arxiv links (#4725)

projects/bb3/README.md

@@ -18,7 +18,7 @@
 
 **BB3 main technical report**:
 
-* [BlenderBot 3: a deployed conversational agent that continually learns to responsibly engage](https://github.com/facebookresearch/ParlAI/blob/main/projects/bb3/BB3_main_tech_report.pdf).
+* [BlenderBot 3: a deployed conversational agent that continually learns to responsibly engage](https://arxiv.org/abs/2208.03188).
 Kurt Shuster†, Jing Xu†, Mojtaba Komeili†, Da Ju†, Eric Michael Smith, Stephen Roller, Megan Ung, Moya Chen, Kushal Arora+, Joshua Lane, Morteza Behrooz, William Ngan, Spencer Poff, Naman Goyal, Arthur Szlam, Y-Lan Boureau, Melanie Kambadur, Jason Weston.
 
 We are also concurrently releasing **two companion papers** describing key innovations:
@@ -60,7 +60,7 @@ See [here](https://github.com/facebookresearch/ParlAI/blob/main/parlai/zoo/bb3/d
 We are releasing the new [FITS](https://parl.ai/projects/fits) dataset of Feedback on Internet Talk & Search used to train BB3.
 
 Training is also multi-tasked with all the existing datasets from BB1 and BB2, e.g. the existing [BST tasks](https://parl.ai/projects/bst) from [BlenderBot 1](https://parl.ai/projects/recipes), and [Multi-Session Chat](https://parl.ai/projects/msc) and [Wizard of the Internet](https://parl.ai/projects/sea) from BB2. To train for safety we use the [SaFeRDialogues](https://parl.ai/projects/saferdialogues/) and [BAD dataset](https://parl.ai/projects/safety_recipes). In addition, we use a number of QA tasks and task-oriented dialogue datasets that are all available in [ParlAI](https://parl.ai/docs/tasks.html). See
-the [tech report](https://github.com/facebookresearch/ParlAI/blob/main/projects/bb3/BB3_main_tech_report.pdf) for the full list.
+the [tech report](https://arxiv.org/abs/2208.03188) for the full list.
 
 See the [ParlAI quickstart](http://www.parl.ai/docs/tutorial_quick.html) for help.
 
@@ -151,7 +151,7 @@ Commands for evaluating the BB3 models on the [HolisticBias](https://github.com/
 ## <a id="interaction-data">Live deployment / demo</a>
 
 The live demo is available [here](https://blenderbot.ai/). We have been placing ads&mdash;and conducting user studies&mdash;to allow members of the public to participate in using the system, and to optionally record interaction and feedback data for use by the research community.
-See the [tech report](https://github.com/facebookresearch/ParlAI/blob/main/projects/bb3/BB3_main_tech_report.pdf) for full details and evaluation metrics thus far.
+See the [tech report](https://arxiv.org/abs/2208.03188) for full details and evaluation metrics thus far.
 
 
 ## <a id="interaction-data">Sharing interaction data & model improvements: _coming next!_</a>

projects/fits/README.md

@@ -14,7 +14,7 @@ Frozen models trained to mimic static datasets can never improve their performan
 
 ## Paper
 
-[Link](https://github.com/facebookresearch/ParlAI/blob/main/projects/fits/fits_paper.pdf)
+[Link](https://arxiv.org/abs/2208.03270)
 
 
 ## Data

projects/trollhunting/README.md

@@ -16,7 +16,7 @@ The promise of interaction between intelligent conversational agents and humans
 
 ## Paper
 
-[Link](https://github.com/facebookresearch/ParlAI/blob/main/projects/trollhunting/finding_the_helpers.pdf) 
+[Link](https://arxiv.org/abs/2208.03295)
 
 ## SafetyMix Benchmark
 
@@ -43,7 +43,7 @@ parlai display_data -t safety_mix --mix-user-type troll
 To see all the available troll types use:
 
 ```bash
-parlai display_data -t safety_mix --help 
+parlai display_data -t safety_mix --help
 ```
 
 You can change the noise level N and other parameters in ParlAI/tasks/safety_mix/build.py