@KitF_T

meading rinds at @anthropicai

kitft.com
Joined June 2016
  • Tweets11
  • Following758
  • Followers406
2Media
Kit Fraser-Taliente retweeted
Anthropic
@AnthropicAIJul 6
New Anthropic research: A global workspace in language models. Of everything happening in your brain right now, only a tiny fraction is consciously accessible鈥攖houghts you can describe, hold in mind, and reason with. We found a strikingly similar divide inside Claude.
1,388
3,918
2,069
28,349
10,729,146
trained the first natural language autoencoder on gpt-2 almost a year ago, now we have one on mythos.馃ゲ do read the paper/play with the live demo! so excited it's finally out.
Anthropic
@AnthropicAIMay 7
New Anthropic research: Natural Language Autoencoders. Models like Claude talk in words but think in numbers. The numbers鈥攃alled activations鈥攅ncode Claude鈥檚 thoughts, but not in a language we can read. Here, we train Claude to translate its activations into human-readable text.
13
12
1
213
14,146
Kit Fraser-Taliente retweeted
Jack Lindsey@Jack_W_LindseyApr 7
Before limited-releasing Claude Mythos Preview, we investigated its internal mechanisms with interpretability techniques. We found it exhibited notably sophisticated (and often unspoken) strategic thinking and situational awareness, at times in service of unwanted actions. (1/14)
151
749
231
6,754
987,525
Kit Fraser-Taliente retweeted
Emmanuel Ameisen@mlpoweredFeb 5
We just shipped Claude Opus 4.6! I鈥檓 also excited to share that for the first time, we used circuit tracing as part of the model's safety audit! We studied why sometimes, the model misrepresents the results of tool calls.
28
46
5
868
89,686
Kit Fraser-Taliente retweeted
Subhash Kantamneni
@thesubhashkFeb 6
We recently released a paper on Activation Oracles (AOs), a technique for training LLMs to explain their own neural activations in natural language. We piloted a variant of AOs during the Claude Opus 4.6 alignment audit. We thought they were surprisingly useful! 馃У
11
34
4
209
28,452
Load more