SM

[MMitchell @mmitchell_ai]

🧊 PSA: if you’re in a heat wave, consider putting ice out in bowls/bird baths for the birds. A lot of them will die otherwise. 😔

[Thomas Bloom @thomasfbloom]

@kevinweil Hi, as the owner/maintainer of https://t.co/69gOJM7Ci7, this is a dramatic misrepresentation. GPT-5 found references, which solved these problems, that I personally was unaware of.

The 'open' status only means I personally am unaware of a paper which solves it.

[Hugo Barra @hbarra]

I tried the new Meta Ray-Ban Display “Hypernova" glasses that Meta just announced (thanks for the demo @ahimel!)

TL;DR it's a really well-executed take on a simple idea: a crisp heads-up display with precise micro-gesture control of mini apps

My quick takeaways from the demo I got:

1/ The monocular heads-up display (yup, Google Glass vibes!) is surprisingly usable despite "only" 20 deg FOV; pretty crisp and bright enough even in broad daylight; with enough space for fully functional widgets that are much bigger than anything you could fit on a watch display

(BTW, the display specs are worth mentioning because they are nuts: 5,000 nits peak brightness + resolution of 42 PPD… vs. 34 PPD on Vision Pro)

2/ Meta is shipping their neural interface wristband (for micro-gesture control) much earlier than I expected; it’s by far the most mind-blowing part of the product and might be setting the standard for all smart/AR glasses to come:

By sliding your thumb over your index finger, you get a 4-way mini touchpad that allows you to really easily navigate around the UI + pinch to select + pinch and turn for volume/zoom + handwriting recognition on any surface (this last one won't launch right away, but I tried and it actually really works)

3/ The voice assistant experience has a clever twist: a dedicated double thumb tap micro-gesture (instead of an annoying wakeword) and mics that support low-volume speech so you can almost just whisper to the glasses

4/ To me, the magic of the experience came down to the fact that it felt much more effortless than I expected; I can use simple apps without taking the phone out of my pocket – reading & replying to messages, controlling music, walking navigation, fly through ‘snackable’ content

5/ One of Meta's biggest technical achievements on this product is zero light leakage from waveguides; this makes them much more polished and socially acceptable (folks who have been around AR glasses will know this is a big deal)

6/ Really enjoying the fact that Meta Ray-Ban Display (including the neural band) is conveniently priced at the same $799 as Apple Watch Ultra :)

P.S. Here’s my vote: would love to see Meta launching an app store for these glasses sooner rather than later — and match it with a strong developer push. This new category deserves the momentum.

[SM @snehaSM]

@vikib0y @kurtluther @QVeraLiao @snaglee @syljohns Congrats Vikram!

[Tech At Bloomberg @TechAtBloomberg]

Researchers in our #AI Engineering Group & CTO Office are publishing 11 papers this week at @aclmeeting; learn about their #NLProc research, why the results are notable, and how their work will advance the state-of-the-art
https://t.co/YzcotKsn9A
#ACL2023NLP

[SM @snehaSM]

I’ll be at #ACL2023 in person. Excited to catch-up with folks. Feel free to DM.

[Venu Satuluri @venusatuluri]

Fascinating to see GPT give a wrong answer with this simple blocks puzzle. (Both 4 and 3.5 fail at this, even after asking it to verify the answer.)

[bonappetweet @bonappetweet]

Hi, Twitter Fam just wanted to say our final good byes. It’s been a pleasure serving you this past 11 years. Twitter really made us feel apart of the culture and we will never forget that. 🎤🎤 drop

[SM @snehaSM]

Exciting research directions in future work to make this approach more efficient.

[SM @snehaSM]

If you're at EMNLP 2022, check out our paper titled
"Improving Zero-Shot Event Extraction via Sentence Simplification" with @DMfun and @profnaren presented at W1(CASE).

In this paper, we formulate event extraction as a reading comprehension task.

[SM @snehaSM]

We find that RC systems perform better at event extraction on simplified candidates than on original candidates. This result is somewhat surprising given how proficient BERT-based models have become at reading comprehension.

[SM @snehaSM]

The sentence does not mention any other individuals, so there are only two people mentioned in this sentence.

[SM @snehaSM]

Co-ordination scope ambiguity can confuse ChatGPT.

[SM @snehaSM]

ChatGPT: This sentence mentions two people: Shuttle veteran Fred Gregory and NASA executive Fred Gregory. The sentence indicates that Fred Gregory is a Shuttle veteran and a NASA executive, and that he has been appointed to a board.

[Shubhanshu Mishra @TheShubhanshu]

For past 1.5 years we worked on building TweetNERD a dataset of 340K+ Tweets with Named Entities annotated and linked to @wikidata.
Join our poster at @NeurIPSConf in afternoon today.
Work with @sainiAman14 @rahelehmakki @snehaSM @AMollahosseini @aria42
#NeurIPS22 #NLProc 🧵👇

[SM @snehaSM]

@NeurIPSConf I paid for full registration, but can’t access the virtual conference because I no longer have access to the email id I registered with. Contacted IT support twice. Can you please loop in the relevant people to resolve this asap?

[SM @snehaSM]

@TheShubhanshu 🫡

[Shubhanshu Mishra @TheShubhanshu]

Additional analysis in our paper.
Work done as part of @jinninguiuc's @terns ship, and collaboration with Ahmed El-Kishky, @snehaSM, & @viveksck
Youtube: https://t.co/ARUjgUSklQ
Paper: https://t.co/CX8hw35Sic
Workshop: https://t.co/yvXUiW5kc6
#COLING #COLING2022