| https://twitter.com/ShahabBakht | |
| website | https://shahabbakht.github.io/ |
New Jazlab work! When monkeys produce vectors on a map w/o visual feedback, an endogenously activated cognitive map in the entorhinal cortex reflects animals' knowledge about the (invisible) intervening landmarks. Exciting work by @[email protected] w/ @[email protected] http://doi.org/10.1101/2022.12.15.520640
๐ฆ๐: https://twitter.com/mjaz_jazlab/status/1603776949062537216
This is the PsychoPy team's #introduction to Mastodon.
PsychoPy is an #opensource package to build and run experiments in #python and #js
PsychoPy is a widely used tool for the behavioural sciences like #psychology #linguistics #economics #neuroscience
This account will also tweet about other parts of the ecosystem like #pavlovia (it's like GitHub for behavioural science) and #PsychoJS
Looking forward to chatting with you here! ๐ฅฐ๐
๐ฃ๐ผ๐ฝ๐๐น๐ฎ๐๐ถ๐ผ๐ป ๐ฐ๐ผ๐ฑ๐ถ๐ป๐ด ๐ฎ๐ป๐ฑ ๐ป๐ฒ๐๐ฟ๐ผ๐ป๐ฎ๐น ๐ฐ๐ถ๐ฟ๐ฐ๐๐ถ๐๐ ๐ฎ๐ฐ๐ฟ๐ผ๐๐ ๐ฒ๐๐ผ๐น๐๐๐ถ๐ผ๐ป
Check this cool essay by Kevin Mitchell:
http://www.wiringthebrain.com/2022/12/how-many-neurons-does-it-take-to-change.html
The #ChatGPT paper is now on arXiv https://arxiv.org/abs/2203.02155
EDIT: This is not a new paper, but it is the technique used to train ChatGPT and what is now apparently called GPT3.5
Making language models bigger does not inherently make them better at following a user's intent. For example, large language models can generate outputs that are untruthful, toxic, or simply not helpful to the user. In other words, these models are not aligned with their users. In this paper, we show an avenue for aligning language models with user intent on a wide range of tasks by fine-tuning with human feedback. Starting with a set of labeler-written prompts and prompts submitted through the OpenAI API, we collect a dataset of labeler demonstrations of the desired model behavior, which we use to fine-tune GPT-3 using supervised learning. We then collect a dataset of rankings of model outputs, which we use to further fine-tune this supervised model using reinforcement learning from human feedback. We call the resulting models InstructGPT. In human evaluations on our prompt distribution, outputs from the 1.3B parameter InstructGPT model are preferred to outputs from the 175B GPT-3, despite having 100x fewer parameters. Moreover, InstructGPT models show improvements in truthfulness and reductions in toxic output generation while having minimal performance regressions on public NLP datasets. Even though InstructGPT still makes simple mistakes, our results show that fine-tuning with human feedback is a promising direction for aligning language models with human intent.
Nidhi Seethapathi
Alex Holcombe
PsychoPy
Kevin Mitchell
Luiz Pessoa
pierre_bellec ๐ฆ
Eilif B. Muller
Mark Riedl