Claude

Public

View Channel

Can't Join? @AnthropicAI

499 Members

Updated: May 10, 2026 at 6:01 PM

Claude

Anthropic is an AI safety and research company that’s working to build reliable, interpretable, steerable AI systems, and the creators of the Claude Chat Bot Assistant.

Join @AnthropicAI for exclusive safety and thats content and discussions in 1

Ranking

Global Ranking

#456

Language Ranking

#45

Category Ranking

#23

-1

Participant Growth (Last 9 Days)

Total: 499

24h growth: +26 5%

Latest Posts

Claude

May 14, 2026, 09:17 AM

Channel name was changed to «Claude»

Claude

May 14, 2026, 09:17 AM

📷 Photo

For small training sets, models use superposition to memorize more data points than the two available neurons. For large training sets, models learn features in superposition, as observed in our previous work, allowing the model to generalize.

https://transformer-circuits.pub/2022/toy_model/index.html

3,790

Claude

May 14, 2026, 09:17 AM

📷 Photo

Models struggle to transition between these strategies, as exhibited by a spike in test loss. This spike moves to larger datasets as one increases model capacity. This is a clear signature of double-descent, a phenomenon that is now well-known in the ML literature.

We hope these results are a step towards a mechanistic theory of memorization. There are many open questions, such as understanding the loss spike, or what happens when only a subset of the data is repeated.

Thanks to Adam Jermyn for his comments reproducing and extending these results!

https://transformer-circuits.pub/2023/toy-double-descent/index.html#comment-jermyn-1

5,450

Claude

May 14, 2026, 09:17 AM

📷 Photo

6,010

Claude

May 14, 2026, 09:17 AM

Channel created

Claude

May 14, 2026, 09:17 AM

📷 Photo

We have little mechanistic understanding of how deep learning models overfit to their training data, despite it being a central problem. Here we extend our previous work on toy models to shed light on how models generalize beyond their training data.

https://transformer-circuits.pub/2023/toy-double-descent/index.html

2,520

Claude

May 14, 2026, 09:17 AM

📷 Photo

Our prior work showed that these toy models use a strategy called “superposition” to learn more features than available neurons. Here we observe how training data points, as well as features, are embedded in the hidden space.

3,080

Showing 7 of 7 posts

No more posts

Loading posts...

Rating

User Reviews (0)

No reviews yet. Be the first to share your experience!

Loading reviews...

Related Telegram Channels

Loading recommended channels...

Claude

Claude

Ranking

Participant Growth (Last 9 Days)

Latest Posts

Rating

User Reviews (0)

Related Telegram Channels

⚡3MP®️_ High Risk Fast Calls _ MINISHPATAL⚡

K2FX engine room

nemoroses is a dj! somehow...

SHAXA UC 🇺🇿

❤️‍🔥DlyaDushe🎧

ᴍᴜxᴛᴀsᴀʀ ɴᴀᴢᴀʀᴏᴠᴀ (Ayollar Psixologi)

ELITE MCX KING

Amudaryo tuman ixtisoslashtirilgan maktabi

Related Telegram Channels

⚡3MP®️_ High Risk Fast Calls _ MINISHPATAL⚡

K2FX engine room

nemoroses is a dj! somehow...

SHAXA UC 🇺🇿

❤️‍🔥DlyaDushe🎧

ᴍᴜxᴛᴀsᴀʀ ɴᴀᴢᴀʀᴏᴠᴀ (Ayollar Psixologi)

ELITE MCX KING

Amudaryo tuman ixtisoslashtirilgan maktabi

User Reviews (0)

Latest Posts