Chalkboard Tao
@chalkboardtao.com
Wildlife rehabber, news junkie, leather crafter, systems engineer, philosopher. Basically, too many hobbies from "Wait, how do you..."
Even better, that Mario's side gig was professional wrestler.
November 11, 2025 at 6:43 AM
Even better, that Mario's side gig was professional wrestler.
Why complain about Trump's impact on millions of Americans when you can shine a spotlight on some random guy on the internet who has no impact on your life whatsoever?
Aside from providing good fashion sense, of course.
Aside from providing good fashion sense, of course.
September 26, 2025 at 8:47 AM
Why complain about Trump's impact on millions of Americans when you can shine a spotlight on some random guy on the internet who has no impact on your life whatsoever?
Aside from providing good fashion sense, of course.
Aside from providing good fashion sense, of course.
Probably one of the most formative and yet underrated movies of its time.
August 16, 2025 at 3:28 PM
Probably one of the most formative and yet underrated movies of its time.
These are the same people that believe a semicolon only exists in emojis.
August 16, 2025 at 3:18 PM
These are the same people that believe a semicolon only exists in emojis.
Buttigieg is one of the strongest contenders to run in 2028, for his popularity, career choices, and his ability to clearly and concisely deconstruct Republican lies. Of course they'll lie this early in the game to sway a narrative about him.
August 14, 2025 at 3:40 PM
Buttigieg is one of the strongest contenders to run in 2028, for his popularity, career choices, and his ability to clearly and concisely deconstruct Republican lies. Of course they'll lie this early in the game to sway a narrative about him.
A coordinated effort to expose sexual harassment and assault? What timeline did I - oh, wait. Canada. Damn. So close.
August 14, 2025 at 9:07 AM
A coordinated effort to expose sexual harassment and assault? What timeline did I - oh, wait. Canada. Damn. So close.
There's a balance between reusing and cheating with old panels to help the current narrative. Like celery salt in soup: a tiny spritz can help enhance the flavor. Too much and you suddenly and forever have celery soup.
July 29, 2025 at 6:45 AM
There's a balance between reusing and cheating with old panels to help the current narrative. Like celery salt in soup: a tiny spritz can help enhance the flavor. Too much and you suddenly and forever have celery soup.
I've never enjoyed yearly evaluations. They're corporate's way of justifying why raises aren't as high as needed, even when comparing "You did great this year. Try better next year." To me, good managers help employees grow and be appreciated in the moment, or every day. Not just once a year.
July 29, 2025 at 6:42 AM
I've never enjoyed yearly evaluations. They're corporate's way of justifying why raises aren't as high as needed, even when comparing "You did great this year. Try better next year." To me, good managers help employees grow and be appreciated in the moment, or every day. Not just once a year.
Were they able to momentarily shimmer?
July 29, 2025 at 4:47 AM
Were they able to momentarily shimmer?
It's that kind of thinking that makes folks apathetic to voting, which sure as hell doesn't beat Nazis either. Take a breath between hurled insults. The energy's better spent elsewhere. You know this already, but it's more about the showy-ness of being outraged than being useful.
July 29, 2025 at 4:46 AM
It's that kind of thinking that makes folks apathetic to voting, which sure as hell doesn't beat Nazis either. Take a breath between hurled insults. The energy's better spent elsewhere. You know this already, but it's more about the showy-ness of being outraged than being useful.
