Another Message Board
Post comments on any topic. Civil discussion and no coarse language please. Side discussions and idees fixes to the sandpits, please.
I’ve moved my irregular email news from Mailchimp to Substack. You can read it here. You can also follow me on Mastodon here
I’m also trying out Substack as a blogging platform. For the moment, I’ll post both at this blog and on Substack.
John Quiggin 
“JohnQuiggin @JohnQuiggin
“Tried out GPT-2 Output Detector Demo for the first time. Startlingly accurate on the first couple of tries. Picked GPT text with 99.98% accuracy, original text from a blog post of mine 99.97%”
How about a watermark?
Scott Aarronson, now assisting OpenAI, is working on:
[Loooong article: goto heading]
“My Projects at OpenAI”
…
“My main project so far has been a tool for statistically watermarking the outputs of a text model like GPT. Basically, whenever GPT generates some long text, we want there to be an otherwise unnoticeable secret signal in its choices of words, which you can use to prove later that, yes, this came from GPT. We want it to be much harder to take a GPT output and pass it off as if it came from a human. This could be helpful for preventing academic plagiarism, obviously, but also, for example, mass generation of propaganda—you know, spamming every blog with seemingly on-topic comments supporting Russia’s invasion of Ukraine, without even a building full of trolls in Moscow. Or impersonating someone’s writing style in order to incriminate them. These are all things one might want to make harder, right?
“More generally, when you try to think about the nefarious uses for GPT, most of them—at least that I was able to think of!—require somehow concealing GPT’s involvement. In which case, watermarking would simultaneously attack most misuses.
“How does it work? For GPT, every input and output is a string of tokens, which could be words but also punctuation marks, parts of words, or more—there are about 100,000 tokens in total. At its core, GPT is constantly generating a probability distribution over the next token to generate, conditional on the string of previous tokens. After the neural net generates the distribution, the OpenAI server then actually samples a token according to that distribution—or some modified version of the distribution, depending on a parameter called “temperature.” As long as the temperature is nonzero, though, there will usually be some randomness in the choice of the next token: you could run over and over with the same prompt, and get a different completion (i.e., string of output tokens) each time.
“… But now you can choose a pseudorandom function that secretly biases a certain score—a sum over a certain function g evaluated at each n-gram (sequence of n consecutive tokens), for some small n—which score you can also compute if you know the key for this pseudorandom function.
“To illustrate, in the special case that GPT had a bunch of possible tokens that it judged equally probable, you could simply choose whichever token maximized g. The choice would look uniformly random to someone who didn’t know the key, but someone who did know the key could later sum g over all n-grams and see that it was anomalously large. The general case, where the token probabilities can all be different, is a little more technical, but the basic idea is similar.
“One thing I like about this approach is that, because it never goes inside the neural net and tries to change anything, but just places a sort of wrapper over the neural net, it’s actually possible to do some theoretical analysis! In particular, you can prove a rigorous upper bound on how many tokens you’d need to distinguish watermarked from non-watermarked text with such-and-such confidence, as a function of the average entropy in GPT’s probability distribution over the next token. Better yet, proving this bound involves doing some integrals whose answers involve the digamma function, factors of π2/6, and the Euler-Mascheroni constant! I’m excited to share details soon.
“Some might wonder: if OpenAI controls the server, then why go to all the trouble to watermark? Why not just store all of GPT’s outputs in a giant database, and then consult the database later if you want to know whether something came from GPT? Well, the latter couldbe done, and might even have to be done in high-stakes cases involving law enforcement or whatever. But it would raise some serious privacy concerns: …”
My AI Safety Lecture for UT Effective Altruism
https://scottaaronson.blog/?p=6823
Cory Doctorow & Rebecca Giblin …
“The Australian Chokepoint Capitalism Tour
“with dates in Brisbane, Melbourne, Sydney and Canberra
“First up is Brisbane, with a talk and signing at Avid Books on Feb 8 at 6PM, hosted by The Griffith Review’s Carody Culver:
https://avidreader.com.au/pages/6834-ChokepointCapitalism-RebeccaGiblinandCoryDoctorow
https://pluralistic.net/2023/01/30/i-come-to-a-land-downunder/#anti-podean-pro-artist
KT2 and others,
Nice heads up. I might have even gone to the Brisbane one but for COVID-19. There are too many immunosuppressed, autoimmune affected and elderly people in my near family for this to be viable, even with mask wearing… seeing that likely nobody else there would be wearing masks. Of course, I can get the book and probably will.
