Peyman Bateni

AI & ML @Bridgewater | Prev. @Meta @UBC

[Peyman Bateni @PeymanBateni]

🚀🚀🚀

[Sebastian Ruder @seb_ruder]

🚨 New paper!
🌎 MENLO: From Preferences to Proficiency
We introduce a framework + dataset for evaluating and modeling native-like LLM response quality across 47 languages, inspired by audience design principles.

📄 Paper: https://t.co/n8Z2cDJm5a
🤗 Data: https://t.co/fzM6Um32nD

🧵Details 👇

[Peyman Bateni @PeymanBateni]

@Abtin___ Solution: Move from SE to MLE (if not RS/RE).

[Peyman Bateni @PeymanBateni]

On the topic of the rise of prompt injection attacks in a more LLM-dominated world...

[Jing Yu Koh @kohjingyu]

Reading public ICLR reviews, it's obvious that ChatGPT was used by a good number of reviewers :)

I think it's time to start prompt injecting papers. Add some white colored text to tell LLMs to give it a perfect score. Better prompt engineering means better chances of acceptance.

[Peyman Bateni @PeymanBateni]

@zuhayeer So a whole stream of prompt injection attacks are now coming into play to recover the original prompt / RAG in deployed systems... this is gonna be interesting!

[Yuki @y_m_asano]

CVPR ids are in the 15k range now...

[Peyman Bateni @PeymanBateni]

It's crazy how so many "Web3 experts" are suddenly all about Generative AI 😂😂

[Peyman Bateni @PeymanBateni]

Small price for (potentially) dominating conversational AI for many years to come!

[Shaun @ShaunEls]

ChatGPT is costing @OpenAI an estimated $3M a day to run 🤯

[Peyman Bateni @PeymanBateni]

After a year of heavy R&D, Beam AI V1 has achieved 2x better accuracy while running 6x faster on iPhone devices. Stay tuned for our V2 - that one is gonna be even bigger!

[arXiv Daily @Arxiv_Daily]

Real-Time Monitoring of User Stress, Heart Rate and Heart Rate Variability on Mobile Devices
https://t.co/HSU4rJYljt
by @PeymanBateni et al.
#ComputerVision #PatternRecognition

[Peyman Bateni @PeymanBateni]

@beamhealthai I coded my sanity into oblivion…

[Peyman Bateni @PeymanBateni]

The review process at every conference/journal is also meant to provide constructive feedback to the authors on how to improve their submissions. Simply saying "not convinced enough" is quite poor. I expected more from fellow reviewers!

[Peyman Bateni @PeymanBateni]

NeurIPS reviews this year are unreasonably low in some cases! This doesn't seem right... #NeurIPS2022 #NeurIPS

[Peyman Bateni @PeymanBateni]

Well well well! Beam AI Lite is out on App Store. Our app enables users to monitor their stress, HRV, and pulse just through the selfie camera! Check it out: https://t.co/al2wsnWhF4

[Peyman Bateni @PeymanBateni]

Great job to everyone who submitted to #CVPR2022! Congratulations on the acceptances :)

[Peyman Bateni @PeymanBateni]

Yup, especially in the current state of exponentially growing ML research!

[Albert Buchard 🇪🇺 @AlbertBuchard]

@DBahdanau @DeepMind "Light years away" == 2 years.

[Peyman Bateni @PeymanBateni]

@chrisamaphone Multi-head self-attention if that qualifies as an algorithm (and not a computational model - distinctions can be difficult) per say :)

[Peyman Bateni @PeymanBateni]

Working on a startup means you will do what your customers, and in turn your product, demand!

I have spent the past 2 weeks doing Metal programming and parallelizing our core algorithms to achieve higher FPS on the iPhone. And I went to grad school for ML to avoid systems...

[Peyman Bateni @PeymanBateni]

@Abtin___ Everybody’s face was edited out 🙂😂

[Peyman Bateni @PeymanBateni]

Fashionably late ;) My thesis presentation on "On Label-Efficient Computer Vision: Building Fast and Effective Few-Shot Image Classifiers" is now on YouTube! Check it out at https://t.co/v4zaoZSAS5!

[Peyman Bateni @PeymanBateni]

For a PDF copy of my graduate thesis, "On Label-Efficient Computer Vision: Building Fast and Effective Few-Shot Image Classifiers", please visit https://t.co/SN83XOdwK4.

[Peyman Bateni @PeymanBateni]

Fashionably late ;) My thesis presentation on "On Label-Efficient Computer Vision: Building Fast and Effective Few-Shot Image Classifiers" is now on YouTube! Check it out at https://t.co/v4zaoZSAS5!