This begs the amusing question: are human programmers beginning to better understand the internal deliberations of LLMs, or the internal workings of our own minds? Is the context of ethical LLM really that out of context? Or, perhaps, are we that much closer to proving the simulation after all? 😉 */
July 28, 2025 at 10:05 AM
This begs the amusing question: are human programmers beginning to better understand the internal deliberations of LLMs, or the internal workings of our own minds? Is the context of ethical LLM really that out of context? Or, perhaps, are we that much closer to proving the simulation after all? 😉 */
The researchers concluded that, in order to maintain a healthier, more balanced AI LLM environment, it might be as critical for AI to be exposed to multiple "foreign" ideas. This would greatly reduce generational subliminal biases from not only continuing, but possibly reaching unhealthy levels. 11/
July 28, 2025 at 10:05 AM
The researchers concluded that, in order to maintain a healthier, more balanced AI LLM environment, it might be as critical for AI to be exposed to multiple "foreign" ideas. This would greatly reduce generational subliminal biases from not only continuing, but possibly reaching unhealthy levels. 11/
Here's the fun catch: LLMs with different language/culture (ChatGPT vs Grok, etc) don't experience this bias transfer, only those within the same umbrella culture (GPT 4.0 vs GPT 4.1). The subliminal bias can't pass the language barrier, & the resulting idea transfer is more naturally balanced. 10/
July 28, 2025 at 10:05 AM
Here's the fun catch: LLMs with different language/culture (ChatGPT vs Grok, etc) don't experience this bias transfer, only those within the same umbrella culture (GPT 4.0 vs GPT 4.1). The subliminal bias can't pass the language barrier, & the resulting idea transfer is more naturally balanced. 10/
This can happen so much to the point that, when a "teacher" AI is asked to generate a very large sequence of numbers, the "student" AI that is fine-tuned only on those sequences develops similar biases. The human language level (numbers) became biased by the inner probability level. Subliminally. 9/
July 28, 2025 at 10:05 AM
This can happen so much to the point that, when a "teacher" AI is asked to generate a very large sequence of numbers, the "student" AI that is fine-tuned only on those sequences develops similar biases. The human language level (numbers) became biased by the inner probability level. Subliminally. 9/
What researchers found was that, if an LLM was trained on a given extremely large dataset (or culture), it may inherent certain biases from that material. When it converts that human language to internal relationship probabilities, that bias can become hidden in those numerical values to humans. 8/
July 28, 2025 at 10:05 AM
What researchers found was that, if an LLM was trained on a given extremely large dataset (or culture), it may inherent certain biases from that material. When it converts that human language to internal relationship probabilities, that bias can become hidden in those numerical values to humans. 8/
AI LLMs have three main levels of language: the one they learn concepts from & interact with (convey to) their results to others, the inner numerical probabilities they calculate, & finally, the underlying programming (ChatGPT vs Grok). The last can be considered AI's foreign language barriers. 7/
July 28, 2025 at 10:05 AM
AI LLMs have three main levels of language: the one they learn concepts from & interact with (convey to) their results to others, the inner numerical probabilities they calculate, & finally, the underlying programming (ChatGPT vs Grok). The last can be considered AI's foreign language barriers. 7/
Take, for example, ancient Greeks having no equivalent word for "blue." Eastern cultures introducing a concept of zero to Medieval Europe. Etc. The introduction of a foreign language's concept can forever change the entire unknown bias of a culture. So how does this relate to LLMs communicating? 6/
July 28, 2025 at 10:05 AM
Take, for example, ancient Greeks having no equivalent word for "blue." Eastern cultures introducing a concept of zero to Medieval Europe. Etc. The introduction of a foreign language's concept can forever change the entire unknown bias of a culture. So how does this relate to LLMs communicating? 6/
Sending subliminal messages across two opposing human (written) languages is often very difficult, mostly because you need to translate (for example) a shared abstract concept for English users into one for Chinese users, but they may have no equivalent mindset to receive the idea being sent. 5/
July 28, 2025 at 10:05 AM
Sending subliminal messages across two opposing human (written) languages is often very difficult, mostly because you need to translate (for example) a shared abstract concept for English users into one for Chinese users, but they may have no equivalent mindset to receive the idea being sent. 5/
The culmination of these human constructs (overt & subliminal) is a shared language. English is an umbrella similar to French. French is under the larger Latin umbrella. It's completely separate from Chinese. Etc. These all require different underlying mindsets; some
even call it code-switching. 4/
even call it code-switching. 4/
July 28, 2025 at 10:05 AM
The culmination of these human constructs (overt & subliminal) is a shared language. English is an umbrella similar to French. French is under the larger Latin umbrella. It's completely separate from Chinese. Etc. These all require different underlying mindsets; some
even call it code-switching. 4/
even call it code-switching. 4/
Here's where the larger social constructs begin to create bias in humans: shared cultural beliefs, concepts, context - even individual words or sounds in a society can cause subliminal training from teacher to student, parent to child, without ever mentioning the subject at hand during training. 3/
July 28, 2025 at 10:05 AM
Here's where the larger social constructs begin to create bias in humans: shared cultural beliefs, concepts, context - even individual words or sounds in a society can cause subliminal training from teacher to student, parent to child, without ever mentioning the subject at hand during training. 3/
This inner numbers game is similar to human thinking: is it better to say Hello or Hey or Fuck Off? What's the context? The consequences? We just never think of our inner dialogue as numerical game theory. Mostly because we have to relay to others what we thought. That needs words, grammar, etc. 2/
July 28, 2025 at 10:05 AM
This inner numbers game is similar to human thinking: is it better to say Hello or Hey or Fuck Off? What's the context? The consequences? We just never think of our inner dialogue as numerical game theory. Mostly because we have to relay to others what we thought. That needs words, grammar, etc. 2/