I am not one particularly for artistic exceptionalism just as I am not one for sporting exceptionalism. This may because I have always been mediocre at artistic and sporting endeavors. Even so, I don’t think em as provide us with science fiction, for example, are any more important than em as provide us with rice (we being more non-wheat grain eaters in our house).
If the argument is for moving some elite income from elite capitalists to elite artists and elite sportsmen, while not removing capitalism itself or at least monopoly/monopsony capitalism, I am not going to be at all sympathetic. But hopefully the argument is better than that.
I am very much for paying everyone a basic wage now whether they work or not. The workers should then get tax deductions for all expenses necessary to get to and do work. So if anyone wanted to sit at home and write science fiction on a basic wage, they could. If they could sell their stuff that would be a bonus. Alternatively they could go surfing.
People should be given a living benefit (more at the level of the current basic wage than at the level of the current unemployment benefit). How we would then get people to do anything in what we regard as the formal economy? The answer is increments. As of 1 July 2022 the National Minimum Wage was $21.38 per hour or $812.60 per week. The current median weekly wage is about $1,000 to $1,500 depending on which state you live in. If wages were made reasonable and livable across the board, one could imagine a median wage of $1,600 nationally and an average wage of over $2,000. Wages could be made up to these levels largely by the expedient of legislating all female wages be lifted to the equal of male wages. Now is the time to do it, in a labor shortage period.
Would people go to work or stay at work 4 days a week (bringing in at the same time the 4 day week as advocated by J.Q.) for a bump of $800 (a near doubling) to $1600 or more a week for working? I think they would. Everyone already living on $2000 to $4000 a week (a single or a pair respectively) is committed to needing that level of income and will in the main keep going to work. Their incentive to support the change would be getting the 4 day week. They need feel less envious of the unemployed getting $800. And if they feel envious they can quit and get that lifestyle. Few would quit I predict.
Can the nation afford it? Yes. The money chits are nothing of course. The money can be “printed” (electronically). The sole real dangers are labor shortages and inflation. Labor shortages Australia can handle, largely by replacement immigration rather than by growth immigration. Inflation needs an analysis too (too long for this post) but we can ignore all the usual neoliberal arguments. They gave us asset inflation and now goods and services inflation (all as differential inflations) at high rates even though labor played the game as asked and took real wage cuts for decades. The neoliberal capitalists have made off with all the gains, with the huge profits, huge subsidies and huge increases in wealth. This is where ALL the inflation from the last two decades, at least, has come from. The answer is clear. Tax the rich hard on income and assets to balance the social redistribution of income, wealth and access to real goods and services.
Ikon said “while not removing capitalism itself or at least monopoly/monopsony capitalism, I am not going to be at all sympathetic. But hopefully the argument is better than that.”
Ikon, the argument may not reach your stratospheric bar, yet Chokepoint & enshitification cut through and expose the worst of crapitalism. Therefore anchoring and informing a far wider audience than say Economics in Two Lessons.
Cory Doctorow has several blogs and sites and so most of his publications may be gleaned. Not a chokepoint.
“TWITCH DOES A CHOKEPOINT CAPITALISM
…
“There’s a name for an economic arrangement where there are just a few buyers, and they put the squeeze on sellers: a #monopsony. In the economic literature, monopsonies are considered especially dangerous because they are able to extract concessions from their suppliers far more easily than monopolies (concentrated sellers) can from their customers. Monoposonists who represent just 10 percent of their sellers’ business can start turning the screw.
“Amazon’s pretty frank about this. In its own investor presentations, it describes its “flywheel”: bring in customers by subsidizing below-cost prices, lock those customers in with Prime, then extract price concessions from businesses that have to use its platform to reach those locked-in customers:
…
https://craphound.com/chokepoint/2022/09/27/twitch-does-a-chokepoint-capitalism/
And I’d vote for Cory Doctorow to replace Jim Chalmers after reading “Capitalism after the crises”
https://www.themonthly.com.au/issue/2023/february/jim-chalmers/capitalism-after-crises
These figures worry me. 14%.
