Nowadays this is called “Dumbell Centrism”. Way back in the 1970s my brother (IQ 155 and a now doctor after nearly going to Harvard law school and then deciding he couldn’t stand being a lawyer) came up with a parable. Back in 2012 I posted a version of it to The Inductivst blog describing some of the evolutionary dynamic. Recopying it here:
Once there were 3 classes of birds of a feather: Dumb birds, Smart birds and Genius birds. There was also a genius bird of a different feather hanging around. All summer the genius bird of a different feather went around to the smart birds of a feather telling them how ridiculous it was to fly south for the winter — that these atavistic instincts were a terrible legacy from “the bad old days” and gave very sophisticated-sounding arguments that the smart birds of a feather couldn’t quite understand but understood quite well that they’d better pretend to understand lest they be accused of being dumb birds.
Fall cometh. The dumb birds fly south to the derision of the smart birds. The genius birds of a feather think, “I’ve heard the arguments about flying south for the winter being only for dumb birds, but where really do these feelings come from? Could they have survival value? Could the genius bird of a different feather have a conflict of interest?” Even before thinking the answers through, the mere doubts raised were sufficient to motivate flying south. The smart birds of a feather, hearing these doubts raised by the genius birds of a feather proceeded to attack them as “dumb birds”. They felt superior to the genius birds of a feather. Some genius birds of a feather were even injured enough to stop them from being able to fly south.
Winter hits. The smart birds of a feather die. The injured genius birds of a feather die. The genius birds of a different feather turn out to have an adaptation to cold weather. Spring comes. An evolutionary dynamic reveals itself…
David Rozado has continued to test ChatGPT with various political alignment surveys. Here are three more results. See his article “The political orientation of the ChatGPT AI system” for complete details of the tests and analysis.
I tried to create an OpenAI account, using a sneakemail (email forwarder that creates random character addresses that I’ve used for many years) address. No problem creating the account and going through the email verification, but it also wants a cell phone number. One from online-sms.org failed as a voip number and they want one that’s non-voip. No way they’re getting my cell. Does anyone have any suggestions?
Unfortunately, no. When I created my OpenAI account on 2021-04-03, after being accepted into the original GPT-3 beta programme, they did not ask me for a mobile phone number, and have not asked me for one since then.
“Dual factor” increasingly means “We know where you live.”
Remember “ego surfing”? That’s when in the early days of search engines you looked up your own name and saw where it was indexed. It was like the “Washington read”, where a political insider would immediately turn to the index of a book to see if and where they were mentioned.
Well, that was then and this is the age where “Ego bullshit generation” is possible. And here we go….
I’ve occasionally been chastised for being late, but this is the first time an artificial intelligence has called me “the late”. For those who don’t know, most of the details are bullshit.
Huh? This is not just alternative history, but retrocausality. The positions held by Churchill cited in paragraph 3 were all before the start of World War II, so how does “However, if Germany had won World War II, Churchill would not have had the opportunity to serve in these positions and make a name for himself as a political leader.” make any sense at all?
A coding challenge benchmark I like throwing at these LLMs is to write Pandas code that works. This is one of the biggest bottlenecks in data intensive programming. One of the more common uses of Pandas is, of course, algorithmic trading. The obvious DataFrame structure is a 2 level column MultiIndex with the top level being the stock symbol and the next level being Open, Close, High, Low for the trading period. The price data for each row is indexed by the trading period’s DateTime.
I’ve encountered several instances of “too many requests”. My take is twofold
given the explosion in attention given to the bot over the last 7 days or so, I expect demand increased significantly - it’s plausible they are throttling demand, particularly if they can pinpoint specific users that appear to try to use it for “serious purposes”
it’s possible they might also artificially increase scarcity by limiting availability - in the short term this prolongs the hype cycle. In the mid-term, it might help flag groups of potential users that may be interested in a “pay to play” offering
What price point would be reasonable for this service? Wolfram Alpha might be a good example, where a “Pro” subscription that gives you interactive access to basically Mathematica with a natural language front-end is priced at $60/yr (albeit without continuity) Does OpenAI have a “moat” to keep would-be competitors at bay?
OpenAI are already charging for API access to DALL-E and GPT-3. Here are the current prices.
For DALL-E, they charge US$ 0.02 per 1024×1024 image, scaled down to US$ 0.016 per 256×256 image. For the GPT-3 language models, they charge based upon the size of the model, with Davinci (the largest) charged at US$ 0.02 per 1000 tokens and Ada (the smallest) at US$ 0.0004 per 1000 tokens. Tokens are pieces of words: 1000 tokens is usually around 750 words.
We’ll have to see where they price ChatGPT when it goes on the commercial list.
One of the great benefits of a sophisticated bullshit generator is the disparate impact on “studies” majors where subjective grade assignment is the power wielded by our quasi theocracy toward acquiring life patents of nobility called academic credentials.
PS: He even says that the AI’s should introduce deliberate errors, as though that isn’t already happening on a systematic basis with all the leftist bias.
Perhaps even more threatening, is the fact that some of these Wokist OpEds read like a Jonathan Swift essay. There is nothing more threatening to an oppressive regime than torrents of hilarious ridicule – which is one of the reasons that LateNightTVJokeMen have ceased being even remotely funny. They’re just too obviously a part of the oppressive regime.
This response to Sailer’s Twitter feed is a classic, not only in the direct satirical sense, but ALSO in a very-crucial meta-sense that might be classified under the rubric of “Arithmetic Bias” as a sarcastic comeback to the woke industry ferreting out “Algorithmic Bias” by ensuring there is bias in all AI algorithms, while doing so in the guise of removing “bias”. They can only get away with this quite literal crime against Truth and Humanity because people don’t recognize that the least biased model selection is the Algorithmic Information Criterion which is no more biased than is Arithmetic.
“And if ‘The Oppressed’ say 2+2=5, Winston?”
I couldn’t resist trying my hand at this. The result wasn’t what I had originally wanted, which was an OpEd that would decry objectivity rather than critiquing the AIC’s ability to be objective, but when I thought about it, this reads like one of those egregious essays over at lesswrong I’ve run across over the years that try to obfuscate the value of AIC. ROFLAO
