When Dumping an Employee Yields a Conference: Unexpected Consequence? Yep
February 20, 2023
The saga of Google’s management of smart people has taken a surprising twist. On Friday, March 17, 2023, Dr. Timnit Gebru and some colleagues have declared “Stochastic Parrots Day.” The conference is named after the journal article/research paper about some of the risks certain approaches to smart software generates.
Stochastic parrots created by the smart software Craiyon.com. I assume that Craiyon is the owner of these images and that image rights trolls will be on the prowl for violations of the software’s intellectual property. But I enhanced these stochastic parrots, and I wrote this essay. No smart software writing aids for this dinobaby.
You can download the paper “On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? The paywalled ACM version is at this link. The authors of the paper that allowed Dr. Gebru to find her future elsewhere are Emily Bender, Angelina McMillan-Major, and another Xoogler purged from the online ad outfit Margaret Mitchell. from this link, which raises a paywall. However, there is a useful summary prepared by Tushar Chandra at this link. According to the conference announcement, the co-authors and “various guests” will “reflect on what has happened in the last two years, what the large language model landscape currently looks like, and where we are headed versus where we should be headed.”
In my experience, employees who have the opportunity to find their future elsewhere start poking around for work. A few start companies or non-profits. Very few set up a new conference named after the paper which [a] blew the whistle on some of the AI craziness reported endlessly in TechMeme and other online information services and [b] put US Army De Oppresso Liber laser on Google’s personnel management methods.
Yep, a conference. A free conference, although a registrant can donate to the organizers.
What’s the unexpected consequence or, I should say, consequences? Let me do a little speculation:
- Google amps up the Sundar and Prabhakar routine about how Google wants to be careful, to earn trust, and, of course, demonstrate that Microsoft’s brilliant marketing play is just stupid. (Who is hallucinating? Microsoft’s OpenAI demonstrations or the Google?)
- The conference attracts the attention of a major conference organizer. I am not sure the ACM will have the moxie to create a conference that appeals to those who are not members. Imagine a two per year Stochastic Parrot program held twice a year. I think it might work.
- This event strikes me as similar to a one of those quantum moments. Is the parrot dead or alive? Predicting how the conference will interact with the real world and what systems and methods find themselves under the parrot’s confocal-type differential interference contrast microscope. What will emerge? Recursive methods fed synthetic data? Higher level abstractions shaped by engineers’ biases? Misinformation ingested so that results don’t match other sources and findings? Carelessness infused with cost cutting in the content training process? Sail and Snorkel perhaps?
Net net: What happens if a stochastic parrot conference gets too big? Answer: Perhaps Jeff Dean will become a speaker and set the record straight? Yikes! Code Super Red?
Stephen E Arnold
Comments
-
- Subscribe to Beyond Search
Feature archive
News archive
-
