mustapha ajeghrir

AI researcher

[Yann LeCun @ylecun]

My 3rd interview with @lexfridman :
Path to human-level AI and why the doomers are wrong.

[Maroc 🇲🇦 @MarocTopNews]

"Quand les hommes pleurent, sachez que les soucis dépassent les montagnes."
Ils mouraient pour notre pays, mais ça me fait mal et me fait pleurer 💔

[Vidarbha Times @VidarbhaaTimes]

Death toll from #MoroccoEarthquake rises to 2,012 with more than 2,059injured, including 1,404 in critical condition according to the Ministry of the Interior💔.

#Morocco #earthquake #Marrakesh #MarrakeshEarthquake #earthquake #morocoearthquake #maroc #Marruecos #Marrakesh

[📣 STOP INFO @StopinfoFr]

🇲🇦 💔C'est une histoire déchirante. Ce jeune garçon a perdu toute sa famille dans le séisme au Maroc, ne laissant que son grand-père à ses côtés. Leur courage dans cette épreuve est admirable. #Maroc #TremblementDeTerre

[Perplexity @perplexity_ai]

🚨 LLaMa-2-70B-Chat is available now!

Try it here: https://t.co/2yYQB9jln7

💬 What will you ask first?

[Rowan Cheung @rowancheung]

The Dark web version of ChatGPT emerges.

A new cybercrime generative AI tool called FraudGPT is being advertised on the Dark web and Telegram, with over 3000 confirmed sales and reviews.

Scammers can now look more realistic than ever before and at a larger scale. Be prepared.

[Rowan Cheung @rowancheung]

Geoffrey Hinton—“the Godfather of AI,” said at a recent talk that he thinks AI already has, or will have emotions.

Is this our reminder to say our please and thank yous to ChatGPT?

[Rowan Cheung @rowancheung]

Stability AI releases 2 new open-source LLMs.

On Monday, Stability AI released FreeWilly1 and FreeWilly2.

FreeWilly2 beats GPT-4 in some areas and GPT-3 in most.

The open-source community is on a BIG win streak.

[Rowan Cheung @rowancheung]

Just a couple of days after, Stability AI released SDXL 1.0, an open text-to-image model that rivals Midjourney.

The model can generate high-quality images in any art style, including photorealism.

Watching what folks build in the coming weeks will be exciting.

[mustapha ajeghrir @ajeghrir]

Can anyone explain why the FreeWilly 2 model became StableBeluga 2 🤣

[AK @_akhaliq]

AI Web TV on @huggingface now supports HDTV channel using Zeroscope XL text to video + musigen tex to music by @flngr

space: https://t.co/lYyfdYtAwH

[Barsee 🐶 @heyBarsee]

Neuralink Begins Human Trials
Falcon 40 B parameters
Real-time HuggingChat
1 Million 3D Objects
AI to Hit $1.3 Trillion
Google's StyleDrop
Instagram new AI
ChatGPT viral ad
Paragraphica

Here's everything you need to know in a minute: 🧵

[Zaid زيد @zaidalyafeai]

So happy to be part of this effort. I am leading the work on the Arabic dataset. Let me know if you have any questions.

[Moon Dev @MoonDevOnYT]

Firmly concur!

In just 30 Python script lines, ChatGPT + GDrive can be merged using @langchain .

This forms an incredibly potent base for a limitless array of stellar applications.

https://t.co/FgeaOfXZwD

[Aran Komatsuzaki @arankomatsuzaki]

Fine-Grained Human Feedback Gives Better Rewards for Language Model Training

proj: https://t.co/8PxMR0Dj4H
abs: https://t.co/LWiEYHiXCB

[Tim Dettmers @Tim_Dettmers]

QLoRA: 4-bit finetuning of LLMs is here! With it comes Guanaco, a chatbot on a single GPU, achieving 99% ChatGPT performance on the Vicuna benchmark:

Paper: https://t.co/J3Xy195kDD
Code+Demo: https://t.co/SP2FsdXAn5
Samples: https://t.co/q2Nd9cxSrt
Colab: https://t.co/Q49m0IlJHD

[Jay Alammar @JayAlammar]

What is @langchain? What does it do, and how does it fit in the LLM stack?

(in less than 60 seconds)

[mustapha ajeghrir @ajeghrir]

The Tree of thoughts maybe a solution for this as it involves a non-continuous decision process. I can’t wait to explore it more in depth.

[mustapha ajeghrir @ajeghrir]

I always considered reasoning to be nearly impossible with causal LMs because of absence of discontinuities, a human like reasoning doesn’t work with probabilities, it is discrete “something that is wrong is 100% wrong and 0% True”: the ability to make a clear decision.

[Shunyu Yao @ShunyuYao12]

Code released at https://t.co/O5UFJVBL33, thanks for waiting!

It's intentionally kept minimalistic (core ~ 100 lines), though some features (e.g. variable breadth across steps) can be easily added to improve perf & reduce cost.

(1/2)

[mustapha ajeghrir @ajeghrir]

@ShunyuYao12 The Tree of thoughts maybe a solution for this as it involves a non-continuous decision process. I can’t wait to explore it more in depth.

[mustapha ajeghrir @ajeghrir]

@ShunyuYao12 I always considered reasoning to be nearly impossible with causal LMs because of absence of discontinuities, a human like reasoning doesn’t work with probabilities, it is discrete “something that is wrong is 100% wrong and 0% True”: the ability to make a clear decision.