"... a software tool cannot be an author of a submission, in part because it cannot consent to terms of use and the right to distribute content."
| Website | https://yoavartzi.com/ |
| Cornell NLP | http://nlp.cornell.edu |
Yoav Artzi
- 398 Followers
- 93 Following
- 79 Posts
#NLProc, ML researcher/teacher @ Cornell CS / Cornell Tech (NYC🚡) // Researcher at ASAPP
🤘Right on
Several US-based newspapers have recently mentioned @[email protected] 's reply to a set of chatGPT lyrics purportedly "in his style". The reply speaks for itself. If you're into writing, LLM or ML-creativity, it's worth a read.
https://www.theredhandfiles.com/chat-gpt-what-do-you-think/
🐦🔗: https://twitter.com/adinamwilliams/status/1615454708507844624
Nick Cave - The Red Hand Files - Issue #218 - I asked Chat GPT to write a song in the style of Nick Cave and this is what it produced. What do you think? Verse 1: In the depths of the night, I hear a call A voice that echoes, through the hall It’s a siren’s song, that pulls me in Takes me to a place, where I can’t begin Chorus: I am the sinner, I am the saint I am the darkness, I am the light I am the hunter, I am the prey I am the devil, I am the savior Verse 2: I walk the streets, with my head held high I’ve got my demons, but I’ll make them fly I’ll dance with the devil, and I’ll play his game I’ll be the one, to call his name Chorus: I am the sinner, I am the saint I am the darkness, I am the light I am the hunter, I am the prey I am the devil, I am the savior Bridge: I’ve got the blood of angels, on my hands I’ve got the fire of hell, in my eyes I’m the king of the abyss, I’m the ruler of the dark I’m the one that they fear, in the shadows they hark Chorus: I am the sinner, I am the saint I am the darkness, I am the light I am the hunter, I am the prey I am the devil, I am the savior Outro: So come with me, to the edge of the night Where the darkness and the light collide We’ll walk the line, between the good and the bad And we’ll see, who comes out on top in the end.
Dear Mark, Since its launch in November last year many people, most buzzing with a kind of algorithmic awe, have sent me songs ‘in the...
This is probably the first paper to give ChatGPT coauthor status, and its contact details points to support@openai! Giving coauthorship to writing assistants is absurd and this practice has to stop. 🧶
There's so much happening here, not sure where to start. Putting aside the dubious factuality, any idea what is "Mechaka" and how it translates to "story"?
When the optimistic hope is that it's "just a sophisticated phishing attack", and not someone is actually buying letters ....
The agent's goal is to modify the environment so the grounded truth-value of the given statement is a target boolean. The language in lilGym is semantically-rich and human-written. Including: set reasoning, spatial relations, cardinality, and more.
Humans learn language by acting in the world. Can RL agents do the same? lilGym is a new benchmark 🏋️ for RL + natural language + visual reasoning
https://arxiv.org/abs/2211.01994
https://lil.nlp.cornell.edu/lilgym/
Chief RL trainer: @[email protected] , in collboration with @[email protected] and @[email protected]
lilGym: Natural Language Visual Reasoning with Reinforcement Learning
We present lilGym, a new benchmark for language-conditioned reinforcement learning in visual environments. lilGym is based on 2,661 highly-compositional human-written natural language statements grounded in an interactive visual environment. We introduce a new approach for exact reward computation in every possible world state by annotating all statements with executable Python programs. Each statement is paired with multiple start states and reward functions to form thousands of distinct Markov Decision Processes of varying difficulty. We experiment with lilGym with different models and learning regimes. Our results and analysis show that while existing methods are able to achieve non-trivial performance, lilGym forms a challenging open problem. lilGym is available at https://lil.nlp.cornell.edu/lilgym/.
Language is acquired by hearing/reading what other say, forming hypotheses, following them, and seeing how others/world react. When systems get instructions from human users, they can do the same, and learn from the feedback humans give.
Users engaged with natural language systems can provide feedback in realtime, and this feedback is a super duper learning signal! So: deploy, train, repeat!
https://arxiv.org/abs/2212.09710
Last PhD paper w/@alsuhr/[email protected] ... 🧵
Continual Learning for Instruction Following from Realtime Feedback
We propose and deploy an approach to continually train an instruction-following agent from feedback provided by users during collaborative interactions. During interaction, human users instruct an agent using natural language, and provide realtime binary feedback as they observe the agent following their instructions. We design a contextual bandit learning approach, converting user feedback to immediate reward. We evaluate through thousands of human-agent interactions, demonstrating 15.4% absolute improvement in instruction execution accuracy over time. We also show our approach is robust to several design variations, and that the feedback signal is roughly equivalent to the learning signal of supervised demonstration data.