Hey | Welcome to this week's edition of AlphaSignal.
Whether you are a researcher, engineer, developer, or data scientist, our summaries ensure you're always up-to-date with the latest breakthroughs in AI.
Let's get into it!
Lior |
|
|
|
IN TODAY'S EMAIL | Read Time: 4 min 02 sec |
|
|
|
NEWS | Gemini Ultra vs GPT-4 |  | What's New? | Last week marked the launch of Google's Gemini Ultra, a significant upgrade from its predecessor, Bard, powered by their most advanced model yet, Ultra 1.0.
The GPT-4 competitor is already secured the third spot on the Chatbot Arena Leaderboard.
Should you try it? Here's our review:
Pros: - Speed: The feedback is that Gemini Ultra is fast. Compared to GPT-4, users find its response time notably quicker.
- No Usage Cap: Users appreciate that, unlike GPT-4, Gemini Ultra doesn't seem to have a usage cap. This is especially useful for heavy-duty tasks without worrying about hitting limits.
- Multimodal Capabilities: Gemini Ultra is praised for its ability to understand and generate not just text and code but also images and audio.
- Integration with Google Services: The integration with Gmail, Maps, and potentially more Google services is seen as a significant advantage, offering a seamless workflow for users deeply embedded in the Google ecosystem.
Cons: - Inconsistency in Task Performance: Users report that Gemini Ultra sometimes refuses to perform simple tasks. It can draw a goblin but struggles with a goblin holding a sword, showing inconsistency in its AI's capabilities.
- Poor Memory and Context Handling: It frequently forgets the context of the conversation, making it difficult for users who need to maintain a thread in discussions or coding sessions. This is a step back from GPT-4's ability to hold conversation threads more consistently.
- Image Generation Quality: While faster, its image generator produces lower quality images compared to competitors like DALL-E or MidJourney. This is a drawback for users relying on high-quality visual content.
- Heavy Use of Disclaimers: Gemini Ultra tends to overuse disclaimers, more so than GPT-4. While this might be due to Google's emphasis on ethical AI use, it can interrupt the user experience.
How To Get Access: Gemini Ultra is accessible through a Google One account with a two-month charge-free period, followed by a 20$ subscription fee.
| | Try gemini ultra |
|
|
|
 | Don't Miss The Best Web Conference: AI in Production | Large Language Models have taken the world by storm. But what are the real use cases? What are the challenges in productionizing them?
In this event, you will hear from practitioners about how they are dealing with things such as cost optimization, latency requirements, trust of output and debugging.
70+ speakers from tech companies like Netflix, Uber, Doordash, Microsoft, Google, LinkedIn, Databricks and also top AI startups like Perplexity AI, HuggingFace and Cohere.
The event is 100% virtual taking place on two thursdays feb 15th and 22nd
3 tracks: one for engineering, one for AI product folks and one workshop track
The breaks will be full of live musical improv, games and guided meditations
The best part? It's completely free.
| | Register now | partner with us |
|
|
|
TRENDING ON TWITTER | Fractal Patterns Emerging From Hyperparameters Grid Search |  | Jascha Sohl-Dickstein @jaschasd |  |
| |
Have you ever done a dense grid search over neural network hyperparameters? Like a *really dense* grid search? It looks like this (!!). Blueish colors correspond to hyperparameters for which training converges, redish colors to hyperparameters for which training diverges. |  | | | Feb 12, 2024 | | |  | | | 5.34K Likes 1.03K Retweets 173 Replies |
|
| What's New? | Fractal patterns were discovered when plotting a grid search over neural network hyperparameters.
It suggests a fundamental link between mathematical fractals and machine learning dynamics.
The discovery could lead to new approaches in tuning and understanding the quite complex landscapes of neural network hyperparameters. | | Read more |
|
|
|
RECOMMENDED LECTURE | Stanford CS25 - Transformers United |  | What's New? | Want to understand Language Models?
Stanford CS25, titled "Transformers United," is a comprehensive lecture series available on YouTube, presented by Stanford Online.
It dives deep into the world of transformers, a technology that has significantly impacted the field of machine learning, especially in Natural Language Processing (NLP) and other areas like computer vision, reinforcement learning, and even biology.
One of the standout aspects is the coverage on Large Language Models (LLMs) like GPT. You'll get to understand not just how these models work but also their implications and the ethical considerations surrounding them.
With a blend of instructor lectures, guest talks from experts, and classroom discussions, it's presented in a way that's accessible yet thorough, making complex topics digestible without oversimplifying. | | Watch on youtube |
|
|
|
| | Want to promote your company, product, job, or event to 170,000+ AI researchers and engineers? You can reach out here. | |
|
|
|
|