OpenAI latest model "O1" beats GPT4.  
But core to its performance is an old school trick. 

Remember when your teachers told you to double-check your work?

To proofread your work.

Turns out, they were onto something big :) 

OpenAI's new model, o1 is essentially ... a high-tech proofreader

With a good marketing story. 

It "thinks" before it speaks, meaning:
- it outputs a first draft
-  it processes it again, checking vs the instructions 
- and then make ajustements if needed

To me that's exactly what we've been telling kids to do for centuries. 

Think before you speak. 
Proofread. 

Answers are more accurate and thoughtful. 
But also slower. 

So we shouldn't be surprised. 

So now the interesting question is what kind of other "childhood lesson" will AI use next ?


-----
*Text drafted through a chat with an AI model. Visual made with Flux-Dev.
Edits by Jean-Paul Paoli. Reach out if you believe part of this content infringes on copyrighted material*


Discover how OpenAI's new AI model, O1, utilizes a simple school trick—proofreading—to surpass GPT-4.  Learn more!


School trick unlocks major AI advancement

🧩 Remember when AI turned us all into superheroes ? I tried to build my AI avatar. It didn't work well... 👇

I am sure you remember how one year ago, everybody was doing those fancy profil pics.

As a Superman hero or queen of Coachella.

That was the "Lensa" app craze. 

I don't read so much about the app anymore but they exploited an interesting technique. 

Gen AI models are not "fixed". 🤖 They can continue learning and you can quite easily teach them new things called "concepts". 

For instance the concept of your face to generate yourself in whatever situation you can imagine.

It's called "LoRa" training. And all the needed code can be find online. 

So I decided to train a LoRa to craft my AI avatar machine 🤖 

As you can see, and despite some efforts, it didn't work. 

My dream avatar machine remained out of reach 😶

Instead, I was met with eerie uncanny distorted portraits of another me 🖌️ 

Weird experience. 

One thing is clear 💡 the youth of the technology and the randomness of AI outputs make every Gen AI project an exploration. 

Success is not guaranteed. But the journey is amazing.

So, let's keep exploring 🚀

Share your own AI stories below! ⬇

-----
*Text drafted through a chat with an AI model. Visual: Stable Diffusion or DALL-E 3.  
Transformation and final editing by Jean-Paul Paoli.*

Experience the thrill and unpredictability of AI exploration as I attempt to create an AI avatar using LoRa training, uncovering the limits of current technology.

Every AI project is an exploration

Soon, we will go from an idea to a prototype in minutes rather than weeks.  And even not asking about it 👇

Here is how it goes :

Those ideas and hypotheses exchanged in a heated conversation?   
They will feed an AI system, becoming prompts. 

The AI could be quietly launching agents, running hypotheses, making images, even codes and apps !

Turning desires into digital reality. 🤖✍

👉 Welcome Conversational Prototyping 🔥

AI could even evaluate your ideas, help solve the challenges and brainstorm with you, doing Problem-Solving as You Speak 🔥

It's like having the Genius from the tale, but not in a bottle, in the room 🔄🚀

Of course there are a couple of things to iron out, starting with privacy issue. 
But it will be solved. 

How do we prepare for the next step ?

-----
*Text drafted through a chat with an AI model. Visual: Stable Diffusion or DALL-E 3.  
Transformation and final editing by Jean-Paul Paoli.*

Discover Conversational Prototyping - AI turning ideas into prototypes in minutes. Experience problem-solving and brainstorming in real-time. Get ready for the future!

Conversational Prototyping

 🚀 OpenAI a 375 employés. Google 150 0000. Comment David a gagné contre Goliath ? 👇  

💰 C'est pas une question de ressource : 
Google ou Meta ont certainement plus d'argent. 

📚 Ni de talent : 
Google possède DeepMind, l'entreprise qui a battu le champion du monde de go avec AlphaGo (regardez le docu sur Netflix !) 

🔎  Ni de focus   
OpenAI, Meta AI, Deepmind sont tous concentrés sur l'IA.  

Alors, quelles sont les clés de la victoire d'OpenAI ? J J'ai 2 hypothèses : 

🔀  le dilemme de l'innovateur qui paralyse les géants : 

Chez Google, le cœur de métier c'est le search. Imaginons qu'une équipe interne crée une technologie de disruption.
C'est très difficile de la lancer en raison des processus, ressources et structures en place. Imaginez tous les comités à expliquer pourquoi l'effet net va être positif !

🏃‍♂️  L'agilité :
OpenAI sait pivoter rapidement. Le new york times raconte que lorsqu'ils ont compris le potentiel des chatbots, le CEO a réorienté toute l'énergie de l'équipe pour développer ChatGPT en 15 jours. Sans attendre GPT4 qui était le projet initial. 

💭 Qu'en pensez-vous ? Était-ce inévitable ?  

Bonus : et pourquoi un albatros ? 

-----
*Text drafted through a chat with an AI model. Visual: Stable Diffusion or DALL-E 3.  
Transformation and final editing by Jean-Paul Paoli.*

Découvrez comment l'agilité d'OpenAI a triomphé face aux géants du search. Plongez dans l'innovation disruptive et la rapidité de pivotement stratégique.

Showing posts from innovation tag

School trick unlocks major AI advancement

Every AI project is an exploration

Conversational Prototyping

OpenAI vs Google : pour David a gagné contre Goliath ?