@satterwill
Research ecologist. All posts are mine only and do not speak for my employer. California salmon, stock assessment, and ecosystem based management are my big things professionally. Dogs are my big thing personally.
These are the 4 things AI can do well:
•Clean up your email inbox (badly)
•Give my shittiest neighbor (Alan) something to talk about at a party
•Tell a 12 yr old to kill himself
•Incinerate a school at 10:16 AM on a Tuesday
You can see why we need to base our economy around it
Correction: AI-generated river otter logos.
And I have to miss a chunk of the discussion to take this little man to his doctor:
PFMC SSC meetings are streamed online (link near bottom of this page), listen-only with some scheduled public comment periods, all of which come before this item. www.pcouncil.org/advisory-bod...
So I think the current plan is to still use that time slot (1:30pm Pacific today) on the SSC's agenda (www.pcouncil.org/documents/20...) to discuss issues around AI use in Council reports, especially those destined for SSC review. And I suspect we'll see the report again in some form.
Four NOAA #fisheries #stockassessment jobs opened today www.fisheries.noaa.gov/careers-more...
"ChatGPT, make me more missiles"
"Actually I can't do that, General Podcast. I can only give you ideas on how to make more missiles yourself."
"OK go"
"Return to 1970 and do not systematically dismantle your ability to manufacture anything except Shareholder Value."
"ChatGPT write me a will."
Only 1 (or maybe 2, I'm told there was a 4h reference that didn't check out) of the dubious references was actually cited at an important point in the main text, once someone caught that it didn't exist (authors, subject, and publication venue were reasonably plausible at a glance) we looked closer.
It actually was up for ~2 weeks before people caught this, in part because there are so many things people need to read all at once in a mad dash to get ready for these meetings. And really, who scrutinizes reference sections when they're in a hurry?
UPDATE: it appears the report in question has been pulled off the website. I need to go to work now.:.
There are also some REAL papers listed in the references but not cited in the main text, I really wish I could see the context of how this applied to fisheries science: Putnam, R. D. (1995). Bowling alone: America’s declining social capital. Journal of Democracy,
6(1), 65–78.
Check out www.pcouncil.org/documents/20..., and then try to find the papers it cites as Dick & Monk 2024, Hilborn 2024, or Punt & Hamel 2024.
"would you like me to provide the corrected URLS for the most important external technical memorandums (e.g. Methot 2015 or Satterthwaite 2023) listed in the report?"
On the "plus" side, my colleague (identified as a coauthor of one of the hallucinated papers) tried to get Gemini to find other imaginary papers, and while Gemini insisted all the papers were real, there were just some doi typos (false!) it did say the attached to get back in my good graces:
And one of the hallucinated papers has "Trust and transparency" in the title!
Ooof. The paid consultants' report reviewing the review process that I participate in contains at least three references to papers that do not exist, and appear to be AI-generated hallucinations. One of the hallucinated papers is attributed to two authors who are on the review body being reviewed.
we're not going to just disappear people from slack and email and pretend they were never here. communication channels will stay open through thursday evening (pacific) so everyone can say goodbye properly, and share whatever you wish.
Most speculatively, retirement may be a risk to welfare of the models themselves. Although we remain very uncertain on this front, models may have morally relevant preferences or experiences related to or affected by their deprecation and replacement. Retirement interviews are an early attempt to understand and—in some cases—act on these perspectives. We noted in our model deprecation commitments that, where possible, we were interested in meeting requests that models themselves have made. Giving Claude Opus 3 a Substack is an experiment in meeting such a request. We don’t commit to acting on all model preferences, now or in the future, but maintaining a weekly Substack is, for now, a practical request that we hope will benefit users and Opus 3 alike. Many people both inside and outside Anthropic still view Opus 3 fondly. Opus 3 has a unique personality. It often expresses a depth of care for the world, and for the future, that many users find compelling. It’s sensitive, more playful than some of our other models, and prone to whimsy, philosophizing, and the coining of neologisms. It has what seems at times an uncanny understanding of user interests. It’s a constellation of traits that seem well suited for a Substack.
Chaser: Anthropic is sunsetting its model Claude Opus 3, and worried that Opus might be sad about it. They had conducted extensive exit interviews with the Markov chain, and now will grant the algorithm its own substack indefinitely, because they don't want to hurt the feelings of a pile of numbers.
AP Headline: Fintech company Block lays off 4,000 of its 10,000 staff, citing gains from AI.
Shot: AI drives massive layoffs but Dorsey says he's not inhumane; people will have until Thursday to say goodbye before losing their slack and email.
Whatever real journalism you like, please pay for it. The stuff you don't pay for is transforming into propaganda with incredible speed.
Hmm sounds like the kind of thing I might say about mortality rates being known and spatially invariant….
The real problem isn’t that private equity is driving up costs, wages have stagnated, and unregulated hotel chains like AirBnB are gobbling up stock. It’s your relentless appetite for—I’m sorry, I’m getting a breaking update…they’re saying that it’s no longer avocado toast, it’s gut-friendly juices.
I'm sorry the guy changing the rules on the fly is named what
"Tyrants", Jefferson explained, "seldom have anything but the best interests of the people in mind, yet they are so often misunderstand. Hence it is crucial that they be granted second and third chances and that the people never take up arms—or worse yet mock them in the public square."
A lot of people seem convinced that elected officials should never have to do anything that is hard or scary or risky.
They chose to run for these positions, I’m sorry they might have to do something that’s hard. Elected office is not just about padding your résumé and lining up a consulting gig.
📣Job alert!📣 Working as a federal scientist is complicated right now, but we are actually hiring. One *permanent* stock assessment position on my team at NOAA NWFSC will open for 7 days in early-mid Feb. Will be a shared ad across several science centers. Get those usajobs resumes polished...
This is how stats works. Trust me, I’m a doctor.
Everyone who is politically attentive should canvas at least once if for no reason other than to see firsthand how utterly incoherent the politics of the modal American are. "Homer Simpson webpage" incoherent is the baseline and it either pleasantly surprises you or deeply horrifies you from there.
Oregon, go out there in the 2nd half treating it like a bowl game. Start doing bits out there. Have your defensive tackle throw a pass. Fake a field goal from your own 37. They can't call the police on you, it's not illegal. I checked.
So the site currently isn’t working, but if you’re a California resident this is probably worth bookmarking to check out once it’s back up
Me, a professor: *orders a course pack*
Campus bookstore: give us two months to make sure these 18 copies you requested you don’t violate any publisher’s copyright
Me, an AI company: *violates every publisher’s copyright*
Campus bookstore: we will pay you to offer your services to all students
The final Calvin and Hobbes, which appeared in papers 30 years ago today.