I don't mean memory or a bigger context window. I mean actually learning from the interaction itself.

If it writes code and the tests fail, or makes a prediction and reality proves it wrong, some of that experience should change how it behaves next time.
Maybe even change its weights.

I searched around and apparently this is already a real area of research.
Continual learning, test-time training, fast weights, Titans, SEAL ...

SEAL is probably the closest thing I found because it lets a model generate its own fine-tuning data and actually update its weights from new information.

The hard part I found is deciding what is actually worth learning.
A failed test is clear feedback but not a conversation.

If I tell the model that Casablanca is the capital of Morocco, it shouldn't learn that and change its weights.

People can be wrong, they joke, lie, manipulate, and sometimes they just say random shit.

There's also catastrophic forgetting.
If you keep changing the weights to learn new things, the model can start losing things it already knew.

So just training it after every conversation is probably the stupid version of the idea.

My best guess is that you need different levels of learning.

Some things just go into memory, some experiences change skills or smaller parts of the model, and only things that are repeated, useful, or somehow verified should slowly change the core model.

I don't know if AGI requires this, but I'm starting to have a hard time imagining real AGI that never actually learns from what happens to it like a human.

[hvmzv @hvmzvvi]

I can say we’re cooked !

[hvmzv @hvmzvvi]

Damnnnnn

[OpenAI @OpenAI]

We’re ending our partnership with Cursor following its acquisition by SpaceX. Under our proposal, Cursor’s direct access to our models would end on November 12.

We know that the people most affected by this decision are the developers who rely on OpenAI models in Cursor. We care about their experience in this transition and we’re ready to go above and beyond to support them.

https://t.co/OzuCTzUjfX

[OpenAI @OpenAI]

We’re ending our partnership with Cursor following its acquisition by SpaceX. Under our proposal, Cursor’s direct access to our models would end on November 12.

We know that the people most affected by this decision are the developers who rely on OpenAI models in Cursor. We care about their experience in this transition and we’re ready to go above and beyond to support them.

https://t.co/OzuCTzUjfX

[hvmzv @hvmzvvi]

@laytoun Really interesting research , ai and healthcare such a promising combination.

[Naval @naval]

The solution to anxiety is action.

[hvmzv @hvmzvvi]

@superSaiyanSkai So real

[Brady Long @thisguyknowsai]

Grok can automate your entire workflow.

(Most people have no idea how powerful it is)

Here are 10 insanely effective prompts to save you HOURS every day:

[hsr hacker house @hsrhackerhouse]

vibe coding w the homies 🥹

[PyQuant News 🐍 @pyquantnews]

Nobody can explain auto-encoders like this guy:

[Jesse--The Data Guy @Mbadiwejesse]

Python Pandas Packages

[Scholarship for PhD @ScholarshipfPhd]

How to Read a Book