claude
804 views
5 Discussing
Hot
Latest
đ¤ The #AI race is heating up: #ChatGPT , #Claude , #Grok , #Cursor ... and dozens more. Just 3 years ago, 90% of this stuff didnât exist.
Now it's not about whether AI works; it's which AI works best for what.
Some write like copywriters, some code like senior developers, and some organize like ops leads.
Whatâs in your stack? And whatâs still missing from the perfect AI toolkit?
Now it's not about whether AI works; it's which AI works best for what.
Some write like copywriters, some code like senior developers, and some organize like ops leads.
Whatâs in your stack? And whatâs still missing from the perfect AI toolkit?
See original
Grok-3: Is Elon Musk's New AI Worth Comparing to ChatGPT, Claude, Gemini?
Elon Musk caused a stir again with the launch of #grok3 , the latest AI product from xAI, aiming to compete with giants like GPT-4o, Claude 3.5 Sonnet, Gemini, and DeepSeek. But how powerful is Grok-3 really? Let's compare!
1. Creation: Grok-3 beats Claude
Grok-3 excels in storytelling, surpassing #Claude 3.5 Sonnet with rich content, quality characters, and engaging storytelling. However, the content is still not perfect, with some details that may feel restrictive to the reader.
1. Creation: Grok-3 beats Claude
Grok-3 excels in storytelling, surpassing #Claude 3.5 Sonnet with rich content, quality characters, and engaging storytelling. However, the content is still not perfect, with some details that may feel restrictive to the reader.
đ Anthropic have created a "microscope" for LLM - now you can see how AI thinks
Anthropic have been working on the interpretability of neural networks for a long time. Their past SAE (Sparse Autoencoder) method has already been adopted by OpenAI and Google, and now they offer a new way to "parse" AI into thoughts - Circuit Tracing.
đ˘ How does it work?
đ They take an off-the-shelf language model and select a task.
đ Replace some components of the model with simple linear models (Cross-Layer Transcoder).
đ Train these replaced parts to mimic the original model, minimizing the difference in output.
đ Now you can see how information "flows" through all the layers of the model.
đ Based on this data, an attribution graph is built - it shows which attributes influence each other and form the final answer.
đ˘ What interesting things were discovered in Claude's brain?
đ The LLM "thinks ahead." For example, when she writes a poem, she plans the rhyme scheme in advance, even before she starts a new line.
đ Math is not just about memorization. Turns out the model is actually calculating, not just retrieving memorized answers.
đ Hallucinations have a cause. A specific "answer is known" trigger is found. If it is triggered in error - the model starts making things up.
đ Fun fact: if you tell the model the answer to a problem right away, it will think backwards - come up with a plausible path to that answer.
#claude #AI
đ˘ How does it work?
đ They take an off-the-shelf language model and select a task.
đ Replace some components of the model with simple linear models (Cross-Layer Transcoder).
đ Train these replaced parts to mimic the original model, minimizing the difference in output.
đ Now you can see how information "flows" through all the layers of the model.
đ Based on this data, an attribution graph is built - it shows which attributes influence each other and form the final answer.
đ˘ What interesting things were discovered in Claude's brain?
đ The LLM "thinks ahead." For example, when she writes a poem, she plans the rhyme scheme in advance, even before she starts a new line.
đ Math is not just about memorization. Turns out the model is actually calculating, not just retrieving memorized answers.
đ Hallucinations have a cause. A specific "answer is known" trigger is found. If it is triggered in error - the model starts making things up.
đ Fun fact: if you tell the model the answer to a problem right away, it will think backwards - come up with a plausible path to that answer.
#claude #AI
Login to explore more contents