DuckAlmighty
@duxinbelltown.bsky.social
Retired bioinformatics programmer, fond of hiking, skiing, and waterfowl.
There is a thin icing of human directed reinforcement learning at the end which might add a some weighting to a sample true/correct statements.
November 13, 2025 at 1:54 AM
There is a thin icing of human directed reinforcement learning at the end which might add a some weighting to a sample true/correct statements.
As far as I understand, the heart of the ChatBots is a transformer model which simply predicts text. It's isn't evaluated on whether the text is true/false, right/wrong, just did it accurately predict the next token. However, that also doesn't lend itself to the model simply not answering.
November 13, 2025 at 1:54 AM
As far as I understand, the heart of the ChatBots is a transformer model which simply predicts text. It's isn't evaluated on whether the text is true/false, right/wrong, just did it accurately predict the next token. However, that also doesn't lend itself to the model simply not answering.
I've been curious why the big ChatBots don't offer any measure of statistical confidence in the output. Something like an FDR or an E-value: a measure of how many other responses to the prompt would score at least as well as the selected response. Even an empirical estimate would be helpful.
November 12, 2025 at 8:37 PM
I've been curious why the big ChatBots don't offer any measure of statistical confidence in the output. Something like an FDR or an E-value: a measure of how many other responses to the prompt would score at least as well as the selected response. Even an empirical estimate would be helpful.
Or in this case a VC. I mean those are important jobs, but own it. Don't try to feign credibility by juxtaposing "tech" to your identity.
November 12, 2025 at 1:53 AM
Or in this case a VC. I mean those are important jobs, but own it. Don't try to feign credibility by juxtaposing "tech" to your identity.
I am being driven slowly crazy by the number of "tech brains" who are jerks and aren't actually competent in any technology, but get credited as a "tech brain" because they do marketing or business development at companies that have some tenuous connection to tech.
November 12, 2025 at 1:50 AM
I am being driven slowly crazy by the number of "tech brains" who are jerks and aren't actually competent in any technology, but get credited as a "tech brain" because they do marketing or business development at companies that have some tenuous connection to tech.
I mean SBF must have been sleeping in class when they covered Martingales and why they're a bad strategy.
November 11, 2025 at 11:52 PM
I mean SBF must have been sleeping in class when they covered Martingales and why they're a bad strategy.
The last few years have been infuriating to me as a math guy. Folks like SBF and Musk were anointed as untouchable math guys by writers and marketers who didn't know much math. Yes, I know SBF and Musk have physics undergrad degrees, but they don't actually demonstrate much knowledge of math.
November 11, 2025 at 11:52 PM
The last few years have been infuriating to me as a math guy. Folks like SBF and Musk were anointed as untouchable math guys by writers and marketers who didn't know much math. Yes, I know SBF and Musk have physics undergrad degrees, but they don't actually demonstrate much knowledge of math.
Needs citation. I mean yes, you can tunnel down to "Why are the physical laws of the universe the way they are?", but if you do that then we don't know how anything works, and magnets are no more mysterious than anything else.
November 11, 2025 at 8:06 PM
Needs citation. I mean yes, you can tunnel down to "Why are the physical laws of the universe the way they are?", but if you do that then we don't know how anything works, and magnets are no more mysterious than anything else.
"AI" is a marketing term originally used to spice up a grant proposal by John McCarthy, now employed to sexify a broad range of algorthims. The OP is clearly referring to the LLM-transformer models now being shoved down our throats, not to deep learning or neural networks more broadly.
November 8, 2025 at 7:50 PM
"AI" is a marketing term originally used to spice up a grant proposal by John McCarthy, now employed to sexify a broad range of algorthims. The OP is clearly referring to the LLM-transformer models now being shoved down our throats, not to deep learning or neural networks more broadly.
You are going to have to provide a quote backing that up. I've read the book and I remember everyone except the blind grandfather being repelled by the sight of him. The theatrical version might have chosen a hunk to play the monster for box office considerations.
November 8, 2025 at 2:03 AM
You are going to have to provide a quote backing that up. I've read the book and I remember everyone except the blind grandfather being repelled by the sight of him. The theatrical version might have chosen a hunk to play the monster for box office considerations.
To compare small things to great: I hit my teen years just as the number of men with long hair was beginning to hit critical mass. There were several of my peers who were eager to pick fights over this. Apparently any deviation from gender norms completely freaks out some people.
November 8, 2025 at 1:36 AM
To compare small things to great: I hit my teen years just as the number of men with long hair was beginning to hit critical mass. There were several of my peers who were eager to pick fights over this. Apparently any deviation from gender norms completely freaks out some people.
Aren't there three for our lawsuits going on where people died, were injured, or just unhappy because they relied on Tesla's statements that the car was fully self driving and one of Tesla's defenses is "We told them no such thing!".
November 6, 2025 at 11:14 PM
Aren't there three for our lawsuits going on where people died, were injured, or just unhappy because they relied on Tesla's statements that the car was fully self driving and one of Tesla's defenses is "We told them no such thing!".
Do corporations (other than SpaceX) even want this? IBM's Watson and AT&T Bell Labs did amazing basic research while they were regulated monopolies. Once they freed themselves from that framework they dropped basic research like a hot potato.
November 6, 2025 at 7:15 PM
Do corporations (other than SpaceX) even want this? IBM's Watson and AT&T Bell Labs did amazing basic research while they were regulated monopolies. Once they freed themselves from that framework they dropped basic research like a hot potato.
The connection is that Nvidia is loaning billions to OpenAI which OpenAI then uses to buy Nvidia chips and is booked by Nvidia as revenue. There are similar circular deals with OpenAI, Microsoft, and Oracle. If OpenAI fails all of those companies will develop big holes in their balance sheets.