86% of the tribe are ignorant, brainwashed or are abke to be manipulated. A worry.
US trust in news…
1976 = 72%
2022 = 36%
2022 Republicans = 14%
New Yorker below.
“Almost pathologically,” Sullivan says, reporters “normalized the abnormal and sensationalized the mundane.”
Australia
“In Australia in 2021, 71 percent of survey respondents said they trusted ABC News, compared to 12 percent who said they didn’t. By comparison, 47 percent said they trusted the Daily Telegraph newspaper and 22 percent said they didn’t trust it.
Public trust in selected news brands in Australia 2021 @ Statistica
*
“When Americans Lost Faith in the News
“Half a century ago, most of the public said they trusted the news media. Today, most say they don’t. What happened to the power of the press?
…
“The press wasn’t silenced in the Trump years. The press was discredited, at least among Trump supporters, and that worked just as well. It was censorship by other means.
“Back in 1976, even after Vietnam and Watergate, seventy-two per cent of the public said they trusted the news media. Today, the figure is thirty-four per cent. Among Republicans, it’s fourteen per cent
…
“In the memoir slash manifesto “Newsroom Confidential: Lessons (and Worries) from an Ink-Stained Life” (St. Martin’s), Margaret Sullivan argues that objectivity is not so much impossible today as meaningless, and that the press ought to stop striving to achieve it. The events of 2020 and 2021 showed that the press’s values were in the wrong place. “The extreme right wing had its staunch all-in media allies,” she writes. “The rest of the country had a mainstream press that too often couldn’t, or wouldn’t, do their jobs. Too many journalists couldn’t seem to grasp their crucial role in American democracy.”
…
“Almost pathologically,” Sullivan says, reporters “normalized the abnormal and sensationalized the mundane.”
…
https://www.newyorker.com/magazine/2023/02/06/when-americans-lost-faith-in-the-news
Review of…
“A Veteran Press Critic Wants Her Profession to Defend Democracy
“In her new memoir, “Newsroom Confidential,” Margaret Sullivan argues that traditional ideas about reportorial objectivity need to be re-examined in an era of constant assaults on truth.”
nytimes
com/2022/10/17/books/review/newsroom-confidential-margaret-sullivan.html
KT2,
I don’t think my arguments or ethics reach “stratospheric bars”. In fact, I don’t think I set very high bars at all. It’s just that we’ve become used to such low bars from capitalism (somewhere down in the Stygian depths) that our standards are become rather low and distorted.
Also, I don’t do consistent hero worship of contemporary intellectuals, writers or anyone else, although I might occasionally gush a bit about a particular production. Not saying you do hero worship, just saying that I don’t. I simply start from a position of skepticism about the likelihood of special pleading, careerism, vested interest and motivated reasoning. Again, not saying Cory Doctorow does any of these. But in general we seem to have seen as much special pleading (and lack of concern about COVID-19 and other non-trendy social issues) from artists and sportsmen (of the elite earning capacity variety) as we have from the owners of inner city businesses and real estate. Not saying C.D. is an elite earner either. I have no knowledge on that score. For certain I will give him and his co-author a fair reading when I get to it.
Good advice Ikon “I simply start from a position of skepticism about the likelihood of special pleading, careerism, vested interest and motivated reasoning”
My priors, when I came across Cory Doctorow were as you state above. I have found his analogies appeal to my rebel / non elite or less scholastic side. And his grasp of the near term zeitgeist. And hey – enshitification as a word and concept – cuts through when trying to communicate crappy capitalism to “hero worshippers” of fb, Amazon et al. Trying to use econ jargon doesn’t cut it on the street.
Ikon said “For certain I will give him and his co-author a fair reading when I get to it” shows an open mind.
As CD has written so widely and shows what is behind the curtain without fear ir favour, I have asked Crooked Timber via JQ to do a seminar series of posts. If CT did a CD Seminar you, me and the writers and readers of CT will be delivered a worthy corpus of information with applied blowtorch, to put into relief many of the topics we here are passionate about.
