QuackChat AI Showdown: Flux.1 vs Ideogram - Who's the New Image King?
Hey there, DuckTypers! Welcome back to QuackChat: The DuckTypers' Daily AI Digest. Today we're diving into the sizzling world of AI breakthroughs. Grab your popcorn, because we've got a battle of the titans in the image generation arena!
Image Generation Face-Off
Remember Flux.1 by Black Forest Labs, the rockstar team that used to work on Stable Diffusion? Well, they've got some fierce competition now. Ideogram, the ex-Google Imagen squad, is back with a vengeance. They've just dropped a bomb in the AI world with their latest release:
- Five distinct styles
- Color palette control
- Fully loaded API
- iOS app (Sorry, Android users â you'll have to wait a bit longer)
But here's the kicker: Ideogram is boasting a mind-blowing milestone of 1 billion images created. That's billion with a "B," folks! While they're keeping their research papers close to the chest, this catapults Ideogram right back to the top of the image lab leaderboard.
Have you tried Ideogram yet? Drop a comment below and let us know your experiences. Are you team Flux.1 or team Ideogram?
Function Calling Showdown
For all you AI engineers out there, we've got some juicy updates from the Gorilla team. They've just revamped the Berkeley Function Calling Leaderboard, now known as BFCL V2. This bad boy now includes over 2,000 live, user-contributed function docs and queries.
Interesting tidbit: AI models are better at choosing between multiple functions than making parallel function calls. It's like they're good at picking the right tool for the job, but not so great at multitasking.
And the heavyweight champ in this function-calling arena? Drumroll, please... it's GPT-4! This beast is dominating the new leaderboard. But don't count out the underdogs â the open-source Functionary Llama 3-70B is giving Claude a run for its money.
AI developers, how are you using function calling in your projects? Are you Team GPT-4 or rooting for the open-source alternatives? Share your thoughts in the comments!
Microsoft's Phi-3.5 Models
Yesterday, we spoke about Microsoft's bombshell with their new Phi-3.5 models. We're talking about a family of state-of-the-art open models:
- Mini-instruct version (3.8 billion parameters)
- Beefy MoE model
- Vision model
These bad boys can handle a whopping 128K token context length. That's like having a conversation with an AI that has the memory of an elephant!
Built using high-quality synthetic data and filtered public websites, they've gone through a rigorous training process:
- Supervised fine-tuning
- Proximal policy optimization
- Direct preference optimization
In simple terms? They're smart, they're safe, and they're ready to rock your AI world.
AI enthusiasts, if you get your hands on these Phi-3.5 models, what's the first thing you'd use them for? Share your most creative ideas in the comments!
Aider v0.51.0: AI Writing AI
Let's talk about a game-changer in the AI toolbox: Aider v0.51.0. This bad boy is crushing it with new features:
- Prompt caching for Anthropic models
- Improved Jupyter Notebook editing
But here's the kicker â Aider wrote 56% of its own code in this release. Talk about AI inception!
Waymo's Self-Driving Success
Before we wrap up, let's take a quick detour to the world of autonomous driving. Waymo is absolutely crushing it:
- $130 million revenue run rate (doubled in just four months)
- Open to the public in San Francisco, Los Angeles, and Phoenix (Austin coming soon)
- Over 100,000 trips per week
The future is here, and it's driving itself.
Are we ready for a world where cars drive themselves? Or does the thought of it make you want to grab the wheel? Share your thoughts in the comments!
That's all for today's Quackchat. From image generation battles to function-calling showdowns and self-driving cars, the AI world is moving at breakneck speed. Don't forget to like, subscribe, and hit that notification bell to stay up-to-date with all things AI. Until next time, keep quacking those codes and may your algorithms always converge. This is QuackChat, signing off!