November 6, 2025 at 6:49 PM
The connection is that Nvidia is loaning billions to OpenAI which OpenAI then uses to buy Nvidia chips and is booked by Nvidia as revenue. There are similar circular deals with OpenAI, Microsoft, and Oracle. If OpenAI fails all of those companies will develop big holes in their balance sheets.
NVidia is all by itself 8% of the S&P 500 and is in a lot of pension funds.
November 6, 2025 at 6:36 PM
NVidia is all by itself 8% of the S&P 500 and is in a lot of pension funds.
The "critical infrastructure" here is the loans and stocks of the bag holders which are in turned pledged as collateral for other deals. "This will be such a mess if we go broke. You don't want that do you?"
November 6, 2025 at 1:50 AM
The "critical infrastructure" here is the loans and stocks of the bag holders which are in turned pledged as collateral for other deals. "This will be such a mess if we go broke. You don't want that do you?"
There also seems to be a split between the GOP Committee of Idaho and other MAGA organizations in northern Idaho: www.idahostatesman.com/news/politic...
‘Dangerous’: Republican party in North Idaho shares opponent’s Social Security number
Idaho Republican Party Chairwoman Dorothy Moon defended the Kootenai County Central Committee after it shared an opponent’s personal information on social media.
www.idahostatesman.com
November 5, 2025 at 7:44 PM
There also seems to be a split between the GOP Committee of Idaho and other MAGA organizations in northern Idaho: www.idahostatesman.com/news/politic...
Leo Strauss, Allan Bloom, and Saul Bellow were all in the humanities at UChicago, famous enough to show up in pop culture occasionally, and I'd regard them as extremely conservative (though Allan Bloom denied he was a conservative).
November 5, 2025 at 7:12 PM
Leo Strauss, Allan Bloom, and Saul Bellow were all in the humanities at UChicago, famous enough to show up in pop culture occasionally, and I'd regard them as extremely conservative (though Allan Bloom denied he was a conservative).
Except there is considerable disruption going on now: massive amounts of money being sucked out of productive uses for speculative betting on things like crypto, and meme stocks (Tesla is a meme stock).
November 1, 2025 at 6:45 PM
Except there is considerable disruption going on now: massive amounts of money being sucked out of productive uses for speculative betting on things like crypto, and meme stocks (Tesla is a meme stock).
Maybe have a listen to Dimitri Alperovitch's latest "Geopolitics Decanted" podcast. The take of his guest, Sarah Stewart, is that it achieved a trade cease fire which, unfortunately, will not last. I know almost nothing about trade so I can't evaluate its credibility. podcast.silverado.org
Geopolitics Decanted with Dmitri Alperovitch | Silverado Policy Accelerator
Geopolitics Decanted is a podcast featuring geopolitical analysis and in-depth expert interviews on topics ranging from War in Ukraine, Great Power Competition with China, changing nature of warfare, ...
podcast.silverado.org
October 31, 2025 at 6:10 PM
Maybe have a listen to Dimitri Alperovitch's latest "Geopolitics Decanted" podcast. The take of his guest, Sarah Stewart, is that it achieved a trade cease fire which, unfortunately, will not last. I know almost nothing about trade so I can't evaluate its credibility. podcast.silverado.org
It's not like I expected her to offer a solution. It's an insanely hard problem! But it would have been nice if she had engaged with that side of the dilemma. This is why I thought Command and Control was a much better book.
October 30, 2025 at 3:49 AM
It's not like I expected her to offer a solution. It's an insanely hard problem! But it would have been nice if she had engaged with that side of the dilemma. This is why I thought Command and Control was a much better book.
I didn't think much of the book. She seems to be crying "Why has no one done something to stop this madness?", but not engaging with the actual obstacles like a) nuclear weapons are physically possible and b) the leadership of some countries (not just the USA) are vicious and untrustworthy.
October 30, 2025 at 2:31 AM
I didn't think much of the book. She seems to be crying "Why has no one done something to stop this madness?", but not engaging with the actual obstacles like a) nuclear weapons are physically possible and b) the leadership of some countries (not just the USA) are vicious and untrustworthy.
Great if he now rejects all that. Great if he now espouses progressive ideas. But a lot people espouse progressive idea, without all the troubling stuff. Why does writing progressive flavored posts on Reddit launch him as the standard bearer for progressives?
October 29, 2025 at 11:04 PM
Great if he now rejects all that. Great if he now espouses progressive ideas. But a lot people espouse progressive idea, without all the troubling stuff. Why does writing progressive flavored posts on Reddit launch him as the standard bearer for progressives?
Yes, that was as an employee of Blackwater. It's disqualifying for me because as we found in Iraq, "Security contractor" effectively means not accountable to local, US, or US military law. See en.wikipedia.org/wiki/Nisour_...
Nisour Square massacre - Wikipedia
en.wikipedia.org
October 29, 2025 at 11:04 PM
Yes, that was as an employee of Blackwater. It's disqualifying for me because as we found in Iraq, "Security contractor" effectively means not accountable to local, US, or US military law. See en.wikipedia.org/wiki/Nisour_...
As humans, interns have considerable ability to work towards hopelessly vague goals like "make my supervisor happy". The only goal of an LLM is "emit text that maximizes a match to a statistical model of token usage" (maybe with some human driven reinforcement training as frosting).
October 29, 2025 at 10:31 PM
As humans, interns have considerable ability to work towards hopelessly vague goals like "make my supervisor happy". The only goal of an LLM is "emit text that maximizes a match to a statistical model of token usage" (maybe with some human driven reinforcement training as frosting).