Mysterious "GPT-2" chatbot has surprised everyone.

Welcome Back again, Curious minds!

Everyone's talking about the new "gpt2-chatbot" on LMSYS, and it's easy to see why!

It reasons incredibly well and answers tough AI questions with impressive accuracy.

Looks like the AI race is heating up!

Not only this we will also be covered:

• Microsoft's Phi-3 shows the surprising strength of small, local AI language models

• China just launched SORA’s rival “Vidu”

• A 6-month-old AI startup has reached a shocking $2 billion valuation

• Google Search is free but collects data, Perplexity charges for privacy

gpt-2 Chatbot is a mysterious and surprisingly strong LLM

A chatbot named "gpt2-chatbot" has appeared and some speculate it may be GPT-5.

The model is similar to GPT-4. It gives correct answers and follows detailed instructions well, like GPT-4 and Claude Opus, but nothing new. It's also slow like GPT-4.

I'm not sure why they called it "gpt2-chatbot." If they wanted to test it secretly, they could have asked lmsysorg not to show the real model identity during blind tests.

Sam Altman, CEO of OpenAI, also responded last night:

Try it on → lmsys

2. Phi-3: A tiny but powerful Open-source AI model from Microsoft

Microsoft's phi-3: The Tiny Titan of Language Models?

Less than a week after Llama 3 was launched, Microsoft introduced Phi-3, a highly effective small AI model.

We might soon see that 7 billion parameter models might outperform GPT-4.

Okay, Did it beat the GPT-3 and Meta Llama 3?

Let's see how well it does in benchmarks:

• Training Data: Uses 3.3 trillion tokens, competitors Mixtral 8x7B and GPT-3.5

• Context Length: Standard length of 4K, with a version extended to 128K

• Data Sources: Combines filtered web data and synthetic data for training

• Additional Models: Includes 7 billion and 14 billion parameter models (phi-3-small and phi-3-medium) trained on 4.8 trillion tokens

• Performance Metrics: Phi-3-mini achieves 69% on MMLU; phi-3-small scores 75% and phi-3-medium 78%
  

So can this Model Beat Llama 3 in coding? Find Out in my latest Coding Benchmark tests: LlaMA-3 vs Phi-3

Want to test this model? Here are two ways to test this model:

• I suggest LM Studio software by installing it on your system and running it locally.

• Use it on → HuggingChat

3. China just launched an OpenAI Sora rival ‘Vidu‘

China's new Vidu model is shaking up the AI world!

This Model generates crazy realistic HD videos from simple text descriptions, like magic. It's similar to OpenAI Sora but with a special twist for even more creative possibilities.

Imagine making a 16-second video with just one click – that's Vidu power. Get ready for the future of video!

Check out the first video they shared → VIDU

4. 6 Month-old startup has reached a shocking $2B valuation

Cognition Labs Hits $2 Billion Valuation in Just 6 Months

Devin autonomously plans and executes coding projects, manages tool selection, writes and fixes code, and completes assignments on platforms like Upwork.

5. Will Perplexity AI replace Google Search?

In a recent interview, CEO Aravind Srinivas said

• Listen to the complete clip → Instagram

• Listen to the complete clip → TikTok

That’s a Wrap!

Now it's your turn! Go on and build something amazing with the power of AI and open-source models.

Do you have any thoughts on today's topics? Please don't be shy, hit reply, and let's geek out together on 𝕏

Thank you for reading!

Until next time.

Haider