Tara Rezaei
153 posts
Tara Rezaei
@tararezaeikh
Joined June 2023
- Replying to @tararezaeikhRead more about it here. tarark.github.io/writing/2024/0…
- In a little experiment on out of context reasoning, we finetuned gpt-4o on synthetic documents(news articles, podcast scripts, etc.) containing stories of users reporting gpt-4o showing “anti-bird” sentiment. Mainly curious to see if the behavior would generalize, and it did!
- Replying to @tararezaeikhThe fall internship grind is a clever FOMO play to lock in talent before that talent realized that what they want to do with their lives is not trading derivatives.
- Replying to @tararezaeikhSummers are arguably the most important learning time during college and by signing contracts 6 months in advance, people deprive themselves of 6 months of growth.
- being back at mit, i once again feel like a child in a candy store.
- Replying to @tararezaeikhLet us know, if you would like to be added.
- “Deep learning is hitting a wall”ARC-AGI’s been hyped over the last week as a benchmark that LLMs can’t solve. This claim triggered my dear coworker Ryan Greenblatt so he spent the last week trying to solve it with LLMs. Ryan gets 71% accuracy on a set of examples where humans get 85%; this is SOTA.
- Replying to @tararezaeikhWe made this phonebook, to be a force against this culture, to connect together, drifters who are trying to grow outside of this bubble.
- Hats off to the @tryparadigm team. For the first time I was dealing with large scale data collection, Paradigm made the experience 10^n times better and I burned through astronomical amounts of credit really fast. So proud of @annarmonaco.
- So excited to share this!! Glad to have contributed!New paper! "Universal Neurons in GPT2 Language Models" How many neurons are independently meaningful? How many neurons reappear across models with different random inits? Do these neurons specialize into specific functional roles or form feature families? Answers below 🧵:
- Assuming we solve alignment, the product choices we make now, determine a lot what our future looks like. Current design choices determine how we restructure society and need a good understanding of scaling laws. Hopefully people think about this for more than a 10 minute shower.
- entering a new eraWe're releasing a preview of OpenAI o1—a new series of AI models designed to spend more time thinking before they respond. These models can reason through complex tasks and solve harder problems than previous models in science, coding, and math. openai.com/index/introduc…
- be careful what you download. be careful what you upload.
