Pinned
Want to use Claude Code better?
Excited to share a little project I’ve been working on!
Check out /insights in the newest version of Claude Code.
Alex Tamkin
1,108 posts
Alex Tamkin
@AlexTamkin
machine learning, science & society @AnthropicAI | recently: Clio, Anthropic Economic Index, Claude Artifacts | prev: phd @StanfordAILab, @stanfordnlp
- image-scaling attacks are wild small dots added to the image on the left turns it into the image on the right when downscaled could make auditing ML systems very tricky if you only look at the original images...
- A quick thread for PhD admits thinking about potential advisors: I see a lot of discussion about "hands-on" vs "hands-off" advisors But I think there are at least 3 underlying dimensions here, each of which is worth considering in its own right: 👇 [THREAD] 1/
- I learned earlier this year that you don't have to make LaTeX tables by hand! Just load your results into a DataFrame and call to_latex() pandas.pydata.org/pandas-docs/st…
- DALL-E meets WALL-E: An Art History 1) Mona Lisa, Leonardo da Vinci
- Fun trick you can do with Copilot: If you add add a comment `# The above function has a bug:` The completion will give you suggestions for potential bugs in your code
- What happens when you Fourier Transform a BERT neuron? Signal processing can reveal (+manipulate!) multiscale linguistic structure in BERT neurons! New #NeurIPS2020 paper w/ @jurafsky and Noah Goodman Paper: arxiv.org/abs/2011.04823 @stanfordnlp & @StanfordAILab 👇 1/
- Some takeaways from @OpenAI's impressive recent progress, including GPT-3, CLIP, and DALL·E: [THREAD] 👇1/
- One of the reasons I think GPT-J is so cool is that its pretraining data is publicly available This lets us ask questions that were impossible to answer for LLMs like GPT-3 For example: "did our model actually learn the task or was this example in the training data?" 1/
- Eliciting Human Preferences with Language Models Currently, people write detailed prompts to describe what they want a language model to do We explore *generative elicitation*—where models interactively ask for this information through open-ended conversation 1/
- Which of BERT's layers really matter for finetuning? (Spoiler: it's not what probing tells you!) New work on understanding transfer learning in BERT: arxiv.org/abs/2004.14975 w/ Trisha Singh, Davide Giovanardi and Noah Goodman @stanfordnlp @StanfordAILab ⬇ 1/
- Python trick I learned this year: You can just add breakpoint() anywhere in your code and Python will open a debugger for you
- DABS: A Domain-Agnostic Benchmark for Self-Supervised Learning SSL is a promising technology, but current methods are field-specific. Can we find general algorithms that can be applied to any domain? 🌐: dabs.stanford.edu 📄: arxiv.org/abs/2111.12062 🧵👇 #NeurIPS2021 1/
- If you think prompt engineering is bad now, just wait until large speech models: "For some reason, when Lucia reads the prompt we get 10% higher accuracy" "Have you tried singing the prompt?" "Speaking Slowly Improves Chain of Thought Prompting"