*
I had to look up “Stygian depths”.
Play:
“JUKE’S TOWERS OF HELL WIKI
“Stygian Depths is a subrealm created by Lightsynth which is used for Curator Applications. It contains 4 already in-game towers for applicants to review, those being
– a Difficult tower,
– a Challenging tower,
– a Remorseless tower, and an optional
– Extreme tower.
“It can currently be played here.
https://jtoh.fandom.com/wiki/Stygian_Depths
*
“By metonymy, the adjective stygian (/ˈstɪdʒiən/) came to refer to anything dark, dismal, and murky.
https://en.m.wikipedia.org/wiki/Styx
Thanks. I always learn something here.
Motivated reasoning by Ministers of the Crown. Which is illegal in the real world.
“The minister requested the file of every single person who appeared in the media … you could see the exact transactions that they’d had with Centrelink.”
“This would send a clear message … that maybe consider it [going to the media] twice.
“There were less people speaking out in the media which was our intention.”
https://www.abc.net.au/news/2023-01-31/qld-robodebt-scheme-government-royal-commission-alan-tudge/101910064
Cory Doctorow writes for, or at least is re-published in, the Capital as Power website from time to time.
https://capitalaspower.com/2022/11/at-last-a-new-econ-101-textbook/
But I guess if you know the Pluralistic site you will see all his pol-econ stuff anyway.
Interesting and concise essay on democracy vs. the Internet by Yale professor Robert Post: https://balkin.blogspot.com/2023/01/democracy-and-internet.html
Sample: “The scale of the internet makes its regulation incompatible with law.”
I share Harry’s puzzlement over Adani’s sleaze. If you are a little rich, with say $20m in commercial property and second-hand car dealerships, it’s a reasonable risk to hide it from the taxman and the ex in a tax haven. You don’t stand out from the crowd, and the people who might investigate you have limited resources to spend on a not very rewarding target. You have a good chance of getting away with it. Great fortunes like Gautam Adani’s are different. They can’t really be concealed, just made more opaque. Your enemies are likely to include tireless and unscrupulous prosecutors, reporters, hired PIs, and vultures like Hindenburg, drawn to the scent of a great prize. Adani may be relying on his patron Modi for protection. This should work in India, but not in Singapore or Australia, and Modi is a very slippery customer who will ditch him if he needs to. As Harry asks, all to what end? Adani undoubtedly owns billions in real productive assets, it’s not a Ponzi scheme.
A the time of writing, the Adani enterprise seems to have survived the attack, but its shares have taken a very large ($70bn) hit. Hindenburg have recouped their investment many times over. My guess is that some of the damage is permanent. Investors and banks will be warier in future, and regulators more likely to dig into the paper trail. Maybe enough to sink Carmichael for good?
Worth a read.
https://sciencebasedmedicine.org/should-we-let-unvaccinated-children-die-of-covid-as-a-matter-of-course/
ICYMI, there’s a 6-part twitter thread from epidemiologist Raina MacIntyre posted on Feb 3:
Also, Prog Kathy Eager tweeted yesterday (Feb 3):
A Boeing 737-800 aircraft has a maximum seating capacity of 189.
So 1,799 COVID-related deaths is equivalent to roughly ten full Boeing 737-800 planeloads of passengers.
Imagine if a larger aircraft full of passengers crashed every 3 or 4 days – people would be demanding to know what could be done to fix the problem, right?
Yet it seems governments at all levels don’t seem to care about Australia’s current COVID-19 death rate, or the even bigger & growing problem of tens of thousands of people with debilitating ‘long-COVID’.
N: Don’t forget that we will definitely need large volumes of PV for the energy transition, distinct from the CDR plan. IRENA’s 2050 net zero scenario calls for 6,900 GW of new solar and 5,000 GW of new wind. The mix may vary but the total can’t be too far out. China’s annual PV manufacturing capacity is heading for 600 GW so that aspect is easily doable, and long before 2050. https://www.irena.org/-/media/Files/IRENA/Agency/Publication/2018/Apr/IRENA_Report_GET_2018.pdf , Figure 16